1 INTERNATIONAL JOURNAL OF CLIMATOLOGY Int. J. Climatol. 26: (2006) Published online 26 January 2006 in Wiley InterScience ( DOI: /joc.1286 MODELLING...
1 University of Massachusetts Amherst From the SelectedWorks of Raymond S Bradley February, 1986 Northern Hemisphere Surface Air Temperature Variation...
1 University of Wisconsin Milwaukee UWM Digital Commons Theses and Dissertations May 2014 Northern Hemisphere Sea Level Pressure Synchronization and I...
1 Quarterly Journalof the RoyalMeteorologicalSociety Q. J. R. Meteorol. Soc. 140: , July 2014 B DOI: /qj.2263 Centennial trends in Northern Hemisphere...
1 Recent warm and cold daily winter temperature extremes in the Northern Hemisphere Kristen Guirguis, Alexander Gershunov, Rachel Schwartz, and Stephe...
1 1OCTOBER 2010 C H O I E T A L Changing Northern Hemisphere Snow Seasons GWANGYONG CHOI Department of Environmental Science, Kangwon National Univers...
1 1 2 3 Unraveling the teleconnection mechanisms that induce wintertime temperature anomalies over the Northern Hemisphere continents in response to t...
1 The Northern Hemisphere Sea ice Trends: Regional Features and the Late 1990s Change Renguang Wu Institute of Atmospheric Physics, Chinese Academy of...
1 Global Nest: the Int. J. Vol 4, No 2-3, pp , 2002 Copyright 2002 GLOBAL NEST Printed in Greece. All rights reserved ON THE RELATION BETWEEN SEA SURF...
1 Title Planetary Wave Activity in the Trop during the Northern Hemisphere Wint Author(s) Shiotani, Masato Citation Journal of the Atmospheric Science...
Holocene temperature trends in the Northern Hemisphere extratropics ACADEMISCH PROEFSCHRIFT ter verkrijging van de graad Doctor aan de Vrije Universiteit Amsterdam, op gezag van de rector magnificus prof.dr. V. Subramaniam, in het openbaar te verdedigen ten overstaan van de promotiecommissie van de Faculteit der Bètawetenschappen op maandag 20 november 2017 om 13.45 uur in de aula van de universiteit, De Boelelaan 1105
door Yurui Zhang geboren te Gansu, China
prof.dr. H. Renssen
copromotor: prof.dr. H. Seppä
prof. dr. R.T. van Balen prof. dr. A.J. Dolman prof. dr. M. Crucifix dr. W.Z. Hoek dr. Q. Zhang,
This research was carried out at Vrije Universiteit Amsterdam Faculty of Sciences Department of Earth Sciences Earth and Climate cluster De Boelelaan 1085, 1081 HV Amsterdam, The Netherlands
University of Helsinki Faculty of Science Department of Geosciences and Geography Division of Geology Gustaf Hällströmin katu 2a, FI-00014 Helsinki, Finland
This research was funded by China Scholarship Council (No. 201306180016).
Author: Yurui Zhang ISBN: 978-94-028-0813-1 Holocene temperature trends in the Northern Hemisphere extratropics In Dutch: Het temperatuurverloop op het noordelijke halfrond tijdens het Holoceen, ten noorden van de tropen
Contents Acknowledgements ................................................................................................... 9 Summary ..................................................................................................................11 Nederlandse Samenvatting ...................................................................................... 15 Chapter 1 Introduction ...................................................................................... 19 1.1 The climate system, its variation and paleoclimate change .......................... 20 1.1.1 The climate system and mechanisms underlying climate change .......... 20 1.1.2 Palaeoclimate change and the Holocene ................................................ 23 1.2 The early-Holocene transition and Holocene climate forcings ..................... 24 1.2.1 The early-Holocene transition in the climate system ............................. 24 1.2.2 Main forcings of the Holocene climate .................................................. 26 1.2.3 Previous studies on the Holocene climate and remaining problems ...... 26 1.3 Methods to study Holocene climate .............................................................. 28 1.3.1 The LOVECLIM climate model ............................................................ 28 1.3.2 Sensitivity experiments .......................................................................... 31 1.3.3 Inter-model comparisons........................................................................ 32 1.3.4 Proxy records and model–data comparisons .......................................... 33 1.3.5 Variation partitioning ............................................................................. 35 1.4 The research questions .................................................................................. 36 Chapter 2 Effects of melting ice sheets and orbital forcing on the early Holocene warming in the extratropical Northern Hemisphere ........................ 39 Abstract ............................................................................................................... 39 2.1 Introduction ................................................................................................... 40 2.2 Model and experimental design .................................................................... 42 2.2.1 The LOVECLIM model ......................................................................... 42 2.2.2 Prescribed forcings ................................................................................. 43 2.2.3 Setup of experiments .............................................................................. 45 2.3 Results ........................................................................................................... 46 2.3.1 Equilibrium experiments at the onset of the Holocene .......................... 46 2.3.2 Transient simulation for the Holocene ................................................... 48 2.4 Discussion ..................................................................................................... 52 2.4.1 Comparison of simulations with proxy records ..................................... 52 2.4.2 Mechanism of climate response to forcings ........................................... 54 2.4.3 Early Holocene warming and climate–ocean system response to freshwater ........................................................................................................ 57 2.5 Conclusions ................................................................................................... 58 Chapter 3 Holocene temperature trends in the extratropical Northern Hemisphere based on inter-model comparisons ................................................. 61 Abstract ............................................................................................................... 61 3.1 Introduction ................................................................................................... 62
3.2 Models and simulations ................................................................................ 63 3.2.1 Models and prescribed climate forcings ................................................ 63 3.2.2 Setup of simulations ............................................................................... 68 3.3 Results ........................................................................................................... 69 3.3.1 Simulated temperature in the NH extratropics ....................................... 69 3.3.2 Temperature over the regions with good inter-model agreements ......... 72 3.3.3 Temperatures over the regions with less multi-model consistency ........ 73 3.4 Discussion ..................................................................................................... 75 3.4.1 Divergent climate variables lead to mismatched temperatures .............. 75 3.4.2 Potential sources contributing to inter-model divergences of climate variables .......................................................................................................... 81 3.5 Conclusions ................................................................................................... 84 Chapter 4 Holocene temperature evolution in the Northern Hemisphere high latitudes—model–data comparisons ................................................................... 87 Abstract ............................................................................................................... 87 4.1 Introduction ................................................................................................... 88 4.2 Methods ......................................................................................................... 90 4.2.1 Data and analysis ................................................................................... 90 4.2.2 Forcings and simulations ....................................................................... 92 4.3 Results and discussion .................................................................................. 96 4.3.1 Temperatures in Fennoscandia ............................................................... 96 4.3.2 Temperatures in Greenland .................................................................... 99 4.3.3 Temperatures in north Canada ............................................................. 101 4.3.4 Temperatures in Alaska ........................................................................ 103 4.3.5 Temperatures in high-latitude Siberia .................................................. 105 4.3.6 Summarized discussion on model–data comparisons .......................... 106 4.4 Conclusions ................................................................................................. 108 Chapter 5 The role of climate, forest fires and human population size in Holocene vegetation dynamics in Fennoscandia ............................................... 111 Abstract .............................................................................................................. 111 5.1 Introduction ..................................................................................................112 5.2 Materials and methods .................................................................................115 5.2.1 Study area ..............................................................................................115 5.2.2 Regional plant abundances and plant functional types (PFTs) .............116 5.2.3 Explanatory variables of Holocene vegetation changes .......................117 5.2.4 Statistical analysis .................................................................................119 5.3 Results ......................................................................................................... 120 5.3.1 Holocene change in PFTs and explanatory variables ........................... 120 5.3.2 Variation partitioning results ................................................................ 122 5.4 Discussion ................................................................................................... 126 5.4.1 From natural vegetation dynamics towards increasing anthropogenic
influence ........................................................................................................ 126 5.4.2 Potential biases in the method and explanatory variables.................... 130 5.5 Conclusions ................................................................................................. 133 Chapter 6 Synthesis and outlook ..................................................................... 135 6.1 Summary of the main findings .................................................................... 135 6.1.1 Effects of the ice sheets on Holocene climate change in the NH extratropics .................................................................................................... 135 6.1.2 Inter-model comparisons of the Holocene climate trend ..................... 138 6.1.3 Model–data comparisons of Holocene temperatures at NH high latitudes ......................................................................................................... 141 6.2 Remaining issues and Outlook.................................................................... 147 Appendix ............................................................................................................... 151 Appendix A: Abbreviations ............................................................................... 151 Appendix B: List of figures and tables ............................................................. 152 Appendix C: Supplementary information ......................................................... 155 Bibliography.......................................................................................................... 167
Acknowledgements By doing this research during the last four years, I finally understand why a PhD experience is often seen as a metaphor of an adventurous journey. Like many journeys, my PhD life possesses all elements of an adventure, including happiness, surprises, worries, anxiety, and all kinds of memorable moments. Luckily, I received countless kind helps and supports from many kind people, which enables me to approach the end of this journey. This journey is enormously valuable for me because it not only leads to academic progresses, but also means the self-improvements and personal development. At this moment, I am excited about to finish it, and also want to express my sincere thanks to all of those who made this special journey to happen. My great gratitude goes to Hans Renssen, my supervisor, for his guidance and enlightening. I feel so lucky to have a chance to work with him and I am grateful for his guidance throughout this journey. I could not imagine what would have happened in this journey without his supervision. I still clearly remember our first meeting and discussion in Amsterdam in 2014. That is the tipping-point of my PhD journey, although I have to admit the critical roles of previous preparation. During that meeting we came out with the basic outline and structure on how the project should go. Since that meeting, I gradually understood where we were going to go. I have also visited him in VU Amsterdam several times, for example for the setup of the LOVECLIM climate model and for participating in the palaeoclimate model course. Thanks to Hans for providing helps and support for arranging my stay, and also for inducing me to the department. I am grateful to Hans for having countless Skype meetings, which provided the great chance to keep updating my work and also to receive feedbacks. I was always impressed by his endless knowledge in climate modeling and relevant fields. Discussions with him always turned out to be enlightening and fruitful. He was always well prepared for all kinds of questions (apologized for some fool questions) that confused me so much. His comments on my each of our manuscripts were enormous benefit. I felt a real upgrade every time after incorporating these comments. Of course, I know that making these valuable comments took a lot of time and I really appreciate all his time spent on my PhD project. Additionally, thanks Hans for translating the summary into Dutch. I want to thank Heikki Seppä, another my supervisor, for showing me the field of palaeoclimatology and for introducing his colleagues to me. He also gave me partial financial support to attend several conferences, which provided me opportunity to meet colleagues on a global scale and educate myself extensively. I believe that this is an essential part my PhD journey and will be conducive to my future academic career as well. Many thanks also go to my colleagues. Hui Tang and Didier Roche are acknowledged for their technical helps at the early stage of this journey. I also want to thank my co-authors, Paul Valdes and Niina Kuosmanen for
nice cooperation. The members of the reading committee of this thesis, Ronald Van Balen, Han Dolman, Michel Crucifix, Wim Hoek, and Qiong Zhang, are acknowledged for their reviewing, suggestions and approval. Thanks also go to my VU colleagues, Fazer, Marinna, Huan and others for helps and sharing information. I also would like to express my thanks to the personnel of VU for providing smooth administrative affairs and to UV friends for making my stays were so much fun. I also would like to express my thanks to my geosciences colleagues and friends in Helsinki for sharing information and nice work environment, and to the personnel of the department for providing supports. In short, I cannot list all of your names, but I would like to take this chance to thank all colleagues and friends. Financial supports are also acknowledged. In particular, this work received financial support from the China Scholarship Council. Doctoral School of Geosciences and Chancellor’s Travel Grant (University of Helsinki) are acknowledged for providing support for attending international conferences and other scientific visits. On a personal note, I would like to thank my family and friends for being my spiritual pillar. I appreciate my family for being always thoughtful and encouraging even from far away. Thanks to many of my Chinese and others friends for sharing nice weekend meals and for their company during the all kinds of festivals and holidays. Finally, I would like to express hearty gratitude to all of my colleagues and friends for your helps and suports that made this journey memorable.
Summary As the latest epoch of the Earth’s history, the Holocene is commonly defined as the last 11.7 ka BP (hereafter referred to as ka) and represents a new phase, encompassing the time span of human civilization. The last deglaciation lasted well into the Holocene, implying that the early Holocene was characterized by a large-scale reorganization with transitions in various components of the climate system. Studying the Holocene can provide insights into how the climate system functions, apart from the theoretical contributions to climate history itself. We first conducted sets of simulations with different combinations of climate forcings for 11.5 ka and for the entire Holocene to investigate the response of the climate–ocean system to the main climate forcings. In particular, two possible freshwater flux (FWF) scenarios were further tested considering the relatively large uncertainty in reconstructed ice-sheet melting. Moreover, we compared four Holocene simulations performed with the LOVECLIM, CCSM3, FAMOUS and HadCM3 models by identifying the regions where the multi-model simulations are consistent and where they are not, and analysing the reasons at the two levels (of the models’ variables and of the model principles and physics) where mismatches were found. After this, these multi-model simulations were systematically compared with data-based reconstructions in five regions of the Northern Hemisphere (NH) extratropics, namely Fennoscandia, Greenland, North Canada, Alaska and high-latitude Siberia. Potential uncertainty sources were also analysed in both model simulations and proxy data, and the most probable climate histories were identified with the aid of additional evidence when available. Additionally, the contribution of climate change, together with forest fires and human population size, to the variation in Holocene vegetation cover in Fennoscandia was assessed by employing the variation partitioning method. With effects of climate forcings, including variations in orbital-scale insolation (ORB), melting of the ice sheets and changes in greenhouse gas (GHG) concentrations, the climate shows spatial heterogeneity both at 11.5 ka and over the course of the Holocene. At 11.5 ka, the positive summer ORB forcing overwhelms the minor negative GHG anomaly and causes a higher summer temperatures of 2–4 °C in the extratropical continents than at 0 ka. The ice-sheet forcings primarily induce climatic cooling, and the underlying mechanisms include enhanced surface albedo over ice sheets, anomalous atmospheric circulation, reduced the Atlantic Meridional Overturning Circulation (AMOC) and relevant feedbacks. In particular, the most distinct feature is a thermally contrasting pattern over North America, with simulated temperatures being around 2 °C higher than those at 0 ka for Alaska, whereas over most of Canada, temperatures are more than 3 °C lower. The geographical variability of simulated temperatures is also reflected in Holocene
temperature evolution, especially during the early Holocene, as constant Holocene cooling in Alaska contrasts with strong early-Holocene warming (warming rate over 1 °C kyr-1) in northern Canada. The early-Holocene climate is sensitive to the FWF forcings and a brief comparison with proxy records suggests that our updated FWF (FWF-v2, with a larger FWF release from the Greenland ice sheet and a faster FWF from the Fennoscandian Ice sheet (FIS)) represents a more realistic Holocene temperature scenario regarding the earlyHolocene warming and Holocene temperature maximum (HTM). Comparison of multiple simulations suggests that the multi-model differences are spatially heterogeneous, despite overall consistent temperatures in the NH extratropics as a whole. On the one hand, reasonably consistent temperature trends (a temporal pattern with the early-Holocene warming, following a warm period and a gradual decrease toward 0 ka) are found over the regions where the climate is strongly influenced by the ice sheets, including Greenland, N Canada, N Europe and central-West Siberia. On the other hand, large intermodel variation exists in the regions over which the ice sheet effects on the climate were relatively weak via indirect influences, such as in Alaska, the Arctic, and E Siberia. In these three regions, the signals of multi-model simulations during the early Holocene are incompatible, especially in winter, when both positive and negative early-Holocene anomalies are suggested by different models. These divergent temperatures can be attributed to inconsistent responses of model variables. Southerly winds, surface albedo and sea ice can result in divergent temperature trends across models in Alaska, Siberia and the Arctic. Further comparisons reveal that divergent responses in these climate variables across the models can be partially caused by model differences (e.g. different model physics and resolution). For instance, the newly adopted formulation of the turbulent transfer coefficient in CCSM3 causes an overestimated albedo over Siberia at 0 ka, which leads to a stronger early-Holocene warmth than in other models. Moreover, the relatively simplified sea ice representation in FAMOUS probably leads to overestimated sea ice cover in the Arctic Ocean. The coarse vertical resolution in LOVECLIM might also introduce strong responses in atmospheric circulation over Alaska. From the perspective of climate features, the transient feature of the early-Holocene climate driven by the retreating ice sheets also influences the inter-model comparisons, as this transient feature induces a large degree of uncertainty into the FWF forcing. Comparisons of multiple model results with compiled proxy data at the sub-continental scale of NH high latitudes (i.e. Fennoscandia, Greenland, north Canada, Alaska and Siberia) reveal regionally-dependent consistencies in Holocene temperatures. In Fennoscandia, simulations and pollen data suggest a summer warming of 2 °C by 8 ka, although this is less expressed in chironomid data. In Canada, an early-Holocene warming of 4 °C in summer is suggested by both the simulations and pollen results. In Greenland, the magnitude of early-Holocene warming of annual mean ranges from 6 °C in simulations to 8 °C in δ18O-based temperatures. By contrast, simulated and reconstructed summer
temperatures are mismatched in Alaska. Pollen data suggest 4°C early-Holocene warming, while the simulations indicate 2°C Holocene cooling, and chironomid data show a stable trend. Meanwhile, a high frequency of Alaskan peatland initiation before 9 ka can either reflect a high temperature, high soil moisture content or large seasonality. In high-latitude Siberia, simulations and proxy data depict high Holocene temperatures, although these signals are noisy owing to a large spread in the simulations and to a difference between pollen and chironomid results. On the whole, these comparisons of multi-model simulations with proxy reconstructions further confirm the Holocene climate evolution patterns in Fennoscandia, Greenland and North Canada. This implies that the Holocene temperatures in these regions have been relatively well established, with a reasonable representation of Holocene climate in the multiple simulations and a plausible explanation for the underlying mechanisms. However, the Holocene climate history and underlying mechanisms in the regions of Siberia and Alaska remain inconclusive. Variation partitioning revealed that climate was the main driver of vegetation dynamics in Fennoscandia during the Holocene as a whole and before the onset of farming. Forest fires and population size had relatively small contributions to vegetation change. However, the size of the human population became a more important driver of variation in vegetation composition than climate during the agricultural period, which can be estimated to have begun at 7–6 ka in Sweden and 4–3 ka in Finland. There is a clear region-dependent pattern of change caused by the human population: the impact of human activities on vegetation dynamics was notably higher in south Sweden and southwest Finland, where land use was more intensive, in comparison with central Sweden and southeast Finland. This thesis investigates the climate responses to the main forcings during the Holocene through various approaches, which has potential implications for the interactions between ice sheets and the climate, the Holocene climate history and current global change. The atmosphere-ocean system was sensitive to the FWF forcing during the early Holocene, implying that existing uncertainties in reconstructions of ice-sheet dynamics can be constrained by applying different freshwater scenarios via a comparison with proxy data. The Holocene climate history in most of the Northern Hemisphere extratropics is relatively well established, especially in regions that were strongly influenced by ice sheets. The implications of our investigation (on the transient early-Holocene) for the current global change are twofold. First, regional heterogeneity of the climate responses implies that regional differences should be taken into account when adapting to the current global change. Second, apart from the different scenarios of GHG forcing, inter-model comparison would be a good option to reduce model-dependency in estimation of the future climate.
Nederlandse Samenvatting Het temperatuurverloop op het noordelijke halfrond tijdens het Holoceen, ten noorden van de tropen Het Holoceen is het laatste geologische tijdvak en wordt gewoonlijk gedefinieerd als de tijd sinds 11.7 duizend jaar geleden (hierna ka). Dit tijdvak vertegenwoordigde een nieuwe fase, waarin de menselijke beschaving vorm kreeg. De laatste deglaciatie duurde voort tot diep in het Holoceen, waardoor het vroege Holoceen gekenmerkt werd door een grootschalige reorganisatie, met veranderingen in verscheidene delen van het klimaatsysteem. Door het bestuderen van het Holoceen kan inzicht worden verkregen over de werking van dit klimaatsysteem, naast theoretische kennis over de klimaatgeschiedenis van het tijdvak. Wij hebben ten eerste een set van computersimulaties uitgevoerd met veschillende combinaties van klimaatforceringen voor 11.5 ka en voor het gehele Holoceen, om zo de reactie van het klimaat-oceaan systeem op de belangrijkste forceringen te onderzoeken. Met name werden twee mogelijke scenario’s getest voor de afvoer van zoet water naar de oceaan (de zogeheten ‘freswater flux’ of FWF). Door middel van deze scenario’s werd de relatief grote onzekerheid onderzocht die gerelateerd is aan het smelten van ijskappen. Bovendien werden nog eens vier Holocene simulaties uitgevoerd met verschillende klimaatmodellen, genaamd LOVECLIM, CCSM3, FAMOUS en HadCM3. Hierbij werd gekeken naar de regio’s waar de verschillende simulaties een consistent beeld lieten zien, en waar juist niet. Vervolgens werden de oorzaken onderzocht van de afwijkingen tussen de verschillende modellen. Dit werd gedaan op twee niveau’s: voor de modelvariabelen en voor de grondbeginselen en fysica in de modellen. Na deze stap werden de verschillende modelsimulaties systematisch vergeleken met klimaatreconstructies welke gebaseerd waren op zogeheten ‘proxy data’, d.w.z. gegevens uit het geologisch archief. Dit werd gedaan voor vijf regio’s op het noordelijk halfrond, te weten Fennoscandinavië, Groenland, Noordelijk Canada, Alaska en het noorden van Siberië. De mogelijke bronnen van onzekerheid werden ook onderzocht in zowel de modelsimulaties als in de proxy data, en het meest waarschijnlijke klimaatverloop werd bepaald aan de hand van aanvullende aanwijzingen, indien beschikbaar. Bovendien werd de bijdrage van klimaatverandering aan de variaties in Holocene vegetatie in Fennoscandinavië onderzocht, naast de invloed van bosbranden en menselijke bevolkingsgrootte. Dit werd gedaan aan de hand van de zogeheten ‘variation partitioning’ methode, welke gebruikt kan worden om het effect van verschillende factoren op een proces te onderscheiden. Het Holocene klimaat werd in belangrijke mate gestuurd door 1) variaties in de ontvangen
zonnestraling als gevolg van veranderingen in de baan van de aarde om de zon, 2) het smelten van ijskappen en 3) variaties in concentraties aan broeikasgassen. De klimaatverandering als gevolg van deze drie factoren laat een grote ruimtelijke heterogeniteit zien, zowel rond 11.5 ka als gedurende de rest van het Holoceen. Rond 11.5 ka was de zomerse instraling groter dan nu en dit veroorzaakte op veel plaatsen hogere zomertemperaturen (met 2 tot 4°C verschil) dan in het preindustriële klimaat (d.w.z. 0 ka). Dit effect was sterker dan dat van de zwak negatieve forcering van de broeikasgassen in de atmosfeer. De ijskappen in Noord Amerika en Noord Europa veroorzaakten daarentegen voornamelijk een afkoeling, met name door de relatief hoge albedo, een verandering in atmosferische circulatie, een verzwakking van de circulatie in the Atlantische Oceaan en aanverwante terugkoppelingsmechanismen. Boven Noord Amerika zorgden deze twee tegengestelde effecten voor een duidelijk contrast in temperatuur tussen Alaska, waar de gesimuleerde temperatuur 2°C hoger was, en een groot deel van Canada, waar de temperatuur meer dan 3°C lager was. De ruimtelijke variabiliteit in de gesimuleerde temperaturen was ook te zien in de evolutie van de Holocene temperaturen. De constante afkoeling in Alaska contrasteerde duidelijk met de opvallende opwarming (met meer dan 1°C per duizend jaar) in het vroege Holoceen in noordelijk Canada. Het vroeg-Holocene klimaat was gevoelig voor FWF forcering, en een vergelijking met onafhankelijke temperatuurreconstructies laat zien dat een bepaald FWF scenario (FWFv2) het meest realistische Holocene temperatuurbeeld geeft, met name wat betreft de opwarming in het vroege Holoceen en tijdens het Holocene thermale maximum. Dit FWFv2 scenario gaat uit van een grotere waterafvoer vanaf de Groenlandse IJskap en een snellere afvoer vanaf de Fennoscandinavische IJskap dan een eerder scenario. Een vergelijking van meerdere simulaties maakt duidelijk dat de verschillen tussen de vier genoemde modellen ruimtelijk heterogeen zijn, ondanks de goede overeenkomst op de schaal van het gehele noordelijk halfrond. In regio’s met een sterke invloed van de ijskappen zijn de trends in temperatuur redelijk consistent tussen de modellen, met een opwarming in het vroeg Holoceen, gevolgd door een warme periode en tenslotte een geleidelijke afkoeling naar 0 ka. Deze regio’s zijn Groenland, noordelijk Canada, noord Europa en westelijk Siberië. In regio’s met een zwakke invloed van de ijskappen treden grote verschillen op tussen de uitkomsten van de modellen. Het gaat hierbij om Alaska, de Arctische Oceaan en oostelijk Siberië. In deze drie regio’s zijn de resultaten van de verschillende modellen niet in overeenstemming voor het vroege Holoceen, met name wat betreft het winterseizoen, waarvoor zowel positieve als negatieve temperatuurafwijkingen werden berekend in de verschillende modellen. Deze afwijkende temperaturen kunnen worden gerelateerd aan inconsistente berekeningen van andere variabelen. Bijvoorbeeld, verschillen in windrichting, albedo van het oppervlak en zeeijsbedekking kunnen resulteren in afwijkende temperaturen in Alaska, Siberië en het Arctisch gebied.
Verdere vergelijkingen laten zien dat dergelijke variaties in de modelresultaten gedeeltelijk veroorzaakt kunnen zijn door verschillen tussen de modellen zelf (bijv. in de modelfysika en het oplossend vermogen). Bijvoorbeeld, de recente formulering van de turbulente uitwisseling in CCSM3 veroorzaakt een overschatting van de albedo in Siberië rond 0 ka, met een relatief sterke opwarming in het vroege Holoceen tot gevolg. Daarnaast kan de overschatting van de zeeijsbedekking in de Arctische Oceaan in FAMOUS worden gerelateerd aan de relatief eenvoudige representatie van zeeijs in dat model. Bovendien kan de sterke reactie van de atmosferische circulatie in LOVECLIM worden verklaard aan de hand van de eenvoudige verticale opbouw in dit model. De grote onzekerheden in de FWF forcering, gerelateerd aan de degaciatie van ijskappen in het vroege Holoceen, maken het vergelijken van verschillende modellen echter lastig. Vergelijkingen van meerdere modelresultaten met proxy data voor verschillende regio’s op het noordelijk halfrond (te weten Fennoscandinavië, Groenland, noord Canada, Alaska en Siberia) laten zien dat de eenduidigheid duidelijk verschilt van gebied tot gebied. De proxy data omvatten met name gegevens van fossiele stuifmeelkorrels (zogeheten pollen) en muggelarven (zogeheten chironomiden). In Fennoscandinavië laten simulaties en pollengegevens een zomerse opwarming zien van 2°C rond 8 ka, terwijl dit mindelijk duidelijk is in de reconstructies welke gebaseerd zijn op chironomiden. In Canada blijkt een vroeg-Holocene opwarming van 4°C duidelijk uit zowel de simulaties als de pollengegevens. De mate van opwarming in Groenland varieert van 6°C in de simulaties tot 8°C volgens reconstructies gebaseerd op δ18O uit het Groenlandse ijs. In Alaska, daarentegen, zijn de gesimuleerde en gereconstrueerde temperaturen in tegenstelling met elkaar. Pollengegevens suggereren een vroeg-Holocene opwarming van 4°C, terwijl de simulaties juist een afkoeling van 2°C laten zien en er geen verandering is waar te nemen in reconstructies op basis van chironomiden. In Alaska is het vroege Holoceen een tijd met grootschalige veenvorming, maar dit kan meerdere oorzaken hebben, zoals een hoge temperatuur, een hoog bodemvochtgehalte of een grote seizoenaliteit. In noord-Siberië zijn relatief hoge temperaturen te zien in zowel simulaties als in de proxy data, al zijn deze waarden onzeker vanwege de grote spreiding in de simulaties en de verschillen tussen reconstructies gebaseerd op pollen en chironomiden. De vergelijkingen van de modelsimulaties met op proxies gebaseerde reconstructies bevestigen de algemene Holocene klimaatevolutie zoals deze bekend was voor Fennoscandinavië, Groenland en noordelijk Canada. Dit betekent dat de Holocene temperaturen in deze regio’s weinig onzeker zijn, blijkend uit een redelijke representatie in de modelsimulaties, en het laat zien dat er een goed begrip is van de onderliggende mechanismen. De Holocene klimaatgeschiedenis en bijbehorende mechanismen in Siberië en Alaska blijven echter onzeker.
Met ‘variation partitioning’ werd aangetoond dat het klimaat de belangrijkste sturende factor was voor vegetatieverandering in Fennoscandinavië gedurende het Holoceen, en in het bijzonder in de tijd voordat landbouw algemeen werd toegepast. Bosbranden en de grootte van de bevolking hadden beiden een relatief kleine bijdrage. Echter, de bevolkingsgrootte werd een belangrijkere sturende factor voor de vegetatiesamenstelling dan klimaat gedurende de tijd dat landbouw algemeen werd. Dit was rond 7-6 ka in Zweden, en 4-3 ka in Finland. Er is een duidelijk regio-afhankelijk patroon van verandering in vegetatie welke werd veroorzaakt door de bevolking: het effect van menselijke aktiviteit op vegetatieverandering was duidelijk groter in zuid Zweden en in zuidwest Finland, waar het landgebruik intensiever was, dan in centraal Zweden en zuidoost Finland. Dit proefschrift onderzoekt de reactie van het klimaat op de belangrijkste sturende factoren gedurende het Holoceen, gebruik makend van verschillende methoden. De resultaten hebben mogelijke implicaties voor ons begrip van de interacties tussen ijskappen en het klimaat, de Holocene klimaatgeschiedenis, alsmede de huidige mondiale klimaatverandering. Het gekoppelde atmosfeer-oceaansysteem was gevoelig voor FWF forcering gedurende het vroege Holoceen, wat impliceert dat bestaande onzekerheden over de reconstructies van ijskapveranderingen kunnen worden verminderd door het toepassen van verschillende FWF scenario’s, in combinatie met proxy data. De Holocene klimaatgeschiedenis op het noordelijk halfrond is nu goed bekend, met name in regio’s waarin de invloed van de smeltende ijskappen groot was. De implicaties van ons onderzoek voor de huidige klimaatverandering zijn tweeledig. Ten eerste, de regionale heterogeniteit van klimaatverandering moet in beschouwing worden genomen bij het bepalen van aanpassingsmaatregelen tegen klimaatverandering. Ten tweede, vergelijkingen van verschillende modeluitkomsten zijn een goede mogelijkheid om de modelafhankelijkheid van simulaties van het toekomstig klimaat te verkleinen.
Chapter 1 Introduction An appropriate climate on Earth is an essential element for the development and sustainment of life and civilization on our planet. Climate conditions expressively affect human lives, and people's livelihoods are closely connected to the local climate, as the types of consumption and housing are influenced by climate conditions. For this reason, food, water and energy supply systems are optimized to the current average climatic conditions, implying that fluctuations or trends in climate can cause serious difficulties for humanity (Rosenzweig & Parry 1994). There is increasing agreement that many decisions need to take climate change into account in order to adapt to the ongoing global changes (IPCC, 2014). In particular, the decisions concerning long-term commitments, such as urbanisation plans, risk management strategies, and infrastructure development for water management, are sensitive to climate change, and the reliability of projections of the future climate is thus relevant to these decisions (Hallegatte 2009). In other words, our society can significantly benefit from precise and reliable climate projections. For instance, a reliable projection of the duration and intensity of extreme climate (events), such as flooding, droughts and heatwaves, can reduce the inflicted damage. The practical value of reliable climate projections is further enhanced by the fact that the frequency of extreme events tends to increase (Coumou & Rahmstorf 2012) under current global warming. Reliable projections require a comprehensive and thorough understanding of the climate system (such as climate variability), as these climate projections are generated by climate models that were built using our understanding of the climate system. Climate changes in the past extend the range of the observed climate, providing an extra constraint for climate models and increasing confidence in the predicted results. Both warmer (e.g. Last Interglacial (LIG, 130–115 ka)) and cooler climates (e.g. Last Glacial Maximum (LGM, ~21 ka)) than the present can be found in geological history, which provides an opportunity to examine the response of the climate system under climate states that differ from the present. For similar reasons, periods characterized by marked climate change, such as the last deglaciation, may shed light on current global changes.
1.1 The climate system, its variation and paleoclimate change 1.1.1 The climate system and mechanisms underlying climate change Earth’s climate is a dynamic characterization of the climate system and is described by geologists as both an agent and a feature of land-surface change. The climate system includes the atmosphere, hydrosphere (ocean), biosphere (vegetation), cryosphere and land surface, with various levels of nested sub-systems (Fig. 1.1), as defined by the Global Atmospheric Research Programme (GARP) of the World Meteorological Organization in 1975. Subsequently, the essential role of the interactions between these components and associated processes in the climate system have attracted increasing attention, as reflected in the definition of the climate system given by the United Nations Framework Convention on Climate Change (FCCC) in 1992. The interactions of heat, energy and momentum between these components occur through physical, chemical and biological processes and feedbacks. For instance, the heat exchanges between the ocean and atmosphere components, mainly through latent heat and associated with the hydrological cycle, influence the heat distribution. Meanwhile, albedo-associated feedbacks link the components of atmosphere and cryosphere together and influence the radiation budget of the climate system. These multiple components of the climate system and their interactions determine climatology as an interdisciplinary field, of which physical processes are controlling the Earth’s climate.
Figure 1.1 A schematic illustration of the components and interactions in the climate system (derived
from Le Treut et al. 2007).
The climate system is a dynamic system, with exchanges of radiant energy, mass, water and, to a lesser extent, carbon and nitrogen among different components. The fluxes of these factors are vectors, as they involve the movement of some quantity from one place to another, and the direction of flow is also crucial to balancing the system (e.g. Trenberth & Stepanial 2004). The net fluxes differ considerably as a function of the time period considered, which results in a dynamic climate over given regions. Climate forcings can exert impacts on these fluxes at various time scales, and thus influence climate conditions. The various climate forcings fall into two types with reference to the climate system itself: external and internal forcings (McGuffie & Henderson-Sellers 2005). External forcings tends to push the climate system to a more stable state that is in equilibrium with the new force, while the forcing itself is not influenced by this forcing. Orbital-scale variation in insolation is one of these external forcings and provides an important explanation for the temporal climate variations of the late Quaternary (e.g. Berger 1978). The astronomical parameters of the Earth’s orbit, including eccentricity, obliquity and precession, determine the solar radiation received at the top of the atmosphere and its distribution over latitudes (Berger 1978). The periodic changes in these parameters primarily explain the periodic alternation of glacial and interglacial periods throughout geological history, part of which are explained by the Milankovitch theory. By contrast, internal climate forcings involve a variety of processes occurring within the climate system, implying that these processes not only impact on the climate condition, but they can also be influenced by the climate state (McGuffie & Henderson-Sellers 2005). For instance, changes in ocean circulation influence the climate locally and globally by adjusting heat and energy transportation, and the strength (changes) of ocean circulations also partially depends on spatial patterns of temperature and atmospheric circulation. However, the boundary between the external and internal forcings is changeable in certain circumstances. For instance, variations in atmospheric CO2 influence the climate through the well-known greenhouse effect and are controlled by carbon fluxes between different components of the climate system. The changing CO2 concentration can be both an external and an external forcing, depending on the time scale of interest. On short time scales (e.g. shorter than the millennial scale), CO2 variations in the atmosphere are an external forcing, which serve as an agent of climate change and are not directly affected by the climate at this time scale. Meanwhile, on a long geological time scale (e.g. a million years or longer), variation in CO2 can also be an internal forcing, as carbon storage in the deep ocean and the atmosphere are linked through the long-term carbon cycle (e.g. Boyle & Keigwin, 1985). Temperatures play an important role in this cycle, such as through their influence on the strength of the oceanic circulation, which impacts on CO2 exchange between the oceans and atmosphere, and temperatures also affect carbon storage in the biosphere, some of which can be ultimately deposited in oceans through geological
CHAPTER 1 movements. Similarly, anthropogenic-related changes in atmospheric CO2 concentrations should be regarded as internal processes when considering these changes on a geological timescale. However, these anthropogenic-related changes can also act as external forcings for climate change on decadal or centennial timescales, because the burning of fossil fuels injects extra greenhouse gases and sulphate aerosols into the atmosphere, and humans change land surface properties through agricultural activities, which exerts critical impacts on climate. However, the intensification of these changes is barely influenced by the climate on this short timescale. Together with these internal and external climate forcings, feedbacks also make important contributions to the variability of the climate system. Feedback occurs when the response of a system to an input further alters the input variable of the system through the involved processes. Through this loop, feedbacks can either amplify forcing-induced changes through positive responses, or dampen the forcing-based changes via negative reaction mechanisms. For instance, albedo-related feedbacks due to changes in the land ice and sea ice have played an essential role in glacial and interglacial transitions by influencing the radiation budget of the Earth (Deblonde et al. 1992; Bintanja et al. 2002). The high albedo of ice both on the land and in the oceans reduces the amount of solar radiation absorbed, which facilitates the formation of ice and triggers positive feedbacks (Curry et al. 1995). In the climate system, some original changes even trigger both negative and positive changes at the same time, and the net effect can therefore either lead to an enhanced or a reduced response, depending on which one is dominant. For instance, an enhanced amount of water vapour in the atmosphere can both increase and decrease the original change in water vapour. On the one hand, more water vapour in the atmosphere leads to a warming climate through the greenhouse effect of water vapour, and increases the evaporation of surface water, which introduces more water molecules into the atmosphere. On the other hand, increased water vapour in the atmosphere accelerates cloud formations, thus reflecting more sunlight back into space and lowering the temperature, which ultimately limits water molecular movement into the atmosphere. Therefore, the final effect of water vapour on the climate depends on the net effect of these two opposite processes, which strongly depends on the vertical distribution of clouds and can overall be positive in the tropics (Inamdar and Ramanathan 1998; Mcguffie & Henderson-sellers 2005). The distinctions between climate forcings and feedbacks are flexible, as they depend on the timescales of processes and the period of consideration. Commonly, a process is regarded as a forcing (boundary condition) when its operating time is much longer than the time scale of interest, which can vary from years to decades (such as feedbacks related to snow, sea ice, the upper ocean, dust and aerosols) to millions of years (such as weathering and the evolution of vegetation) (Palaeosens Project members, 2012). Similarly, the process serves as a feedback when its timescale is shorter than the time period of interest.
1.1.2 Palaeoclimate change and the Holocene With the effects of all these external and internal forcings and feedbacks, the regional and local climate varies spatially, which is a function of the geographic location (e.g. general latitudinal patterns with positive gradients from low latitudes toward high latitudes). Meanwhile, the climate system as a whole is also temporally dynamics and has fluctuated in a distinctive way over the history of the Earth, owing to various scales of temporal variation in these factors. Natural variation in climate is not simply the sum of the effects of these factors, but is the result of non-linear responses in linked climatic components to these changes. Throughout the history of the Earth, the temperature has undergone considerable changes. As recorded by multiple proxies, the climate history is punctuated by multiple glacial and interglacial cycles. For example, temporal variation in δ18O in benthic foraminifera has shown cyclic changes with an amplitude of about 2‰ during the last 600 ka (EPICA Community Members, 2004; Lisiecki et al. 2005). These changes imply that temperatures fluctuated by more than 3–4°C (Johnsen et al. 2001; Jouzel et al. 2003). Combining different proxies gives an overview of the temperature history of the Earth. For instance, from the last interglacial (LIG, ~130 ka) to the last glacial maximum (LGM, ~21 ka), the temperature dropped by 5–6 °C (CAPE Project, 2006). Following the LGM, the climate warmed up with ongoing deglaciation, although this was accompanied by abrupt cooling events (Hughen et al. 1996; Peltier 2004; Shakun et al. 2012). During the last deglaciation, the climate system experienced a major reorganization that lasted well into the Holocene. As the latest epoch of the Earth history, the Holocene refers to the last 11.7 ka and represents a relatively warm climate in comparison with preceding periods. The Holocene embraces the time span of human civilization, such as the evolution of agriculture with the domestication of plants and animals and its spatial expansion (Gupta 2004). The land surface has became progressively covered by forests as the ice retreated (e.g. MacDonald et al. 2000; CAPE project, 2001). In the late Holocene, the forest cover decreased again as mankind’s demand for timber and agricultural land grew (e.g. Kaplan et al. 2009). In particular, the transition in the climate system during the early Holocene is analogous to currently ongoing global change in terms of the overall warming trend of global temperatures and transient features of the climate system, even though they have different causes. In addition, according to Vaughan et al. (2013), the total volume of ice currently existing on land is equivalent to a rise of more than 60 m in the sea level, including the Antarctic (58 m) and Greenland (7 m) ice sheets. This is similar to the volume of melting ice during the early Holocene (Lambeck et al. 2014). In addition to these similar magnitudes of ice, the mechanisms behind the melting ice sheets are also the same, such as the effects on albedo, atmospheric circulation and perturbation of meltwater, although these ice sheets are situated in very different geographical locations. Therefore, studies on the Holocene can accumulate information on how the climate system functions in response to changing climate forcings, thus providing insights into the current global
CHAPTER 1 change.
1.2 The early-Holocene transition and Holocene climate forcings 1.2.1 The early-Holocene transition in the climate system During the last deglaciation, the climate system experienced a major reorganization that lasted well into the Holocene, for which relatively abundant proxy records are available. In particular, the early part of the Holocene (11.5–7 ka) was characterized by transitions in various components of the climate system (Fig. 1.2). In the cryosphere, the Laurentide Ice Sheet (LIS) and Fennoscandian Ice Sheet (FIS) continuously waned until they finally disappeared around 6.8 ka and 10 ka, respectively (Dyke et al. 2003; Occhietti et al. 2011; Cuzzone et al. 2016). At the onset of the Holocene, the LIS and FIS partially covered N America and Fennoscandia, with a maximum thickness of up to almost 2 km and 200 m respectively, as evidenced by multiple lines of geological data (Peltier 2004; Ganopolski et al. 2010). The total amount of melt water from these ice sheets was equivalent to a sea-level rise of up to 60 m over the course of the Holocene (Lambeck et al. 2014). The presence of the decaying ice sheets also induced temporalspatial heterogeneity in the earlyHolocene climate, as these retreating ice sheets exerted multiple influences on the climate system (Renssen et al. 2009).
Figure 1.2 The Holocene period and early Holocene transition in the climate system. (A) Planktonic foraminifera δ18O at the core (OCE326-GGC5) from the subtropical Atlantic, representing the nearsurface temperature and/or local salinity (McManus et al. 2004). (B) Estimated 231 Pa/230Th, indicating the strength of AMOC (McManus et al. 2004); (C) δ18O of NGRIP (Vinther et al. 2006); (D) Ice volume locked in the land, shown as the equivalent sea level (m) (Lambeck et al. 2014); (E) Radiative forcing of greenhouse gases, shown as for CO2 and total GHG (CO2, CH4 & N2O) (Luthi et al. 2008).
Proxy-based vegetation reconstructions also suggest a transition in the biosphere during the early Holocene. The biomass of vegetation tends to increase after the retreat of ice sheets, even though the absolute magnitudes of changes are spatially heterogeneous. Vegetation reconstructions have revealed a northward expansion of boreal forest in the circum-Arctic region by 9–8 ka (MacDonald et al. 2000; CAPE project, 2001; Fang et al. 2013). This expansion of boreal forest into regions that were not previously vegetated or were covered by tundra caused a reduction in the surface albedo and induced a positive feedback in the warming trend (Claussen et al. 2001). Therefore, these interactions between the climate and vegetation composition also had important influences on the transient climate during the early Holocene. The ocean, one of the main constituents of the climate system, also experienced a critical transition, as indicated by marine sediment core data and numerical modelling. Firstly, multiple proxies have suggested that considerable changes in sea surface temperatures (SSTs) occurred during the early Holocene. For instance, the variation in stable isotopes (δ18O and δ13C) measured from planktonic foraminifera (Fig. 1.2) reflects a rise in SSTs in the North Atlantic (Bond et al. 1993; Kandiano et al. 2004; Hald et al. 2007). Moreover, both proxy (e.g. 231Pa/230Th measurements on sediment cores) and modelling studies have found that the Atlantic Meridional Overturning Circulation (AMOC) was relatively weak during the early Holocene due to ice-sheet melting (McManus et al. 2004; He et al. 2010; Blaschek & Renssen 2013), which led to reduced northward heat transport and extended sea-ice cover (Renssen et al. 2010; Roche et al. 2010; Thornalley et al. 2011, 2013). Important adjustments in the carbon cycle occurred in the early Holocene. The carbon cycle describes the exchange of carbon between the carbon pools of the biosphere, pedosphere, geosphere, hydrosphere and atmosphere, implying that carbon can be reused through biogeochemical processes. Changes in the global biomass and ocean temperatures and salinities control the exchanges between the main global carbon reservoirs (the oceans, terrestrial biosphere and atmosphere). CO2 is a well-mixed atmospheric trace gas responding to carbon flux variations in both the Southern and Northern hemispheres. During the early Holocene, for example, atmospheric CO2 rose by 20–30 ppm that is equivalent to 0.3 W m-2 increase in radiative forcing (Fig. 1.2) (Luthi et al. 2008). CO2 reconstructions based on stomatal indices suggest even larger variabilities (Wagner et al. 1999; Jessen et al. 2007). The reorganization of the ocean circulation during the early Holocene (Jessen et al. 2007) has been suggested as the main cause of these CO2 changes, such as through influences on the deep water upwelling in the Southern Hemisphere, enhancing the ocean-to-atmosphere CO2 flux. These transient climatic components are actively interconnected by multiple processes (Fig. 1.1). For instance, the dynamic early-Holocene climate could have impacted on CO2 concentrations in the atmosphere via the carbon cycle, and CO2 concentrations have
CHAPTER 1 inverse influences on the climate via the well-known greenhouse effect. Meanwhile, ice sheets served as both causes and effects of changes in the climate system. Dynamic ice sheets are sensitive to climate changes, which also exert multiple impacts on the atmosphere–ocean system. Investigation of the climate response to decaying ice sheets during the early Holocene could shed light on these interactions. 1.2.2 Main forcings of the Holocene climate Orbital-scale solar insolation (ORB) has played a role in the climate history of the Earth (Berger 1988). During the Holocene, orbital-induced summer insolation has showed a decreasing trend at mid- and high latitudes (Berger 1978; Denton et al. 2010). Meanwhile, the melting FIS and LIS also exerted in important influence on Holocene climate evolution until they eventually disappeared around 10 and 6.8 ka, respectively, which further adjusted the spatial patterns of the climate (Dyke et al. 2003; Renssen et al. 2009; Blaschek and Renssen 2013; Cuzzone et al. 2016). First, the surface albedo was much higher over the ice sheets compared to ice-free surfaces, which resulted in relatively low temperatures. Second, the ice-sheet topography could have also influenced the climate through the adjustment of atmospheric circulation (Felzer et al. 1996; Justino & Peltier 2005; Langen &Vinther 2009). A large-scale ice sheet could have generated a glacial anticyclone, which could have locally reduced the temperature over the ice sheet (Felzer et al. 1996), but it may also have caused warming by 2–3 °C over the North Atlantic during the LGM (Pausata et al. 2011; Hofer et al. 2012). Third, it has been found that melting ice sheets caused a weakening of the maximum AMOC during the early Holocene, which led to reduced northward heat transport and extended the sea-ice cover, and thus influenced the high-latitude climate (Renssen et al. 2010; Thornalley et al. 2011, 2013). Overall, the net effect of ice sheets on the early-Holocene climate can be expected to have tempered the orbitally-induced warming at mid- and high latitudes. Additionally, radiative forcing due to variation in the concentration of greenhouse gases (GHG) in the atmosphere has also affected the Holocene temperatures. From the onset of the Holocene, the total effect of GHG dominated by CO2 showed a rapid rise by 10 ka, reaching a local peak of -0.3 W m-2 relative to the pre-industrial period, which was followed by a slight decrease to a minimum at 7 ka, before gradually increasing towards 0 ka (Loulergue et al. 2008; Joos & Spahni 2008; Schilt et al. 2010). 1.2.3 Previous studies on the Holocene climate and remaining problems The impact of these forcings on the Holocene climate has been partially examined in previous modelling studies. Renssen et al. (2009) found that the Holocene climate was sensitive to the ice sheets and that LIS cooling effects delayed the Holocene Thermal Maximum (HTM) by up to thousands of years. Blaschek and Renssen (2013) later revealed that melting of the Greenland ice sheet (GIS) had an identifiable impact on the climate
over the Nordic Sea. However, the focus in these studies has been on the period after 9 ka by investigating the influence of the decay of the LIS and GIS on the climate relative to other climate forcings (Renssen et al. 2009; Blaschek & Renssen, 2013). The most important challenges in simulating climate during the initial phase of the early Holocene are the inherent uncertainties in the ice-sheet forcings in terms of the ice-sheet dynamics and particularly the associated meltwater release. The ocean–atmosphere system is sensitive to freshwater release, including the total volume, the location, timing and rate (Roche et al. 2007). Different approaches appear to suggest slightly different rates of icesheet melting. For instance, recent deglaciation studies based on cosmogenic exposure dating indicate slightly older ages of deglaciation in some regions than suggested by radiocarbon dating data (Carlson et al. 2014; Cuzzone et al. 2016), primarily because of large uncertainty in the bulk organic sample ages and the possibility of old carbon contamination (Stokes et al. 2010; Carlson et al. 2014). Furthermore, the Younger Dryas stadial ended at 11.7 ka (Fig. 1.2), and may still have influenced the early-Holocene climate due to the long response time of the deep ocean (Renssen et al. 2012). Ref
Table 1.1 Inter-model and model–data comparison studies included in PMIP Period Topic
Braconnot 2007 Gladstone 2005 Kageyama 2013 Harrison 1998 Braconnot 2000
MH & LGM MH MH & LGM MH MH & LGM
Comparison of PMIP1 & PMIP2 NAO & SLP (compared the present day) Comparison of IPSL_CM4 & IPSL_CM5A Vegetation and climate (T & P) African monsson (inter-model)
The last 1 kyr The last 1 kyr The last 2 kyr & LIG MH MH MH MH
Inter-model comparison Inter-model comparison of EMICs Multi-model–data comparison Model–data comparison in Europe Data–model comparison in Europe Data–model comparison at high latitudes East Asian summer monsoon
Apart from these modelling studies, the mid-Holocene (MH, 6 ka) has been relatively well investigated by the Palaeoclimate Modeling Inter-comparison Project (PMIP) in terms of inter-model comparisons and model–data comparisons, as summarized in Table 1.1. In order to evaluate the uncertainties of Holocene simulations regarding model-dependent variations, studies have compared multiple simulations with a focus on snapshot experiments for the mid-Holocene (Harrison et al. 1998; Braconnot et al. 2000, 2007). Other currently available transient inter-model comparisons have examined periods shorter than the whole Holocene, such as 8–2 ka and the last 2 kyr (Bothe et al. 2013; Eby et al. 2013; Bakker et al. 2014). Meanwhile, PMIP has also conducted several model–data comparisons to investigate the different processes, with a focus on the mid-Holocene as well (e.g. Masson et al. 1999; Bonfils et al. 2004; Brewer et al. 2007; Zhang et al. 2010;
CHAPTER 1 Jiang et al. 2013). Liu et al. (2014) have also recently compared model results with proxybased reconstructions to investigate the contradiction of Holocene temperature trends (i.e. the reconstructed cooling and the simulated warming during the Holocene on large-scale climate change, such as over 30–60°N and 60–90°N), although some of their simulations did not include FWF forcing. However, it remains unclear how similar or different model simulations of the early Holocene climate are, and how these results compare with proxybased datasets also remains unknown. Nevertheless, this is important to examine in detail, as the early Holocene provides an analogous context for the current global warming.
1.3 Methods to study Holocene climate 1.3.1 The LOVECLIM climate model With the rapid development of computing power, climate models have become useful tools to investigate climate changes on various timescales. Climate models are mathematical representations of our understanding of the climate system, including movements of heat and mass within components of the climate system and also interactions between different components (e.g. via physical processes). Various (e.g. dynamical) equations are essential elements of climate models, conducting model simulations thus means solving these equations with given forcings and boundary conditions. The testing of different scenarios, termed sensitivity test, allows us to explore the plausible mechanisms behind climate changes and to analyse the temporal-spatial variability. We used the LOVECLIM climate model, a three-dimensional Earth system model, to investigate the Holocene climate history. LOVECLIM stands for the five coupled subsystems, namely LOCH, VECODE, ECBILT, CLIO and AGISM models, representing the carbon cycle, vegetation, atmosphere, ice sheets and the ocean components, respectively (Fig. 1.3). The atmospheric component, ECBILT, is governed by the quasi-geostrophic potential vorticity equation (Opsteegh et al. 1998), with three vertical layers (at 800, 500 and 200 hPa) and T21 (~5.6°) horizontal resolution. Processes are simplified at various degrees. For instance, the humidity in the atmosphere is determined by the total precipitable water content between the surface and 500 hPa. The diabatic heating due to radiative fluxes, the release of latent heat, and the exchange of sensible heat with the surface is parameterized. The land surface model is also part of ECBILT, including the moisture budget, snow cover and the surface albedo. The surface albedo is a function of the fraction of the grid box covered by ocean, sea ice, trees, desert and grass. The ocean module is represented by CLIO, which couples a 3-layer thermodynamic–dynamic sea-ice model with an ocean general circulation model (GCM). The ocean GCM has a free-surface with 20 vertical layers and 3° × 3° (lat×lon) resolution (Goosse & Fichefet 1999). The albedo formulation in the sea-ice model considers the state of the surface (frozen or melted) and the thickness of the snow and ice covers. The momentum balance of sea ice results
from dynamical interaction with the atmosphere and the ocean within a two-dimensional continuum frame. The coupled ECBILT-CLIO model further includes a terrestrial vegetation component that is represented by VECODE (Vegetation Continuous Description model). The VECODE model consists of three sub-modules: a vegetation structure model that calculates plant functional type fractions (PFTs), a biogeochemical model that estimates net primary productivity (NPP) and a vegetation dynamics model. With these sub-structures, dynamic vegetation cover is obtained in terms of the PFT (such as desert, trees and grasses) in each land grid cell, according to their climate variables (Brovkin et al. 1997). The cryosphere component is the three-dimensional thermo-mechanical AGISM (Antarctica and Greenland Ice-Sheet Model) (Huybrechts 2002). AGISM simulates dynamic ice sheets by taking into account the ice flow, the solid Earth response and the mass balance at the ice-sheet surface and ice–ocean interface. The terrestrial carbon cycle on land is represented in VECODE by dynamic carbon pools on an annual time step. The oceanic carbon cycle is represented by LOCH (Liege Ocean Carbon at Heteronomous), a three-dimensional model that simulates carbon exchange between the atmosphere and ocean, and also the biological pump processes (Mouchet et al. 1996). However, it is important to note that LOCH and AGISM were not activated in our version of LOVECLIM. Accordingly, the ice-sheet configuration and greenhouse gases are prescribed based on reconstructions, as explained in Figure 1.3. A more detailed description of LOVECLIM can be found in Goosse et al. (2010a). The reasons for performing Holocene simulations with the LOVECLIM climate model are multiple, mainly including its reasonable performances for the present-day climate, its capability to simulate climate change in the past and its computational efficiency, which allows several thousands of years of transient simulation be performed in a reasonable time. The performance of LOVECLIM has been systematically examined by Goosse et al. (2010a), who found that the model simulates the modern climate reasonably well, despite some minor biases in the tropics. To be more specific, the key variables of the climate system in the simulation generally agree with the observations. The model can reproduce the main characteristics of the observed surface temperature distribution. For instance, the LOVECLIM results are in good agreement with the observed location of the 0 °C isotherm, even though the 25 °C isotherm is located too far away from the equator with an overestimated temperature. LOVECLIM simulates reasonably well the large-scale pattern of near-surface circulation. The key patterns of pressure fields in the model roughly agree with observations, such as a decrease of 800 hPa geopotential height with latitude in the N Atlantic and N Pacific, despite underestimated gradients around the Aleutian low. The sea ice extent in LOVECLIM agrees more closely with observations in the N Pacific than in the N Atlantic, since the seasonal change produced by the model is too weak in the Baffin Bay and the Labrador Sea. Finally, the strength of the ocean circulation in LOVECLIM shows relatively good agreement with data-based estimates and with other models
CHAPTER 1 (Ganachaud & Wunsch 2000; Goosse et al. 2010a). For instance, the maximum AMOC strength reaches 22 Sv (Sverdrup = 1×106 m s-1), with deep convection occurring both in the Greenland-Norwegian Sea and the Labrador Sea. These values are in relatively good agreement with the data-based estimates and at the high end of the values given by other models (Ganachaud and Wunsch 2000; Gregory et al. 2005; Goosse et al. 2010a).
Figure 1.3 The climate components of the LOVECLIM model (adapted from Goosse et al. 2010a). Filled arrows depict the interactions between these components. Dashed borders and crossed-out arrows indicate inactive components and associated interactions, and open colour arrows lead to corresponding procedures in the present study.
LOVECLIM is capable of simulating plausible climates in the past, which is helpful to explain the climate variability discovered in proxy records. For instance, it has been used in PMIP2 and PMIP3 studies to investigate the climate evolution during the LIG (Bakker et al. 2014; Loutre et al. 2014), LGM (Roche et al. 2012), and the Holocene (Renssen et al. 2009, 2012; Blaschek & Renssen 2013). Climate sensitivity to radiative forcing in LOVECLIM is about 2 °C (i.e. a sea surface temperature increase by 2°C after 1000 yr with a doubling of the CO2 concentration, according to Goosse et al. 2010a), which is at the lower end of GCM estimates (Flato et al. 2013). The sensitivity of the AMOC in LOVECLIM to perturbations of idealized freshwater fluxes (FWF) agrees with other models. For instance, in response to a 0.1 Sv freshwater release in the N Atlantic, the AMOC deceases by about 20–30%. This agrees well with the ensemble mean which includes several AOGCMs and EMICs (Stouffer et al. 2006).
1.3.2 Sensitivity experiments Sensitivity experiments are a set of simulations performed with different forcings and/or boundary conditions relative to a control experiment. Systematic comparison of these sensitivity experiments can improve our understanding of the variability of the climate system and can accumulate our knowledge on the sensitivity of the climate to these changes in forcings and boundary conditions. This understanding is essential for the development of climate theory. The sensitivity experiments have huge potential to explore climate change in the past when no observations are available. Uncertainties in modelbased palaeoclimate estimations are partially caused by uncertain forcings, which can be evaluated through sensitivity test. The first step is to perform sensitivity experiments with all possible forcing-scenarios. After this, simulated variables can be compared with evidence that is independent of the forcing data used in order to determine the most likely scenario. For instance, previous studies have investigated model responses to varying forcings, such as GHG and ORB, under the different benchmarking periods of the LGM (representing the latest glaciation) and MH (a warm period with abundant proxy data), and explored the potential role of different processes and feedbacks (e.g., Hewitt & Mitchell 1997; Kutzbach & Liu 1997; Weaver et al. 1998; Claquin et al. 2003). Comparison of these experiments provides insights into the potential mechanisms behind various responses and into functioning of the climate system under different conditions. Another good example of sensitivity experiments is related to multiple FWF scenarios. Numerous studies have examined the sensitivity of the atmosphere-ocean system to different FWF scenarios, improving understanding of the feedbacks and underlying mechanisms of abrupt climate change events that punctuated the warming trend of the last deglaciation (Teller et al. 2002; Roche et al. 2007). In Chapter 2, we conducted sets of simulations with different combinations of climate forcings, as outlined in Figure 1.4. First, two equilibrium experiments for 11.5 ka, namely OG11.5 and OGIS11.5, were performed with identical ORBGHG forcings, and additionally including the effect of melting ice sheets in the latter experiment (OGIS11.5). The spatial patterns of the early-Holocene climate and the underlying mechanisms were analysed by comparing different responses of associated variables (e.g. surface albedo, atmospheric circulation and AMOC) in these two experiments. Moreover, the temporal features of Holocene temperature were tracked by analysing transient Holocene simulations. Similarly to the equilibrium experiments, the first transient simulation, namely ORBGHG, was forced only by the corresponding ORBGHG forcing. Another two sets of simulations were conducted by further including the ice sheets. These two simulations, namely OGIS_FWF-v1 and OGIS_FWF-v2, tested two possible FWF scenarios, considering the large uncertainties in FWF discharges. The differences between them were a larger FWF release from the Greenland ice sheet and a faster FWF from the FIS in version 2 than in version 1 (Fig. 2.3). By comparing these simulations, the associated
CHAPTER 1 processes and feedbacks regarding the responses of the ocean–atmosphere system to these forcings and how these responses ultimately influenced the Holocene temperature history were investigated. 1.3.3 Inter-model comparisons Uncertainties in simulated palaeoclimates can sometimes be rather large, especially for transitional climate periods such as the early Holocene. The magnitude of the uncertainty in paleoclimate simulations depends on not only on the forcings used in the simulations, but also on the model uncertainty, such as the physical principles, complexity and resolution (Hawkins & Sutton 2009). Different models may have slightly different sensitivity estimates under certain climate conditions, as various feedbacks may be involved at different spatial and temporal scales. Multi-model comparisons provide the option to validate the reliability of model performance and increase the confidence in simulations. The reliability of the model and simulations will increase if independent models consistently indicate the same or similar results. Otherwise, varying model performances or even biases will be implied. PMIP has performed a wide range of inter-model comparisons that cover different periods with various climate forcings or processes. Stage 1 of PMIP was designed to test the climate models (atmosphere component) in the context of the LGM by investigating the climate response to large ice sheets, cool oceans and lower GHG concentrations, and also to investigate vegetation distribution during the MH (Harrison et al. 1998; Braconnot et al. 2000). Stage 2 applied coupled ocean–atmosphere and ocean–atmosphere–vegetation models to examine the role of oceans and vegetation or the land surface (Harrison et al. 2003; Braconnot et al. 2007; Jiang et al. 2008; Wohlfahrt et al. 2008). Stage 3 was extended to include the LIG and the later Holocene (Braconnot et al. 2012) and transient simulations (130–125 ka & 8–2 ka) (Bakker et al. 2014). Overall, PMIP inter-model comparisons have mainly been conducted for certain periods with snapshot experiments, and the currently available transient inter-model comparisons mainly focus on relatively warm periods (e.g. the last 1 kyr and the LIG (Bothe et al. 2013; Eby et al. 2013; Bakker et al. 2014). For the Holocene, the current multi-model comparisons studies, however, are limited to the MH by analyses of equilibrium simulations (Braconnot et al. 2012) and short transient simulations during the last 2 ka (Bothe et al. 2013; Eby et al. 2013). This implies that multi-model comparisons for the early-Holocene climate remain unexamined. To evaluate simulations of the Holocene, we compared four simulations performed with LOVECLIM, CCSM3, FAMOUS and HadCM3 in Chapter 3, as explained in Figure 1.4. More information on these models and associated simulations is provided in Tables 3.1 and 3.2. The simulations were firstly compared in the time windows of 11.5 and 6 ka in order to determine the basic spatial pattern. To track the temporal variability of the simulations,
we compared the transient temperature trends among these simulations over an array of regions to identify the regions where the multi-model simulations are consistent and where they are not. Furthermore, we diagnosed the models’ variables when mismatches were found, to determine the potential causes of these inconsistencies. After this, these multimodel discrepancies were analysed at a deeper level (e.g. why models overestimated or underestimated these variables from the perspective of the model principles) by further referring to relevant studies by others.
Figure 1.4 Schematic illustration of the approaches employed in the thesis. The four different colour tones illustrate the main procedures applied in the four chapters, with orange representing Chapter 2, blue indicating Chapter 3, green depicting Chapter 4 and yellow denoting Chapter 5.
1.3.4 Proxy records and model–data comparisons Proxy-based reconstruction is another well-developed approach for investigating the palaeoclimate, in addition to numerical model simulations. Climate parameters (e.g. temperature) can be reconstructed based on various proxies (e.g. pollen and chironomids) through transfer functions that are based on the relationship between the climate and biological datasets. Transfer-function-based reconstructions facilitate the conversion of past fossil assemblages to quantitative temperature, precipitation and other climate
CHAPTER 1 parameters, assisting direct comparison with model results. Considerable improvements have been achieved with this proxy-based method. In particular, an abundance of climate records are available for the Holocene (e.g. as collected in Shakun et al. 2012; Marcott et al. 2013; Sundqvist et al. 2014), which provides multiple lines of evidence for the climate history. However, proxy data still contain inherent uncertainties or even biases. Comprehensive comparisons of proxy-based reconstructions with model simulations provide opportunities to detect and investigate these inherent uncertainties, and thus are considered to facilitate better climatic interpretations of proxy results. Meanwhile, although climate models are useful for piecing together separate of proxy records and understanding the mechanisms behind climate change, the development of climate models still requires proxy data under certain circumstances. In particular, proxy data are needed to validate the climate models at an early developmental stage (Braconnot et al. 2012) and to evaluate the simulations when multiple models perform differently. This combination can add value to our understanding of climate mechanisms through reciprocal validation of these two approaches, as the principles of model simulation and proxy-based reconstruction are different and independent. Recent progress on proxy-based reconstructions and newly established databases provided the basis for a systematic investigation of Holocene temperature evolution. For instance, based on the Holocene proxy database (Sundqvist et al. 2014), temperature changes in the north Atlantic–Fennoscandia (Sejrup et al. 2016), Alaska (Kaufman et al. 2016), the Canadian Arctic and Greenland (Briner et al. 2016) have recently been examined. Studies have also compared model results and proxy data (e.g. Liu et al. 2014 who focused on large-scale climate change), but comparisons between multi-model simulations and proxy data on a detailed sub-continental scale have not yet been conducted. Meanwhile, considering the transient features and high uncertainty of the early-Holocene climate, comparisons of climate model results with the proxy data at a sub-continental scale are needed to evaluate Holocene simulations and to improve our understanding of Holocene climate evolution. Chapter 4 presents a systematic model–data comparison (i.e. comparison between composite reconstructions and multiple model simulations) over different regions for Holocene temperature trends, as illustrated in Figure 1.4. We first selected 36 pollen-based and 19 chironomid-based temperature records, and six δ18O records of ice cores from the NH high latitudes with consideration of the location, time frame, temporal resolution and dating interval (as explained in Chapter 4). These individual proxy records were compiled into five composite regional reconstructions for Fennoscandia, Greenland, north Canada, Alaska and high-latitude Siberia. We compared these composite reconstructions with the model simulations (ensemble mean and spread of multi-simulation) over different regions in terms of Holocene temperature trends. In addition, potential uncertainty sources were
also analysed from both model simulations and proxy data perspectives, and then used to identify the most plausible climate history of different regions with the aid of additional evidence when available. 1.3.5 Variation partitioning Variation partitioning is a statistical method to partition the variation of a given response variable into a set of explanatory variables (descriptors) whose interaction often results in an overlaid effect (Borcard et al. 1992). The method was originally used to determine how much variation in species abundance data can be simultaneously related to the explanatory variables of environmental and spatial components (Borcard et al. 1992). Recently, it has been used in palaeoecology to investigate vegetation history. For instance, it has been employed to investigate the relative contribution of the climate, forest fires and human population size to variation in Holocene vegetation dynamics (e.g. Reitalu et al. 2013; Kuosmanen et al. 2015, 2016). With the final melting of the FIS by around 10 ka, the development of vegetation development in Fennoscandia was one of the critical processes that occurred over the course of the Holocene, which is relevant for multiple factors. In Chapter 5, we use the statistical method of variation partitioning to assess the relative importance of climate, forest fires and human population size on the variation in actual vegetation cover of different PFTs (Fig. 1.4). The analysis uses PFTs derived from the REVEALS (Regional Estimates of VEgetation Abundance from Large Sites model) dataset as response matrix, and climate, human population size and forest fires as explanatory variables. The REVEALS model (Sugita 2007) was used to reconstruct the regional plant abundances from pollen percentage data, which reduces biases in pollen analysis caused by intertaxonomic differences in pollen productivity, dispersal and depositional characteristics, as well as basin size. The climate data were from the LOVECLIM simulation forced by the OGIS_FWF-v2 scenario, which is a realistic scenario regarding early-Holocene warming and timing of the HTM. The frequency of forest fires was reconstructed based on sedimentary charcoal records from 19 lakes, which provide a signal of Holocene biomass burning across the study region. The human population size variable was derived from the 14 C dating results of archaeological findings obtained from the archeological sites. Variation partitioning analysis with climate, human population size and fire as explanatory variables was first performed for the whole study period in all regions (S & C Sweden, SW & SE Finland). To further track the transient changes in the contributions of these factors, variation partitioning was performed in two sub-periods, before and after the onset of farming, which are regionally dependent. A moving window approach (Reitalu et al. 2013) was applied to S Sweden and SE Finland to assess how the relative importance of climate and human population size changed through time.
1.4 The research questions The topic of the thesis is the Holocene temperature history, and different approaches are adopted in the relevant chapters, as illustrated in Figure 1.4. These chapters are arranged in the order in which they address the following research questions. How does the Holocene climate in the NH extratropics respond to the main forcings? (Chapter 2) This general question is addressed from both spatial and temporal perspectives. For the spatial aspect, the question of how the climate in equilibrium experiments responds to the main forcing at the onset of the Holocene is a key issue. This issue can be further disentangled as follows: a) What are the main spatial characteristics of temperature at the onset of Holocene under different climate forcings, especially for ice-sheet decay? b) What would the possible mechanisms be? The climate forcings show various spatial patterns during the Holocene. Thus, answering question a) can enable us to disentangle the impact of these key forcings on the Holocene climate. Question b) can guide us to analyse the response of various climate variables, such as surface albedo, atmospheric circulation and AMOC strength. From the temporal perspective (in transient simulations), the specific sub-questions include the following: a) What are the temporal trends of Holocene temperatures over different regions? b) How much do different forcings contribute to these temperature trends? c) Which FWF scenarios provide more a plausible Holocene temperature evolution regarding the early-Holocene warming and timing of the HTM? With these temporally varied forcings, answering question a) provides a basis for further analysis. By comparing the simulations with and without the ice-sheet forcings, question b) can be addressed. Answering question c) can provide insight into the Holocene temperature history and improve understanding of climate variability. To what extent are multiple simulations (performed by different models) consistent? (Chapter 3) This general question can also be divided into three levels. First, what are the consistencies and divergences among these Holocene simulations (namely LOVECLIM, CCSM3, HadCM3 and FAMOUS)? In other words, to what degree and over which reigions are these Holocene simulations consistent and divergent. Second, what climate variables in models cause these inter-model divergences? In particular, detection of the variables in simulations causing these multi-model variations from the model response perspective can facilitate to the identification of their direct reasons when divergences are found. Third, where do these discrepancies originate? This implies tracking down of the original sources
of these inconsistencies from the models themselves, such as different or even biased model principles. To what extent are model simulations and proxy data consistent? (Chapter 4) There are three sub-questions under this broad question. First, to what degree do proxybased reconstructions agree with model results? Expressly, we identify the regions where data and model results are consistent and where there is model mismatch with proxy data at the sub-continental scale of NH high latitudes. Second, what are the potential sources of uncertainty in simulated temperatures and proxy-based reconstructions? The answer to this question can provide insights into the interpretation of proxy results and validation of simulations. Third, what are the most probable Holocene temperature trends? Specifically, we conclude on the most probable temperature trends over these regions, with aid of other available evidence that is independent of the proxy data used. How much does climate dynamics?(Chapter 5)
Estimating the contributions of various potential factors to Holocene vegetation dynamics in S Fennoscandia be can divided into two aspects. Firstly, how much do different factors contribute to the Holocene vegetation dynamics and what are the main drivers of vegetation dynamics? Explicitly, we quantitatively evaluate the contribution of potential factors in sub-regions of S Fennoscandia (i.e. including South Fennoscandia, S Sweden, C Sweden and S Finland). Secondly, are there any changes in dominant patterns over different periods? Alternatively, are these patterns different before and after the onset of farming activity?
Chapter 2 Effects of melting ice sheets and orbital forcing on the early Holocene warming in the extratropical Northern Hemisphere _________________________________________________________________________ Based on: Zhang, Y., Renssen, H., Seppä H.: Effects of melting ice sheets and orbital forcing on the early Holocene warming in the extratropical Northern Hemisphere, Clim. Past, 12, 1119–1135, 2016.
Abstract The early Holocene is marked by the final transition from the last deglaciation to the relatively warm Holocene. Proxy-based temperature reconstructions suggest a Northern Hemisphere warming, but also indicate important regional differences. Model studies have analyzed the influence of diminishing ice sheets and other forcings on the climate system during the Holocene. The climate response to forcings before 9 kyr BP (referred to hereafter as kyr), however, remains not fully comprehended. We therefore studied, by employing the LOVECLIM climate model, how orbital and ice-sheet forcings contributed to climate change and to these regional differences during the earliest part of the Holocene (11.5–7 kyr). Our equilibrium experiment for 11.5 kyr suggests lower annual mean temperatures at the onset of the Holocene than in the preindustrial era with the exception of Alaska. The magnitude of this cool anomaly varied regionally, and these spatial patterns are broadly consistent with proxy-based reconstructions. Temperatures throughout the whole year in northern Canada and northwestern Europe for 11.5 kyr were 2–5 °C lower than those of the preindustrial era as the climate was strongly influenced by the cooling effect of the ice sheets, which was caused by enhanced surface albedo and ice-sheet orography. In contrast, temperatures in Alaska for all seasons for the same period were 0.5–3 °C higher than the control run, which were caused by a combination of orbital forcing and stronger southerly
CHAPTER 2 winds that advected warm air from the south in response to prevailing high air pressure over the Laurentide Ice Sheet (LIS). The transient experiments indicate a highly inhomogeneous early Holocene temperature warming over different regions. The climate in Alaska was constantly cooling over the whole Holocene, whereas there was an overall fast early Holocene warming in northern Canada by more than 1 °C kyr-1 as a consequence of progressive LIS decay. Comparisons of simulated temperatures with proxy records illustrate uncertainties related to the reconstruction of ice-sheet melting, and such a kind of comparison has the potential to constrain the uncertainties in ice-sheet reconstruction. Overall, our results demonstrate the variability of the climate during the early Holocene, both in terms of spatial patterns and temporal evolution.
2.1 Introduction The early Holocene from 11.5 to 7 kyr BP (hereafter noted as kyr) is paleoclimatologically interesting as it represents the last transition phase from full glacial to interglacial conditions. This period is characterized by a warming trend in the Northern Hemisphere (NH) that has been registered in numerous proxy records and indicated by stacked temperature reconstructions (Shakun et al. 2012). Oxygen isotope measurements from ice cores in Greenland (Dansgaard et al. 1993; Grootes et al. 1993; Rasmussen et al. 2006; Vinther et al. 2006, 2008) and the Canadian High Arctic (Koerner and Fisher 1990) consistently show an increase in δ18O by up to 3–5‰, which indicates an approximate warming in the climate system (Vinther et al. 2009). Moreover, this early Holocene warming is also registered in biological proxies. For example, a 4–5 °C warming in western and northern Europe is indicated by chironomid and macrofossil data obtained from lake sediments (Brooks and Birks 2000; Brooks et al. 2012; Birks 2015). In addition, this transition is recorded in other high-resolution records from further east in Eurasia, such as in the speleothems from China (Yuan et al. 2004; Wang et al. 2005). Comparable trends have been identified in marine sediment core data, such as sea surface temperature (SST) rise in the North Atlantic reflected by the variation in δ18O and in planktonic foraminifera (Bond et al. 1993; Kandiano et al. 2004; Hald et al. 2007). Although these proxy records provide a general view of early Holocene warming, their detailed expression in different regions and the reasons for this spatial variation are poorly known. The orbitally induced increase in NH June insolation was one of the main external drivers of climate change during the last deglaciation (Berger 1988; Denton et al. 2010; Abe-Ouchi et al. 2013; Buizert et al. 2014). This increase peaked in the earliest Holocene (Berger 1978) and resulted in warming over large areas. However, the early Holocene was also characterized by adjustments in components of the climate system that further affected the temperature through various feedback mechanisms. In the cryosphere, the Laurentide Ice
Sheet (LIS) and Fennoscandian Ice Sheet (FIS) were melting at a fast rate and eventually disappeared at around 6.8 and 10 kyr, respectively (Dyke et al. 2003; Occhietti et al. 2011), which exerted multiple influences on the climate system (Renssen et al. 2009). First, the surface albedo was much higher over the ice sheets compared to ice-free surfaces, which resulted in relatively low temperatures. Second, the ice-sheet topography could have also influenced the climate through the mechanism of adjustment to the atmospheric circulation (Felzer et al. 1996; Justino and Peltier 2005; Langen and Vinther 2009). For instance, a large-scale ice sheet could have generated a glacial anticyclone that locally could have further reduced the temperature (Felzer et al. 1996), but it may also have caused a 2–3 °C warming over the North Atlantic in the Last Glacial Maximum (LGM) (Pausata et al. 2011; Hofer et al. 2012). Third, both modeling and proxy studies have found that the Atlantic Meridional Overturning Circulation (AMOC) was relatively weak during the early Holocene due to the ice-sheet melting, which led to reduced northward heat transport and extended sea-ice cover (Renssen et al. 2010; Roche et al. 2010; Thornalley et al. 2011, 2013). Overall, the net effect of ice sheets on the early Holocene climate can be expected to have tempered the orbitally induced warming at the mid- and high latitudes. Important adjustments in the carbon cycle occurred in the early Holocene, as evidenced by the rise in atmospheric CO2 levels by 20–30 ppm that contributed to a warming (Schilt et al. 2010). Changes also happened in the biosphere during the early Holocene. Vegetation reconstructions revealed a northward expansion of boreal forest in the circum-Arctic region after the retreat of the ice sheets (MacDonald et al. 2000; Bigelow et al. 2003; CAPE project 2001; Fang et al. 2013). This expansion of boreal forest into regions that were not previously vegetated or were covered by tundra caused a reduction of the surface albedo and induced a positive feedback to the warming trend (Claussen et al. 2001). The impact of these forcings on the Holocene climate has been examined in modeling studies. The focus in these studies has been on the influence of the decay of the LIS and Greenland Ice Sheet (GIS) on the climate after 9 kyr relative to other climate forcings (Renssen et al. 2009; Blaschek and Renssen 2013). Renssen et al. (2009) used transient simulations performed with the ECBilt-CLIO-VECODE model and found that the Holocene climate was sensitive to the ice sheets and that the LIS cooling effects delayed the Holocene Thermal Maximum (HTM) by up to thousands of years. Blaschek and Renssen (2013) applied a more recent version of the same model (renamed to LOVECLIM) and revealed that the GIS melting had an identifiable impact on the climate over the Nordic Seas. However, these Holocene modeling studies only started at 9 kyr. The most important challenges in simulating climate during the initial phase of the early Holocene are the inherent uncertainties in the ice-sheet forcings in terms of the ice-sheet dynamics and the related meltwater release. Recent deglaciation studies based on cosmogenic exposure dating indicate slightly older ages of deglaciation in some regions than suggested by radiocarbon dating data (Carlson et al. 2014; Cuzzone et al., 2016), primarily because of a
CHAPTER 2 large uncertainty in bulk organic sample ages and the possibility of old carbon contamination (Carlson et al. 2014; Stokes et al. 2015). Furthermore, the Younger Dryas stadial ended at 11.7 kyr and may still have influenced the early Holocene climate due to the long response time of the deep ocean (Renssen et al. 2012). Therefore, the climate system’s response to forcings before 9 kyr, especially those of the ice sheets, is poorly understood. We have extended the study of Blaschek and Renssen (2013) back to 11.5 kyr to explore the early Holocene climate response to these key forcings. By employing the same climate model of intermediate complexity LOVECLIM, we first analyzed the impact of forcings on the climate at 11.5 kyr and subsequently investigated the influence of two ice-sheet deglaciation scenarios in transient simulations. The comparison of these different simulations enables us to disentangle how the ice sheets influenced the early Holocene climate. More specifically, we have addressed the following research questions. (1) What were the spatial patterns of simulated temperature at the onset of the Holocene (11.5 kyr)? (2) What were the roles of the forcings, especially ice-sheet decay, in shaping these features? (3) What was the spatiotemporal variability in the simulated early Holocene temperature evolution?
2.2 Model and experimental design 2.2.1 The LOVECLIM model We conducted our simulations with version 1.2 of the three dimensional Earth system model of intermediate complexity LOVECLIM (Goosse et al. 2010a), in which the components of the atmosphere, ocean including sea ice, vegetation, ice sheets and carbon cycle are dynamically included. However, in our version, the components for the ice sheets and the carbon cycle were not activated. Therefore, the ice-sheet evolution and greenhouse gases were prescribed in our present study. The atmospheric component is the quasigeostrophic model ECBilt, which consists of three vertical layers and has T21 horizontal resolution (Opsteegh et al. 1998). CLIO is the ocean component, which consists of a freesurface, primitive-equation oceanic general circulation model (GCM) coupled to a threelayer dynamic–thermodynamic sea-ice model (Fichefet and Maqueda 1997). The ocean model includes 20 vertical levels and a 3°×3° latitude–longitude horizontal resolution (Goosse and Fichefet 1999). These two core components were further coupled to the biosphere model VECODE, which simulates the dynamics of two main terrestrial plant functional types, trees and grasses, in addition to desert (Brovkin et al. 1997). More details on LOVECLIM can be found in Goosse et al. (2010a). The LOVECLIM model is a useful tool to explore the mechanisms behind climate change, and it has made critical contributions to our understanding of the climate history and
variabilities observed in proxy records (Renssen et al. 2005, 2006, 2010). For example, it has helped with investigations of the potential forcings behind abrupt climate events (Renssen et al. 2002; Wiersma and Renssen 2006; Renssen et al. 2015) as well as understanding the role of the decaying LIS and GIS in temperature evolution over the last 9 kyr (Renssen et al. 2009; Blaschek and Renssen 2013). Moreover, the LOVECLIM model simulates a reasonable modern climate (Goosse et al. 2010a). It also simulates the meridional overturning stream function reasonably well and reproduces a large-scale structure of atmosphere circulation that agrees with observations and with other models (Goosse et al. 2010a). In addition, the model’s sensitivity to freshwater perturbation is reasonable compared to that of other models (Roche et al. 2007), and its sensitivity to a doubling of atmospheric CO2 concentration is 2 K, which is at the lower end of coupled general circulation model (GCM) estimates (Flato et al., 2013). 2.2.2 Prescribed forcings We included the major climatic forcings in terms of greenhouse gases (GHG) in the atmosphere, astronomical parameters (orbital forcing, or ORB) and decaying ice sheets. In all simulations, the solar constant, aerosol levels, the continental configuration and bathymetry were kept fixed at preindustrial values. We based the concentrations of CO2, CH4 and N2O on ice core measurements for GHG forcing (Loulergue et al. 2008; Schilt et al. 2010). The radiative GHG forcing anomaly (relative to 0 kyr) in Wm-2 (Ramaswamy et al. 2001), representing the overall GHG contribution, at first showed a rapid rise with a peak of -0.3Wm-2 at 10 kyr, which was followed by a slight decrease towards a minimum at 7 kyr, and gradually increased towards 0 kyr (Fig. 2.1). The astronomical parameters (eccentricity, obliquity and longitude of perihelion) determine the incoming solar radiation at the top of atmosphere and were derived from Berger (1978). An example of the resulting change in insolation is shown as the anomaly for June at 65° N in Figure 2.1, which shows the gradual decrease over the course of the Holocene. While the global annual mean insolation stayed at almost the same level (not shown), both changes in obliquity and precession are resulting in insolation variations on the multi-millennial timescale of the Holocene. At the beginning of the Holocene (11.5 kyr), the orbitally induced insolation anomaly in the NH was positive in summer and negative in winter (Fig. S2.1). Overall, this setup of GHG and ORB forcing is in line with the PMIP3 protocol ((http://pmip3.lsce.ipsl.fr), except that our simulation excluded the increase in GHG levels during the industrial era (Ruddiman 2007). Accordingly, the terms preindustrial (era) and 0 kyr are considered equivalent in the present text and indicate modern conditions without anthropogenic impacts. We took three aspects into account concerning the ice-sheet forcing, namely their spatial extent, their thickness and their meltwater discharge. The reconstructions of ice-sheet spatial extent were based on the dating of geological features and on the correlation of
CHAPTER 2 these geological data sets between different regions (Dyke et al. 2003; Svendsen et al. 2004; Putkinen and Lunkka 2008). According to these reconstructions, the FIS at 11.5 kyr covered most of Fennoscandia except for southern Scandinavia and eastern Finland (Svendsen et al. 2004; Putkinen and Lunkka 2008; Clark, unpublished data). The LIS occupied most of the lowland area north of the Great Lakes region and filled the whole Hudson Strait (Licciardi et al. 1999; Dyke et al. 2003; Occhietti et al. 2011). The thickness of LIS was up to 2000 m, and for the FIS this thickness was only about 100m (Ganopolski et al. 2010), which is comparable with the ICE-5G reconstruction (Peltier 2004). Both the spatial extent of the ice sheets and their thickness were updated every 250 years in our transient experiments, and they decreased rapidly during the earliest Holocene, followed by a more gradual deglaciation rate from 8 kyr onward (Fig. 2.2a).
Figure 2.1 Evolution of greenhouse gas concentrations (GHG) shown as the radiative forcing’s deviation from the preindustrial level (with solid lines corresponding to the left axis), and June insolation at 65°N derived from orbital configuration (with red line and the axis on the right).
We applied the meltwater release for 1200 years in our equilibrium experiments for 11.5 kyr by adding 0.11 Sv (1 sverdrup is 106 m3 s-1) of freshwater at the St. Lawrence River, 0.05 Sv at Hudson Strait and Hudson River, 0.055 Sv coming from the FIS, and 0.002 Sv from the GIS (Licciardi et al., 1999; Jennings et al., 2015). The total freshwater volume added to the oceans in our transient experiments was about 1.46×1016 m–3 in the first 4700 years (Fig. 2.2b), which roughly matches the estimated ice-sheet melting volume during the early Holocene (Dyke et al., 2003; Ganopolski et al., 2010; Clark, unpublished data). The volume of meltwater was slightly lower than the volume of the estimated 60m sea level rise that took place during the early Holocene (Fig. 2.2c) (Lambeck et al., 2014), which suggests a coeval Antarctic melting contribution that is not considered here. Given the lack of a direct imprint left by meltwater on terrestrial records and hence the relatively large uncertainty, we used two versions of the freshwater flux (thick dashed lines and solid lines with symbols in Fig. 2.2b) that represent two possible deglaciation scenarios of the GIS and FIS, named FWF-v1 and FWF-v2. The GIS FWF_v1 scenario is derived from the ICE_5G reconstruction, and FWF_v2 is based on the reconstruction of Vinther et al.
(2009), which suggests a faster GIS thinning. The two FIS freshwater flux (FWF) scenarios are based on two estimations of the FIS melting, since the recent cosmogenic dating (FWF_v2) supports a faster melting (Clark, unpublished data) than previously thought (FWF_v1). However, we kept the freshwater discharge from LIS the same as in version 1, since the LIS deglaciation has been relatively well studied and we are more certain about its contribution. Figure 2.2 The prescribed ice-sheet forcing during the early Holocene. (a) Variation in ice-sheet extent (km2) displayed as the black lines with the axis on the left and their maximum thickness (m) indicated by the green lines with the axis on the right. A relatively minor change in GIS is not shown due to its small scale. (b) Two freshwater flux scenarios (in mSv), FWF-v1 (thick dashed lines) and FWF-v2 (solid lines). (c) Total meltwater discharge in equivalent sea level (m).
2.2.3 Setup of experiments We performed two types of experiments: equilibrium and transient simulations. First, the equilibrium experiments of OG11.5 and OGIS11.5 with boundary conditions for 11.5 kyr (Table 2.1) were designed. The OGIS11.5 experiment included icesheet forcing, whereas no ice sheets were included in OG11.5 (Table 2.2). Each of these experiments was initiated from the model’s default modern condition and was run for 1200 years, of which the last 200 years of data was used for the analysis. Renssen et al. (2006) demonstrated that a 1200-year spin-up is sufficient to reach a quasi-equilibrium in all components of the model. The data from the end of the 1200-year equilibrium run were then taken to initialize the transient experiments that covered the last 11.5 thousand years. In the first transient simulation named ORBGHG, both GHG and ORB varied on an annual basis. In the second simulation OGIS_FWF-v1, the ice-sheet topography (Fig. 2.2a) and FWF-v1 (thick dashed lines in Fig. 2.2b) were additionally included. A third experiment (named OGIS_FWF-v2)
CHAPTER 2 was performed with the freshwater version 2 (solid lines with symbols in Fig. 2.2b) and with the same ice-sheet topography as in OGIS_FWF-v1 to further investigate the climate response to the relatively uncertain freshwater forcing. Both OGIS_FWF-v1 and OGIS_FWF-v2 were initialized from the OGIS11.5 experiment. A preindustrial simulation (PI) was run for 1200 years from the model’s default (representing modern conditions) with the boundary conditions that are shown in Table 2.1 and, similarly to the other equilibrium experiments, the results of the last 200 years were used as a reference. These simulations and their forcings are summarized in Table 2.2. All temperature values in this study are shown as deviations from the PI simulation (indicates the climate at 0 kyr). Temperatures presented here are simulated near-surface temperature values without the environmental lapse rate corrections to the sea level temperature, which imply approximately a 0.5 °C cold bias over ice-sheet covered regions when compared with sitespecific proxy records. Table 2.1 Boundary conditions for 11.5 kyr and the preindustrial (PI) era PI a
11.5 ky Greenhouse gases (GHG) Orbital para. (ORB) Ice sheets (relative to PI)
CO2 253 ppm ecc. 0.019572 Size 69.2*105 km2
CH4 511 ppb obl. 24.179° Max. thickness 2331 m
N2O 245 ppb lon of perih. 270.209° FWF 220 mSv
CO2 280 ppm ecc. 0.016724
CH4 760 ppb obl. 23.446°
N2O 270 ppb lon of perih. 102.040°
PIa: GHG for 1750 AD; ORB for 1950 AD.
Table 2.2 Experiments and corresponding setup Equilibrium
FWFa: only one freshwater scenario is considered in the OGIS11.5 equilibrium experiment.
2.3 Results 2.3.1 Equilibrium experiments at the onset of the Holocene Simulation with only ORB and GHG forcings at 11.5 kyr (OG11.5) In the experiment OG11.5, summer temperatures were 2–4 °C higher over most of the extratropical continents than in the PI simulation, with a maximum deviation of 5 °C in the central parts of the Northern Hemisphere continents (Fig. 2.3a). The warming over the oceans was about 1.5 °C, and less conspicuous than that over the continents. These warmer conditions were caused by the orbitally induced positive summer insolation anomaly, as all atmospheric greenhouse gas levels were lower at 11.5 kyr than in the preindustrial era (Fig. 2.1). The most obvious feature of simulated winter temperatures was the marked contrast
between high latitudes and areas more to the south (Fig. 3b). For instance, mid-latitudes were 1.5–3 °C cooler with the strongest cooling in the central continents, whereas the highlatitude Arctic was clearly warmer with a maximum up to +3 °C than the PI. This latitudinal gradient can be seen in annual mean temperatures as well (Fig. 2.3c). Annual mean temperatures over the Arctic were about 1–4 °C higher than the PI. The warming was slightly larger in winter than in summer, and this seasonal difference mirrors the Arctic Ocean damping effect on a seasonal signal due to a large heat capacity. Temperatures at lower latitudes were annually unchanged (mostly within ±0.5 °C) with a stronger seasonality (with warmer summers and cooler winters), which is consistent with the insolation change at 11.5 kyr (Fig. S2.1).
Figure 2.3 Simulated temperatures for 11.5 kyr, shown as deviation from the PI. Left column shows the simulation with only GHG and ORB forcings (OG11.5). For the right column, the ice-sheet forcing is included (OGIS11.5). Upper, middle and lower panels present summer (JJA), winter (DJF) and annual mean temperatures, respectively.
Climate response to melting ice sheets at 11.5 kyr (OGIS11.5) Our simulation OGIS11.5 (including the impact of ice sheets) suggests a much cooler climate than that of OG11.5. Most notably, ice sheets induced a strong summer cooling over ice-covered areas, and reduced temperatures up to 5 °C compared to the PI simulation, with the strongest cooling at the center of the LIS (Fig. 2.3d). Additionally, SSTs were also more than 1.5 °C lower over the North Atlantic Ocean. In contrast, temperatures over the ice-free continents were mostly above the preindustrial level, but still lower than those found in OG11.5, except for the Alaska region. Although the area with colder conditions clearly expanded more in the OGIS11.5 simulation than in OG11.5, the central Arctic was still warmer in OGIS11.5 relative to PI (Fig. 2.3e). Alaska was the only continental region
CHAPTER 2 where winter temperatures exceeded the preindustrial values by up to +3 °C. The strongest cooling effect was present in the regions covered by ice sheets, for instance more than 3 °C cooler over the LIS. Simulated annual mean temperatures in OGIS11.5 clearly showed overall lower values than the PI due to the ice-sheet impacts (Fig. 2.3f). The Eurasian continent was mostly 1.5–3 °C cooler, and a maximum temperature reduction of more than 5 °C was found over the LIS. Only two areas were still warmer: Alaska, including the adjacent sector of the Arctic Ocean, and the Nordic Seas. The most distinct feature was thus a thermally contrasting pattern over North America, with simulated temperatures being around 2 °C higher than those the PI for Alaska, whereas over most of Canada temperatures were more than 3 °C lower. 2.3.2 Transient simulation for the Holocene It is clear in our analysis of Sect. 2.3.1 that the climate showed different responses in the following areas: the Arctic, northwestern Europe, northern Canada, Alaska and Siberia (marked in Fig. S2.2). Therefore, these areas were selected for special examination and the temperature evolutions of these regions will be shown. Our major focus was on millennial scale temperature trends; therefore, we applied a 500-year running mean to our simulated time series that effectively filtered out high-frequency variability. Temperature evolution in the Arctic Arctic summer temperatures in ORBGHG continuously decreased, which resulted in a total cooling of 2 °C during the Holocene (Fig. 2.4). Winter temperatures showed an even stronger overall cooling of -3 °C. Figure 2.4 Simulated temperature evolution, shown as the anomalies compared to PI, since the early Holocene at high latitudes (north of 70°N). Panels (a), (b) and (c) represent the summer, winter and annual values, respectively. The slope indicates the overall warming rate and is based on the least-squares regression over the period from the 11.5 to 6 kyr, as from 6 kyr the temperature starts to decrease. It is only a general estimation, and thus uncertainty ranges are not provided. The warmest peak is marked by a shaded bar and represents the simulated peak during which the temperature was over 1 °C higher than the PI. Both slope calculation and warm peak are based on the OGIS_FWF-v2 simulation.
The simulation OGIS_FWF-v1 with full forcings reveals a more complicated Arctic climate evolution compared to that of ORBGHG. The effect of ice sheets at the onset of the
Holocene caused temperatures in both summer and winter to be more than 2 °C lower than those indicated by ORBGHG. The final deglaciation of the FIS happened at 10 kyr, and the corresponding deglaciation for LIS occurred at 6.8 kyr. Therefore, their cooling effects no longer existed after 6.5 kyr, and all three runs showed similar temperatures after that time. As a consequence, the temperature evolution curve of OGIS_FWF-v1 first showed a warming, with the peak being reached at around 7 kyr, when the cooling effects of the ice sheets had been counterbalanced by the insolation anomalies. This was subsequently followed by a gradual cooling that was controlled by a decrease in the orbital forcing. Simulated temperatures initially had increased by 6.5 kyr at rate of 0.26, 0.21 and 0.44 °C kyr-1 for summer, winter and annual temperatures, respectively. The larger warming rate in annual mean than in summer and in winter was due to a largest response in the winter half year. The Arctic (here we refer to the region located north of 70°N) has a large part of ocean where the maximum response was delayed by a few months (Renssen et al., 2005). The study found the largest response in the winter half year (especially in fall) due to above delayed response that was ultimately caused by the thermal inertia over the oceans (Renssen et al., 2005). Indeed, this explanation was furthermore supported by the simulated large warming rate in fall (up to 0.78 °C kyr-1). The OGIS_FWF-v1 simulation indicates that Arctic summer climate experienced a slightly faster warming at the beginning, followed by a more gradual warming toward the maximum anomaly of 1 °C warmer than PI at about 7.5 kyr (Fig. 2.4a). Simulated winter temperatures stayed at a level of 2 °C lower than that of ORBGHG before 7 kyr, which was followed by a rapid increase of about 1.5 °C within a 500-year period, and then reached a temperature peak of about 1.5 °C warmer than the PI (Fig. 2.4b). Simulated annual mean temperatures showed a relatively stable rise until 6.5 kyr, which reached a maximum of about 1.5 °C warmer than that of PI (Fig. 2.4c). The simulation OGIS_FWF-v2 gives similar results for the Arctic but had an even cooler climate before 9 kyr than in OGIS_FWF-v1, with the maximum cooling of up to 0.3 °C for all seasons at 10.5 kyr. Temperature evolution in northwestern Europe The ORBGHG simulation indicates smaller climate variability in northwestern Europe than in the Arctic. Temperatures declined by around 1.5 °C through the entire period in summer and less than 0.5 °C for annual mean, and rose by 0.5 °C in winter (Fig. 2.5), which implies a decreasing seasonality toward the preindustrial era. This contrasts markedly with the clear cooling of climate in each season in the Arctic. The OGIS_FWF-v1 simulation shows an overall cooler climate in northwestern Europe at the onset of Holocene, with temperature anomalies of -1.5 °C in summer, -3°C in winter and -2.8 °C in annual mean compared to the PI simulation. Temperatures increased from this point (11.5 kyr) toward 6 kyr at an overall rate of 0.28, 0.48, and 0.54 °C kyr-1 for summer, winter and annual mean, respectively. The most important feature in summer was
CHAPTER 2 a sharp temperature rise from a negative anomaly (-1.5 °C) to a positive one (±1 °C) by 10 kyr, when the first peak was reached. Subsequently, a slight cooling was noted before 8 kyr, followed by another temperature increase, which led to a second warming peak at 7.4 kyr. The climate in winter showed a relatively stable warming by 6.5 kyr with no identifiable warm peak. Annual temperatures reflected the same phases of warming as in summer, one before 10 kyr and another before 7.5 kyr, but without a clear early temperature peak. Temperatures in all seasons from around 7 kyr followed the ORBGHG simulation. It is worth noting that the OGIS_FWF-v2 simulation indicates a further cooling in summer between 11.5 and 9 kyr compared to OGIS_FWF-v1, which was also reflected in annual mean temperatures. As a result, there was only one clear thermal maximum in summer for northwestern Europe, which peaked at around 7.4 kyr.
Figure 2.5 Same as Fig. 2.4 but for northwestern Europe (5°W–34°E, 58–69°N).
Figure 2.6 Same as Fig. 2.4 but for northern Canada (120–55°W, 50– 69°N).
Temperature evolution in northern Canada Simulated temperatures in ORBGHG for northern Canada decreased by 2.2 °C in summer, by 0.6 °C in winter and by 1.1 °C for annual average during the Holocene (Fig. 2.6). The stronger cooling in summer than in winter reflected a strong early Holocene seasonality, which decreased over the whole period. The OGIS_FWF-v1 simulation describes a much cooler climate in northern Canada during
the early Holocene than that indicated by ORBGHG. This cooling was up to 5 °C for all seasons at the onset of Holocene. The climate dramatically warmed with an overall high rate of more than 1 °C kyr-1 in both winter and summer during the early Holocene, which was due to the impact of the decaying LIS. The early Holocene warming was, however, not linear because an initial phase with more rapid warming was followed by a more gradual temperature increase. In summer, this warming resulted in a thermal peak at around 7.4 kyr, which was about 1.5 °C warmer than the PI. From 7.4 kyr onwards, the climate experienced a gradual cooling that was very similar to that of ORBGHG. Simulated temperatures in winter and annual mean did not show such a clear warm peak in comparison to summer. The results of OGIS_FWF-v2 only indicated marginal differences relative to OGIS_FWF-v1 for all seasons. Overall, the most significant feature of simulated temperatures in northern Canada was the strong warming that took place in the early Holocene. Temperature evolution in Alaska The ORBGHG simulation shows an overall cooling in Alaska for all seasons. Simulated summer and annual mean temperatures experienced a decrease of more than 2 °C throughout the whole period. Winter temperatures had slightly increased by 10 kyr, and then stayed about 2 °C higher for a period of 800 years, which was followed by a constant decrease toward the preindustrial value.
Figure 2.7 Same as Fig. 2.4 but for Alaska (170–120°W, 58–74°N).
Figure 2.8 Same as Fig. 2.4 but for Siberia (62–145°E, 58–74°N).
CHAPTER 2 In contrast to other areas, both summer and winter temperatures in OGIS_FWF-v1 showed an overall cooling trend in Alaska during the entire Holocene (Fig. 2.7), which was slightly higher than in our ORBGHG simulation. The OGIS_FWF-v1 simulation indicates a 2 °C decline in summer temperature over the whole period, with a slightly faster rate between 7 and 6.5 kyr. Simulated winter temperatures decreased by 3.5 °C during the early Holocene, with two small declines at 9.5 and 6.7 kyr. Annual temperatures in the OGIS_FWF-v1 simulation reflected a 2.3 °C cooling during the Holocene. The OGIS_FWF-v2 simulation represents an Alaskan temperature trend that is rather similar to that of OGIS_FWF-v1. Temperature evolution in Siberia The ORBGHG simulation describes an almost 2 °C decline of summer temperatures over Siberia during the last 11.5 thousand years (Fig. 2.8). Simulated winter temperatures showed a smaller variation, as it decreased by less than 1 °C, and annual mean temperatures decreased by around 1 °C over the course of the Holocene. The evolution of simulated temperatures in ORBGHG over Siberia was on a similar scale to that of northwestern Europe. The difference of simulated Siberian temperatures between ORBGHG and OGIS_FWF-v1 varied in summer and winter. On the one hand, simulated summer temperatures in OGIS_FWF-v1 were generally similar to that in ORBGHG with the exception of a small warming of 0.7 °C before 10 kyr. On the other hand, winter temperatures in the OGIS_FWF-v1 simulation were around 2 °C lower than in ORBGHG before 7 kyr, followed by a rapid increase over the next 500 years, after which it followed the ORBGHG simulation. Consequently, simulated early Holocene warming lasted much longer in winter than in summer. Simulated Siberian temperature evolution in OGIS_FWF-v2 generally followed that of OGIS_FWF-v1.
2.4 Discussion We will evaluate our results by briefly comparing the simulations with proxy-based reconstructions, which will be followed by an analysis of the mechanism behind the simulated temperature patterns. The impact of freshwater forcing will also be discussed based on the two FWF scenarios. 2.4.1 Comparison of simulations with proxy records At the onset of the Holocene, the overall cool climate indicated by the reconstructions generally matches that of our OGIS11.5 simulation, which shows lower annual temperatures at 11.5 kyr than the PI. Climate reconstructions based on proxy data generally show a cooler early Holocene over northern Europe than at 0 kyr both in the summer and winter (Heiri et al. 2104; Mauri et al. 2015). Terrestrial and ocean sediment data also
suggest a cooler early Holocene climate over eastern Siberia (Klemm et al. 2013; Tarasov et al. 2013) and slightly lower SSTs over the North Atlantic Ocean (Came et al. 2007; Berner et al. 2008). Cooler conditions over the Barents Sea and Greenland are also indicated by multiple proxies (Peros et al. 2010; de Vernal et al. 2013; Vinther et al. 2008). Therefore, these proxy data agree with simulated lower temperatures over these areas. However, there is less agreement with proxies in places where the reconstructions are sparse. The only available pollen-based reconstruction from the western side of the Ural Mountains suggests similar early Holocene summer temperatures (within 1 °C anomaly) compared to the preindustrial era (Salonen et al. 2011), whereas OGIS11.5 indicates that summer temperatures were slightly higher at 11.5 kyr over most areas. At high latitudes, the sea-ice cover reconstructions serve as an indirect paleo-temperature proxy due to the scarcity of temperature records, and reveal an inconclusive temperature signal over the Canadian Arctic (de Vernal et al. 2013), whereas our simulation reflects an overall warmer climate in the west and cooler conditions in the east. Proxies indicate significantly different climate patterns over the east and the west of northern America. The later initiation and termination of HTM over northern Canada imply lower temperatures during the early Holocene in the east (Kaufman et al. 2004). However, the higher-than present early Holocene temperatures over central Beringia and Alaska are reflected by peat accumulation and by northward expansion of animal species (Kaufman et al. 2004; Jones and Yu 2010). This thermal contrast agrees with those simulated patterns in the OGIS11.5 simulation, which indicates warmer temperatures for Alaska and a much cooler climate over Canada. However, this interpretation of high temperature was recently challenged by Kaufman et al. (2016), who argued that the highest summer temperature in Alaska occurred as late as 8–6 kyr. Hence our simulation agrees better with the interpretation of Kaufman et al. (2004). In general, our simulation with full forcings was able to capture main temperature features indicated in proxy-based reconstructions. Shakun et al. (2012) and Marcott et al. (2013) stacked multiple proxies to construct a record of temperatures since the LGM. Both above stacked reconstructions and our simulation OGIS_FWF-v2 show that the Holocene was generally characterized by an initial warming and subsequent Holocene warm period over the NH extratropics, which indicates the broad consistency between simulation and proxy data. However, there are some disagreements related to seasonality (Fig. 2.9). Marcott et al. (2013) interpreted the stacked temperature reconstruction as representative of the annual mean climate, whereas it shows a better agreement with our simulated summer temperature than with annual mean value (Fig. 2.9). One potential explanation for this seasonal mismatch is that some proxy records have seasonal bias toward summer conditions, as has been suggested recently for many marine-based SST reconstructions from high latitudes (Lohmann et al. 2013).
CHAPTER 2 Further region-by-region comparisons of these warming rates with proxy records are beyond the focus of this work and will be dealt with in a future publication.
Figure 2.9 Model–data comparison over the latitudinal band of 30–90°N, shown as a deviation from the PI. The stacked temperature reconstruction with 1σ uncertainty (grey band) is based on Marcott et al. (2013).
2.4.2 Mechanism of climate response to forcings It is clear from our data that the spatial patterns of climate response at the onset of the Holocene can be attributed to the variation in the dominant forcings prevailing in the different areas. Orbital-scale insolation variations are important driving factors for the early Holocene climate. For instance, higher temperatures in Alaska could be attributed to the orbitally induced positive insolation anomaly in combination with an anomalous atmospheric circulation caused by the remnant LIS. The air descended over the cold LIS surface, which created a high surface pressure anomaly that produced a clockwise flow anomaly at the surface, as indicated by the 800 hPa geopotential height (Fig. 2.10). This induces stronger southerly winds over Alaska, which advected relatively warm air from the south. A potentially different early Holocene atmospheric circulation near the North Atlantic was also found in a proxy record of Steffensen et al. (2008), who reported an abrupt transition of deuterium excess that indicates a temperature change of precipitation moisture sources and is thus indirectly connected to atmospheric circulation changes. The strong influence of the ice sheets on early Holocene temperatures has been found in previous studies (Renssen et al. 2009, 2012). Simulated lower summer temperatures over northern Canada and northwestern Europe in our GIS11.5 simulation were the result of such ice-sheet-induced cooling, which would have fully overwhelmed the warming effect of the positive summer insolation anomaly. The ice-sheet cooling effect could partly be explained on a local scale by the enhanced albedo over the ice sheets and by the climate’s high sensitivity to albedo change (Romanova et al., 2006). Indeed, the summer surface
albedo over the ice sheets was much higher (up to 0.8) than over ice-free surfaces, where the values varied from only 0.1 to 0.5, depending on the vegetation type and the fractional snow cover (Fig. 2.11). Temperatures could be further reduced by the ice-sheet orography impact. The elevation of ice sheets introduced descending air over the ice-sheet surface, which caused locally cooler conditions. There was also an approximate 0.5 °C cold bias induced by the lapse rate effect when compared with the site based records.
Figure 2.10 Geopotential height anomalies from global mean (in m2 s-2) at 800 hPa in the extratropical Northern Hemisphere. Panels (a), (b) and (c) show the control condition PI and the simulations OG11.5 and OGIS11.5, respectively.
Figure 2.11 Summer surface albedo in the extratropical Northern Hemisphere. Panels (a), (b) and (c) represent the control run (PI) and the simulations without ice sheets (OG11.5) and with ice sheets (OGIS11.5), respectively.
Changes in vegetation and land cover during the early Holocene contributed to climate change as well, especially over ecotonal regions. Modeling studies suggest that deforestation in boreal regions could decrease regional temperatures by up to 1 °C due to an increase in surface albedo and related positive feedbacks (Levis et al. 1999; Claussen et al. 2001; Liu et al. 2006). Taking Siberia as an example, the insolation-induced warming was partially offset by the overall higher summer albedo (Fig. 2.11) induced by the southward expansion of the tundra and/or bare ground and related feedbacks at 11.5 kyr, resulting in a minor warming in summer. The albedo-related feedbacks and the smaller annual insolation anomalies jointly result in a 0.5–2 °C cooler in annual climate at 11.5 kyr. We are aware of the potential role of permafrost at high latitudes; however, the discussion
CHAPTER 2 of the impact of permafrost thaw is hindered by the fact that our model version did not include a dynamic permafrost module. A version of LOVECLIM that is coupled to a permafrost module (VAMPERS) is currently in development (Kitover et al. 2015), and should enable us to quantify the role of permafrost in a future study.
Figure 2.12 Meridional overturning stream function (Sv) in the Atlantic Ocean basin. Panels (a), (b) and (c) indicate the control run (PI) and the simulations OG11.5 and OGIS11.5, respectively. On the left-hand side, depth is indicated in kilometers. Positive values indicate a clockwise circulation. Maximum AMOC strength value was 22 Sv (reached at about 1200m depth) in the PI and OGIS simulation, while it was only about 14 Sv (reached at 600–700m depth) in OGIS11.5.
Figure 2.13 Minimum sea-ice thickness (m) in September for PI (a), OG11.5 (b) and OGIS11.5 (c).
Meltwater release and sea-ice-related changes also had a footprint in the early Holocene climate. The OGIS11.5 simulation produces a sluggish AMOC in the North Atlantic with the largest decrease being more than 3 Sv. It was also reflected in a shallower overturning circulation at 11.5 kyr compared to the PI simulation as a response to meltwater release (Fig. 2.12). This slowdown also coincides with the foraminifera data from the Arctic Ocean and the Fram Strait that suggest a reduced northward oceanic heat transport (Thornalley et al. 2009). The slowdown and reduced heat transport led to slightly lower temperatures at high latitudes (western Arctic Ocean) at 11.5 kyr than that at 0 kyr. Likewise, after the meltwater fluxes of the LIS diminished around 7 kyr, strong intensification of the AMOC followed. This sudden intensification of AMOC would explain the rapid Arctic temperature increase that occurred at this time (Fig. 2.4). However, it is important to note that the
temperature decrease was not simply inversely linear with the amount of northward transport of heat since the sea-ice feedbacks further reinforce this change (Roche et al. 2007). In fact, sea-ice coverage in the OGIS11.5 simulation was much more extensive over the Davis Strait (northern Labrador Sea) than the corresponding value in OG11.5 (Fig. 2.13). This extended seaice cover in this region was stronger than the direct cooling effect of the reduced oceanic heat transport. Such an anomaly might be explained by positive feedbacks involving sea ice being active (Renssen et al., 2005). The Greenland Sea warming could be attributed to enhanced convective activity that releases more oceanic heat into the atmosphere. This enhanced convective activity was caused by the shift of deep water formation from the eastern Greenland Sea to the west, which was initially induced by the freshwater discharge from ice-sheet melting. The net response of the climate reflects the impact of a combination of forcings and feedbacks, which showed a high temporal-spatial variability. 2.4.3 Early Holocene warming and climate–ocean system response to freshwater The simulation of OGIS_FWF-v2 indicates a stronger cooling (before 9 kyr) in the Arctic and northwestern Europe than those found in the OGIS_FWF-v1 with the strongest temperatures reduction at around 10 kyr. The enhanced freshwater influx from the GIS and the redistributed meltwater from the FIS caused an alteration in the surface ocean freshening in the Nordic Seas, which reduced convective activity (Renssen et al. 2010; Blaschek and Renssen 2013). Indeed, this reduction led to a further slight reduction of the northward heat transport by the Atlantic Ocean, which was associated with a further AMOC weakening (by 1–2 Sv than in FWF_v1): this in turn produced a slightly stronger cooling at 10 kyr (Fig. S2.3) and a sea-ice expansion over the Denmark Strait (Figs. 2.14 & S2.4). However, the efficiency of the above meltwater flux freshening effect is determined by multiple aspects. The most important factor is the maximum flux of meltwater that was added to the ocean, while the total freshwater amount had only a second-order effect (Roche et al. 2007). Numerous investigations on the behavior of the coupled atmosphere– ocean system suggest that the application of freshwater will not lead to a disruption of the North Atlantic Deep Water production (NADW) as long as a certain threshold is not crossed (Ganopolski et al. 1998; Rahmstorf et al. 2005). Apart from the intensity and duration, the ocean circulation response to freshwater also depends on the location where this freshwater is released. For instance, it is more sensitive to the release of freshwater in the eastern Norwegian Sea than at the St. Lawrence River outlet since the former is closer to the main site with NADW formation (Roche et al. 2010). This is consistent with a previous study by Blaschek and Renssen (2103), who found that freshwater from the GIS did have a tangible impact on the Nordic Seas, even though the total amount was minor. Since the second freshwater scenario (OGIS_FWF-v2) includes a slightly larger FWF from
CHAPTER 2 the GIS (compared to that in OGIS_FWF-v1) and the FWF was released in a sensitive area, the location-dependent sensitivity could also partially explain further AMOC weakening in the OGIS_FWF-v2 simulation compared to OGIS_FWF-v1. The OGIS_FWF-v1 simulation indicates two peaks in the temperature evolution over northwestern Europe, at around 10 and 7 kyr. High temperatures at 7 kyr are recorded in proxy-based reconstructions as well. However, no warm peak at 10 kyr was observed in pollen-based reconstructions, which actually suggests a cooler climate prevailed at 10 kyr than in the preindustrial Europe (Mauri et al. 2015). In contrast to the climate simulated in OGIS_FWF-v1, the simulation with updated freshwater (OGIS_FWF-v2) produced a warming trend that is consistent with a highest temperature around 7 kyr. Moreover, the OGIS_FWF-v1 produced a temperature decrease between two peaks, whereas the proxies indicated a rapid temperature increase at the beginning followed by a more gradual warming (Brooks et al. 2012). Therefore, from the viewpoint of temperature evolution in northwestern Europe, the OGIS_FWF-v2 represented a more realistic climate than OGIS_FWF-v1 did, which implies that the existing uncertainties in the reconstructions of ice-sheet dynamics can be evaluated by applying different freshwater scenarios. Further comparison with proxy data and with other model transient simulations will be conducted in a future paper.
2.5 Conclusions We performed both equilibrium and transient simulations by employing the LOVECLIM climate model to explore the spatial patterns of the climate response to forcings at the onset of the Holocene and temperature evolution over the last 11.5 thousand years. We focused on three research questions in our analysis, which are outlined below with the main finding: 1) What were the spatial patterns of simulated temperature at the onset of the Holocene? The temperature anomalies relative to PI at 11.5 kyr were regionally heterogeneous, which are shown as a range of annually negative anomalies over many areas but which were positive in Alaska. The climate in eastern northern Canada and northwestern Europe was much cooler than in other regions, with temperature anomalies of -2 to -5°C relative to 0 kyr throughout the year. The climate over the northern Labrador Sea and the North Atlantic was also 0.5–3 °C cooler. Temperatures in Siberia were 0.5–3 °C and 1.5–3 °C lower in winter and annually, respectively, and summer temperatures showed only a small deviation (between +0.5 and -1.5 °C) compared to 0 kyr. Simulated summer temperature anomaly in the eastern Arctic Ocean was also small (between +0.5 and -0.5 °C), and annual temperatures were 0.5–2 °C lower. In contrast to cooler conditions in other areas, temperatures in Alaska were 1.5–3 °C higher than the preindustrial period for all seasons.
2) What were the roles of forcings, especially ice-sheet decay, in shaping these features? The ice-sheet cooling effect in northern Canada and northwestern Europe overwhelmed the warming impact of the positive insolation anomaly, which caused the relatively cold climate at 11.5 kyr. In particular, the enhanced surface albedo over the ice sheets and the orographic effect were important in promoting these cold conditions. The cooler climate over the northern Labrador Sea and the North Atlantic was related to both reduced northward heat transport and enhanced sea-ice feedbacks. A small summer temperature anomaly was found in Siberia, where the positive insolation anomaly was partially offset by the cooling effect of the higher albedo associated with the relatively extensive tundra cover in the early Holocene. Overall, lower winter and annual temperatures at 11.5 kyr over central Siberia can be attributed to both vegetation-related albedo feedbacks and to the relatively small negative insolation deviation compared to the preindustrial level. The dominant factors driving the climate in eastern Arctic Ocean climate were the amount of northward heat transport associated with the strength of ocean circulation and the orbitally forced insolation variation. Annual mean temperatures at 11.5 kyr were lower than at 0 kyr because the cooling effect of a reduced northward oceanic heat transport (induced by weakened ocean circulation) was larger than the insolation-induced warming. During summer, these two factors were of similar magnitude and temperatures were similar to those of the preindustrial era. Temperatures in Alaska were higher for all seasons in response to the dominant positive insolation anomaly and the enhanced southerly winds induced by the LIS, which advected relatively warm air from the south. Therefore, this regional heterogeneity is the result of the climate response to a range of dominant forcings and feedbacks. 3) What was the spatiotemporal variability in the simulated early Holocene evolution? Above, geographical variability is also reflected in the Holocene temperature evolution, especially in the early Holocene warming. In Alaska, the climate was constantly cooling throughout the Holocene due to the decreasing insolation and atmospheric circulation variability. In contrast, northern Canada experienced a strong warming with an overall warming rate over 1 °C kyr-1, and this warming lasted until 7 kyr. Although different forcings and mechanisms played different roles in northwestern Europe, the Arctic and Siberia, the overall warming effect was similar for these regions, with a rate of around 0.5 °C kyr-1. In addition, the comparison of early Holocene temperatures over northwestern Europe with proxy records suggests that the OGIS_FWF-v2 represented a more realistic climate condition than the OGIS_FWF-v1 does, and it implies that the uncertainties with regard to the ice-sheet decay can potentially be constrained by applying different deglaciation scenarios and comparing then with networks of proxy records. Overall, our results demonstrated a large spatial variability in the climate response to diverse forcings and feedbacks, both for the early Holocene temperature distribution and for the early
CHAPTER 2 Holocene warming, and this data–model comparison also helps in understanding the difference between proxy records.
Chapter 3 Holocene temperature trends in the extratropical Northern Hemisphere based on inter-model comparisons ________________________________________________________________________ Based on: Zhang, Y., Renssen, H., Seppä, H., Valdes, PJ. Holocene temperature trends in the extratropical Northern Hemisphere based on inter-model comparisons. Journal of Quaternary Science, minor revisions.
Abstract Large uncertainties exist in Holocene climate estimates, especially during the early Holocene when large-scale reorganization was occurring in the climate system. To analyse these uncertainties, we compare four Holocene simulations performed with the LOVECLIM, CCSM3, HadCM3, and FAMOUS climate models. The simulations are generally consistent on large-scale Northern Hemisphere extratropics, while the multisimulation consistencies are heterogeneous on sub-continental scale. Consistently simulated temperature trends are found in Greenland, N Canada, NE and NW Europe and central-west Siberia. These Holocene temperatures show a pattern of an early-Holocene warming, mid-Holocene warmth and gradual decrease toward the pre-industrial, and the extent of early-Holocene warming varies spatially, with 9 °C winter warming in N Canada compared with 3 °C warming in central-west Siberia. In contrast, mismatched temperatures are detected: in Alaska, the warm early-Holocene winter in LOVECLIM primarily results from strongly enhanced southerly winds induced by the ice sheets; in E Siberia, the intense summer warmth in CCSM3 is caused by large negative albedo anomalies due to overestimated snow cover at 0 ka; in the Arctic, cool winter temperatures in FAMOUS can be attributed to too extensive sea ice coverage probably due to simplified sea-ice representations. Thus the Holocene temperature trends in these regions remain inconclusive.
3.1 Introduction The Holocene is an important period for investigating climate variability and improving our understanding of the climate system due to the detectable changes in climate variables and the abundance of available proxy records. The early Holocene (11.5–7 ka) was a transitional phase, accompanied by reorganizations in several components of the climate system, as recorded by proxy data (e.g. CAPE project 2001; Dyke et al. 2003; Shakun et al. 2012). Investigating this transient period can increase the knowledge of variabilities of the climate system, in addition to provide further information on the climate history. Climate modelling, especially by transient simulations is a useful tool to investigate the climate change, and thus have a potential to improve our understanding of this indeterminate earlyHolocene climate. Recent model simulations have either included or specifically focused on this critical time phase to investigate the climate responses to dynamic climate forcings (He 2011; Zhang et al. 2016). However, uncertainties related to the melting of the ice sheets adds challenge in accurately simulating this period. The ice-sheet related uncertainty during the early Holocene has been tested by Zhang et al. (2016), who investigated two different freshwater scenarios in model simulations to identify a more reasonable scenarios regarding the degrees of the early-Holocene cooling and the timing of HTM. Uncertainties in the early-Holocene simulations are not only caused by ambiguity in climate forcings, but can also result from model-dependent variations. The performance of models and their sensitivities to a given forcings are primarily determined by the physical representations of various climate processes in the models. For instance, climate sensitivity to radiative forcings (CSr) usually refers to the change in the global annual mean surface air temperature (in °C) experienced by the climate system after it has attained a new equilibrium in response to a doubling of the atmospheric CO2 concentration (Knutti and Hegerl 2008). Estimates of CSr vary from 2.1 to 4.4 °C among the models in the Coupled Model Inter-comparison Project (CMIP3), due to various feedbacks involved in the models (Randall et al. 2007; Flato et al. 2013). In addition, it has found that CSr is greater for cold than for warm climates, and the sensibilities to changes of climate condition differ among individual studies owing to non-uniform representations of cryospheric processes or dynamical ocean feedback processes (Boer and Yu 2003; Randall et al. 2007; Kutzbach et al. 2013). Variations among models in the estimates of key climate process or parameters, such as the climate-state-dependent CSr, add extra difficulties to simulate the earlyHolocene climate. Multi-model comparison provides an option to validate the reliability of model performances and hence to increase the confidence in simulations. Reliability of and confidence in simulations increases if similar features are observed in other independent model results; conversely, further investigations on the involved variables are required when multi-model simulations differ. The Paleoclimate Modeling Inter-Comparison
Project (PMIP) has performed a wide range of inter-model comparisons, covering a series of periods, such as the Last Glacial Maximum (LGM, 21 ka), the Last Interglacial (LIG, 130–115 ka) and the mid-Holocene (6 ka) (e.g. Harrison et al. 1998; Braconnot et al. 2000, 2012; Lunt et al. 2013), which has increased our understanding of climate changes. However, the PMIP comparisons for the Holocene primarily focus on the mid-Holocene with snapshot experiments, and the current transient inter-model comparisons have been conducted for periods shorter than the whole Holocene, such as 8–2 ka and the last millennium (e.g. Bakker et al. 2014). Therefore, the question of how similar or different the model simulations are during the early Holocene is still unknown, and inter-model comparisons spanning to the entire Holocene are greatly demanded. Here, we evaluate the robustness of four different Holocene simulations in detail by conducting inter-model comparisons and analyzing their uncertainties. These models include LOVECLIM, CCSM3, HadCM3, and FAMOUS, and they differ in multiple critical aspects, as summarized in Table 3.1. Our goal is to analyze how the climate responds to dominant forcings in these different models and to evaluate how robust these responses are through comparisons of multiple Holocene transient simulations. In particular, our comparisons aim to: 1) identify the agreements and divergences among these Holocene simulations, 2) detect which variables in simulations cause these multimodel variations, and 3) further examine the potential origin of these inconsistencies from the models themselves, such as different parameterizations and biases in the models. It is worth to mention that a detailed data–model comparison is, however, beyond the scope of the present study, and will be the topic of a future publication.
3.2 Models and simulations 3.2.1 Models and prescribed climate forcings Models The climate models employed in this study include the LOVECLM, FAMOUS, CCSM3 and HadCM3, which have various resolutions and complexities. The LOVECLIM model explicitly represents the components of the atmosphere, ocean and sea ice, vegetation, ice sheets and carbon cycle (Goosse et al. 2010a) with various complexities. However, the components for the ice sheets and the carbon cycle were not activated in our version, and thus prescribed ice-sheet configurations and greenhouse-gases concentrations (based on reconstructions) are required. The atmospheric component is the quasi-geostrophic model ECBilt coupled to a land-surface model and consists of 3 vertical layers and has T21 horizontal resolution (Opsteegh et al. 1998). The ocean model, CLIO, is an oceanic GCM coupled to a three-layer dynamic-thermodynamic sea-ice model (Fichefet and Maqueda 1997) and includes 20 vertical levels and a 3 × 3° (lat × lon) horizontal resolution (Goosse and Fichefet 1999). These two core components are further coupled to the biosphere model
CHAPTER 3 VECODE, which simulates the dynamics of two main terrestrial plant functional types, trees and grasses, as well as desert (Brovkin et al. 1997). More details on LOVECLIM can be found in Goosse et al. (2010a). LOVECLIM reproduces reasonably well the main characteristics of the observed surface temperature distribution (Goosse et al. 2010a). The large-scale structure of the near surface circulation is well reproduced by the model with a general decrease of the geopotential height with latitudes and local minima in the North Atlantic, although there are some discrepancies in specific location of these weather systems (Goosse et al. 2010a). The model is also able to simulate reasonably well the sea ice extent, even though the amplitude of the seasonal cycle of the sea-ice concentration is slightly weak in some regions. The maximum AMOC reaches 22 Sv (Sverdrup = 1×106 ms-1), with deep convection occurring both in the Greenland-Norwegian Sea and the Labrador Sea. These values are in relatively good agreement with the data-based estimates and in the high end of the values given by other models (Ganachaud and Wunsch 2000; Gregory et al. 2005; Goosse et al. 2010a). LOVECLIM has made critical contributions to our understanding of climate changes in the past, such as during the Holocene (Renssen et al. 2009; Blaschek and Renssen 2013). The HadCM3 model belongs to the GCM family of model and consists of coupled models for the atmosphere, ocean and sea ice (Gordon et al. 2000; Pope et al. 2000). The horizontal resolution of the atmospheric model is 2.5 × 3.75° (lat × lon), with 19 unequally spaced levels in the vertical. The land surface scheme includes a representation of the freezing and melting of soil moisture, and terrestrial evaporation (Cox et al. 1999). Interactive vegetation is also incorporated in the version we used. The spatial resolution of the ocean model in HadCM3 is 1.25 × 1.25° (lat × lon) by 20 unequally spaced layers (Gent and McWilliams 1990). The sea ice model uses a simple thermodynamic scheme and contains parameterization of ice drift and leads (Cattle and Crossley 1995). The temporal variability on annual to decadal timescales is in good agreement with observations, and the thermal contrast pattern between the ocean and continent is qualitatively similar to that observed as well, although there is an overestimation of temperature variability in some regions (e.g. in western Europe and North America due to dry soil in the simulation), and regional biases in ocean temperature variability (Collins et al. 2001). The model also represents the realistic atmospheric circulation, with well-agreed NAO index in comparison with the observations (Collins et al. 2001). Sea ice extents in the Arctic are overestimated when compared to the observed data (Gordon et al. 2000). For instance, the Barents Sea remains ice covered throughout the year and in the N Atlantic region sea ice extends too far to the south during the winter. FAMOUS is a fast and low-resolution version of HadCM3, and its parameterizations of physical and dynamical processes are almost identical to those of HadCM3 (Smith et al. 2008). The ocean component is based on the Cox-Bryan model (Pacanowski et al. 1990), and is run at a resolution of 2.5 × 3.75° (lat × lon), with 20 vertical levels. The atmosphere
is based on primitive equations, with a resolution of 5 × 7.5° (lat × lon), with 11 vertical levels (Smith and Gregory 2009). It has approximately half the spatial resolution of HadCM3, and runs around nine times quicker. FAMOUS can capture the general atmospheric features suggested by HadCM3 (e.g. temperature and atmosphere circulation patterns), except for a warm bias in the tropics and a cold bias in the northern high latitudes due to slightly overestimated sea ice (Jones et al. 2005). A warm bias of FAMOUS in the tropics (especially over land) and a cold bias in the northern high latitudes are represented compared with observations and HadCM3 (Jones et al. 2005). This cold bias in the north Atlantic is associated with the presence of too extensive sea-ice. The NH sea-ice fraction in FAMOUS is generally greater than in HadCM3 (Jones et al. 2005). The AMOC has a similar pattern as in HadCM3, but the overall strength is about 17 Sv, which is a little weaker and does not penetrate as far north (Smith et al. 2008). CCSM3 is a coupled ocean–atmosphere–sea-ice–land climate model without flux adjustment (Collins et al. 2006). The atmospheric model, CAM3, has 26 hybrid coordinate levels in the vertical and ~3.75° resolution in the horizontal. The land model uses the same resolution as the atmosphere with a dynamic global vegetation module, and each grid box includes a hierarchy of land units, soil columns, and plant functional types (Oleson et al. 2003; Yeager et al. 2006). The ocean model is the NCAR implementation of the Parallel Ocean Program (POP) with 25 levels in the vertical. The longitudinal resolution is 3.6° and the latitudinal resolution is variable, with finer resolution near the equator (~0.9°). The sea ice model with the same resolution as POP is the Community Sea Ice Model (CSIM), a dynamic-thermodynamic model that includes a sub-grid-scale ice thickness distribution (He 2011). Temperatures and atmospheric circulations in CCSM3 are generally consistent with present-day observations (Otto-Bliesner et al. 2006). The NH sea ice edge is too extensive in a small region of the Greenland Sea (Bryan et al. 2006). Multi-decadal averages of AMOC in CCSM3 are roughly 16 Sv, which is at the lower end of observational estimates, with a similar pattern of latitude-depth distribution as the observations (Bryan et al. 2006). The CCSM3 simulation included here has been performed by the TraCE-21ka project (http://www.cgd.ucar.edu/ccr/TraCE/). The resolutions and complexities among these models are varied (Table 3.1). LOVECLIM represents the ocean component in more detail than the atmosphere module, while HadCM3 has a more detailed atmosphere component. The CCSM3 is in the middle among the models in terms of complexity of ocean and atmosphere components. Moreover, the sensitivities to climate forcings vary among the involved models. The ranges of global temperature increments spread from 2 °C in LOVECLIM to 4 °C in FAMOUS in response to a doubling of CO2 radiative forcing. In addition, the sensitivity of Atlantic meridional overturning circulation (AMOC) to freshwater perturbations between the models is not constant. The AMOC decreases by almost 70% in response to a 0.5 Sv freshwater release
Table 3.1 Summary of participating climate models
Ocean, Sea ice, Atm, Veg
Ocean, Sea ice, Atm, Veg
Ocean, Sea ice, Atm, Veg
Ocean, Sea ice, Atm
Sea ice model
dynamic-thermodynamic model (CLIO)
dynamic-thermodynamic model (CSIM)
zero-layer thermodynamics zero-layer model thermodynamics model
Resolution ocean (lat×lon)
ECS_radiative forcingb (°C)
1.9; Goose et al. (2010)
2.7; Randall et al. (2007)
3.0; Jones et al. (2005)
4.2; Jones et al. (2005)
0.1 Sv in 100yr => 21% decrease in AMOC (Stouffer 2006); 0.1 Sv in 1 kyr => 30% decrease in max AMOC (Goose 2010)
0.1 Sv in 300yr => 52% decrease in AMOC (He 2011); 0.1 Sv in 100yr => 22% decrease in AMOC (of CCSM2c, Stouffer 2006)
0.1 Sv in 100yr => 25% decrease in AMOC (Stouffer 2006)
0.5 Sv in 100yr => 70% decrease in AMOC (Smith & Grogory 2009)
AMOC sensitivity to FWF (under PI state)
a: FAMOUS is a fast version of HadCM3; b: under PI climate state; c: an earlier version of CCSM3 Table 3.2 Main features of the setup of involved simulations
Berger & Loutre (1991)
Berger & Loutre (1991)
Loulergue et al. 2008; Schilt et al. 2010
Joos & Spahni 2008
Spahni et al. 2005; Loulergue et al. 2008
Spahni et al. 2005; Loulergue et al. 2008
Eq_11.5 ka (1.2 kyr)
Ref. of simulation
Zhang et al. 2016
Singarayer & Valdes, 2010
Prescribed forcing & ref.
for 100 yr in FAMOUS, which is larger than in other models (Smith and Gregory 2009). For instance, in response to 100 yr freshwater release at the rate of 0.1 Sv in the N Atlantic, AMOC reduces about 21%, 22% and 25% in LOVECLIM, HadCM3, and CCSM2 (earlier version of CCSM3) respectively (Stouffer et al. 2006). In addition, the AMOC decreases by 52% in CCSM3 model when 0.1 Sv freshwater is appliedfor 300 yr (He 2011), and declines by 30% in LOVECLIM when 0.1 Sv freshwater lasted for 1 kyr (Goosse et al. 2010a). The AMOC sensitivities among the involved models are slightly smaller than in another ensemble mean of more models (Stouffer et al. 2006), in which about a 30% AMOC weakening in response to 0.1 Sv water release for 100 yr are reported. Stouffer et al. (2006) also show that the EMICs generally obtain similar AMOC responses as GCMs, although the latter have more pronounced patterns and detailed boundaries. Prescribed climate forcings We included the major climatic forcings in terms of insolation variation on orbital scale (ORB), greenhouse gases (GHG) in the atmosphere and decaying ice sheets. Other forcings, such as the solar constant and aerosol levels, were kept fixed at preindustrial values in all simulations. On the orbital scale, the seasonal and latitudinal variations of incoming solar radiation at the top of the atmosphere are determined by the astronomical parameters, including eccentricity, obliquity and longitude of perihelion. These parameters in the simulations were either obtained from Berger (1978) or Berger and Loutre (1991), which indicates that the ORB forcing is comparable across all simulations. The ORB forcing in the extratropical Northern Hemisphere (NH) is primarily characterized by decreased summer insolation over the course of Holocene. For instance, June insolation at 65°N decreased by 50 Wm-2 (Fig. S3.1). GHG concentrations are based on the ice core measurements of CO2, CH4 and N2O (Loulergue et al. 2008; Joos and Spahni 2008; Schilt et al. 2010). The GHG radiative forcing increases by about 1 W/m2 during the Holocene, with a small peak at 10 ka and a low value at 8 ka before gradually increasing to 0 ka level (Fig. S3.1). The GHG forcing in all simulations is comparable, because the GHG concentrations were exactly the same in LOVECLIM, HadCM3, and FAMOUS; and the description on GHG in CCSM3 matches well with these GHG concentrations (He 2011). Generally, the setup of GHG and ORB forcing was in line with the PMIP3 protocol (http://pmip3.lsce.ipsl.fr). Ice-sheet forcings (ISC), including the spatial extent and topography of the Laurentide Ice Sheet (LIS) and Fennoscandian Ice Sheet (FIS), were loosely constrained by geological evidence, as the PMIP4 guidelines were not yet available at the time of the simulations. The spatial extent in the LOVECLIM simulation is based on the dating results of geological remain of glaciers (e.g. morines) and their correlations (Dyke et al. 2003; Occhietti et al. 2011; Hughes et al. 2015; Cuzzone et al. 2016). The extents of ice sheets in the CCSM3 and HadCM3 simulations were based on the ICE-5G reconstructions (Peltier
CHAPTER 3 2004), and are slightly larger than in the LOVECLIM simulation (Fig. 3.1). The topography of ice sheets in LOVECLIM was based on the results of Ganopolski (2010), and was comparable with the ice sheet topography in CCSM3, FAMOUS and HadCM3 that was derived from ICE-5G reconstruction (Peltier 2004). Based on above reconstructions, the retreating ice sheets were updated every 250-yr in LOVECLIM and CCSM3 and every 1-kyr in FAMOU and HadCM3. In FAMOUS and HadCM3 the landsea mask was also updated every 1000 yrs, according to the ice sheet reconstruction of Peltier (2004). Figure 3.1 Ice sheet related forcings during the early Holocene. (a) indicates prescribed freshwater flux (in mSv) into oceans, and (b) shows the area of ice sheets (in km2) in the simulations
The freshwater flux (FWF) in the simulations was constrained by geological data, including the indicators of retreating ice sheets and freshwater routing proxies, such as detrital carbonate and ice rafted detritus (Carlson et al. 2007; Jennings et al. 2015). Meltwater from the Antarctic ice sheet was treated as a background signal and was not taken into consideration in the present study, because high frequent FWF discharges mainly contribute to short-term variations and the overall contribution was relatively small in the long term (Golledge et al. 2014). The FWF was updated at irregular time intervals (the smallest interval was 20 yr) as shown in Figure 3.1. The FWF in the LOVECLIM simulation was generally larger until 9 ka, and ended earlier than in the CCSM3 simulation. The freshwater flux in FAMOUS was slightly different with enhanced FWF peaks at around 11 and 8 ka (Fig. 3.1). However, the Holocene CCSM3 and FAMOUS simulations were parts of longer transient simulations that started at the LGM, and thus with a huge freshwater release at the beginning of the Younger Dryas that may still have impacted the climate due to the large thermal inertia of oceans (Renssen et al. 2006). In the HadCM3 simulation, no additional freshwater fluxes were applied, except for the input from rivers. 3.2.2 Setup of simulations The LOVECLIM simulation is an 11.5 kyr long transient run that was initialized from an
equilibrium experiment (more details are provided in Zhang et al. 2016). The simulation was forced by the annually-varied ORB and GHG throughout the whole period. In addition, the prescribed ISC was included with a time step of 250 yr until 6.8 ka when the LIS eventually disappeared, and associated FWF was also applied with a stepwise time series (Fig. 3.1). The CCSM3 simulation was taken from the TraCE-21ka project, which is a 21kyr long simulation forced by transient ORB, GHG, ISC and freshwater forcings. The ISC were modified every 500 yr based on the ICE-5G reconstructions. Freshwater was discharged with irregular time steps until 6 ka (He 2011). The FAMOUS simulation is also the Holocene part of a 21-kyr simulation that was forced by the transient GHG and ORB forcings, with prescribed ISC and freshwater from the melting of ice sheets. The HadCM3 results were derived from two sets of snapshot experiments at 1-kyr temporal intervals; and the GHG, ORB and ice sheet forcings were updated in each snapshot. The main reason for including these experiments is the high spatial resolution of the HadCM3 model. The differences between these two sets of experiments are the ice sheet configurations that were based on ICE-I5G and ICE-I6G respectively, and they were accordingly named as HadCM3-I5G and HadCM3-I6G. Given the similarity between HadCM3-I5G and HadCM3-I6G, they are generally considered as the whole HadCM3 simulation unless specifically indicated. Each of these snapshot experiments was initialized from a spun-up pre-industrial simulations (Singarayer and Valdes, 2009) and were run for at least 300 yr with fixed GHG and ORB forcings, of which the last 30yr average was taken as representative of the climate during the corresponding time window. Main information on these simulations are summarized in Table 3.2. The model name is also used as the indicator of corresponding simulation in the present study to reduce redundancies. Temperatures presented are simulated (near) surface temperatures, and shown as the deviations from 0 ka (refers to the pre-industrial). To obtain the overall temperature trend throughout the Holocene, the ensemble mean was calculated by averaging all transient simulations. This implies that the HadCM3 results were not included in the ensemble and are separately shown in figures.
3.3 Results 3.3.1 Simulated temperature in the NH extratropics Although HadCM3 and FAMOUS show a slightly cooler climate until 5 ka than CCSM3 and LOVECLIM, the simulated summer temperature trends in the NH extratropics (30– 90°N) are generally consistent across the models (Fig. 3.2a). The overall patterns reveal an early-Holocene warming, a maximum temperature at 10~7 ka, and a cooling toward 0 ka. In summer, the magnitudes of early-Holocene anomalies in these models roughly correlate to their climate sensitivities, as high sensitivity implies a large climate response to given forcings (Fig. 3.2b, Table 3.1). The simulated annual mean temperature shows a gradual
CHAPTER 3 warming of about 2 °C by 4 ka, after which the temperature stays at the 0 ka level (Fig. 3.2b). There is a 1 °C spread in simulated annual mean temperature before 5 ka, with the coolest climate in FAMOUS and warmest in LOVECLIM. An abrupt cooling at ~8.5 ka is found both in the CCSM3 and FAMOUS simulations. In order to approximately evaluate the reliability of the simulations, the simulated results are briefly compared with proxy data. The proxy data are stacked temperatures using records from the 30–90°N latitudes band (Marcott et al. 2013). The brief model–data comparison reveals an overall agreement on temperature trends, despite a slightly stronger early-Holocene warming until 9.5 ka suggested by the proxy data. In particular, the simulated summer temperatures show better agreement with proxy data than annual mean temperatures, which is consistent with the suggestion of potential seasonal biases in the biological proxy data (Lohmann et al. 2013; Liu et al. 2014). Figure 3.2 Comparison of stacked proxy reconstruction with simulated summer (a) and annual mean temperature (b) in NH extratropics (over 30–90°N), shown as a deviation from 0 ka in °C. The stacked temperature reconstruction with 1σ uncertainty (grey band) is based on Marcott et al. (2013). The proxy curve is the same in (a) and (b), although the authors interpreted it as annual mean.
Strong spatial patterns of simulated temperatures are found when zoomed into regional scale. To further illustrate the regional climate response to relevant forcings, two periods, 11.5 ka and 6 ka, were selected as specific time windows. These two time windows either represent the period when the dynamic ice sheets played important roles, or serve as a benchmarking epoch (i.e. by PMIP) Simulated temperatures generally show larger negative anomalies at 11.5 ka in comparison with that at 6 ka (Fig. 3.3). At 11.5 ka, the anomalies of simulated annual mean temperatures are about -1 to -5 °C, with the exception over Beringia where positive temperature anomalies are found in LOVECLIM and CCSM3. The temperatures at 6 ka show latitudinal pattern. At the mid-latitude, simulations suggest similar or slightly lower temperature in comparison with 0 ka. At the high-latitudes, annual mean temperatures are 1 to 3 °C warmer in CCSM3 and LOVECLIM; while similar or -0.5 °C cooler climate is represented by FAMOUS and HadCM3. Another significant feature of climate during the early Holocene is that the multi-model consistencies are regionally heterogeneous and are generally larger in winter than in summer. Target regions
Figure 3.3 Spatial distribution of simulated temperature anomalies (in °C) in NH extratropics for the time windows of 11.5 ka (a) and 6 ka (b). Provided anomalies are relative to 0 ka
were further selected according to the spatial pattern of climate response to dominant forcings (Fig. S3.2), and the Holocene temperature trends over these regions were analyzed. Based on their consistencies, these selected regions were divided into two groups, representing consistent and inconsistent areas. The first group is formed by Greenland, N Canada, NE Europe, NW Europe and central-west Siberia, as all models give similar early-
CHAPTER 3 Holocene temperature trends. The second group with mismatched temperatures includes Alaska, Arctic and E Siberia, as opposite temperature trends (especially in winter) across the models are found. 3.3.2 Temperature over the regions with good inter-model agreements The simulations show overall good agreements over N Canada, NW Europe, NE Europe, Greenland and central-west Siberia. The ensemble mean temperatures generally rise from the cold initial state until 6~7 ka, followed by gradual decrease to the 0 ka level, with exception of summer temperature in NE Europe and central-west Siberia. As indicated by the ensemble mean, the magnitudes of this cool early Holocene vary with regions (Fig. 3.4). In particular, a considerably cool early Holocene climate is found in N Canada, with 5 °C lower ensemble mean in summer and 10 °C lower in winter; whereas only a minor cooling is present in NE Europe and central-west Siberia, with about 4 °C cooling in winter and 1– 3 °C warming in summer. Temperatures in Greenland and NW Europe show intermediate values, with a cooling of 2–3 °C in summer and around 8 °C in winter. Only minor inter-model variations are found in Greenland and NE Europe within the range of 3 °C during the early Holocene. In Greenland, all simulations indicate about 3 °C cooler summer conditions at 11.5 ka in comparison with 0 ka, and this low temperature rise to 1 °C at around 7.5 ka. In winter, all simulations suggest low temperature at 11.5 ka with about -8 °C anomaly of the ensemble mean temperature, notwithstanding the temperature spread between individual simulations up to 5 °C. In NE Europe, the simulations suggest a warming of 1 °C in summer and cooling of 4 °C in winter at 11.5 ka, with a 2 °C multisimulation spread until 9 ka. From 11.5 ka onward, simulated winter temperatures slowly rise to the preindustrial level, while in summer the temperatures anomalies show a gradual decrease of about 1 °C. Over central-west Siberia, the temperatures in winter show 3–4 °C warming trends during the Holocene, while the simulations indicate 3 °C cooling summer. Within the frame of overall consistent trends, slightly different temperatures are found in N Canada and NW Europe during the winter. In N Canada, the small early-Holocene temperature anomalies in LOVECLIM might be related to the use of a fixed modern landsea mask throughout the simulation, as which implies underestimation of the earlyHolocene albedo over the Hudson Bay. The jump of simulated winter temperature in FAMOUS at 8 ka might be related to the spike in FWF forcing (Fig. 3.1). In NW Europe, the magnitudes of this cooling generally increase in the order of LOVECLIM, CCSM3, HadCM3 and FAMOUS. One exception is that winter temperature anomalies in HadCM3 and FAMOUS rapidly rise to a distinct peak of +2 °C at around 9~10 ka, leading to their high temperatures at round 9 ka. The warm peaks in FAMOUS and HadCM3 are mainly caused by a response of sea ice to the opening of the Bering Strait (Fig. S3.4). Overall, these roughly consistent patterns demonstrate that over the above regions, forced millennial climate change exceed the climate variability during the early Holocene.
Figure 3.4 Temperature trends (in °C) over the regions where multiple simulations have good agreements, and corresponding multi-model ensemble mean (based on three transient simulations). The grey indicates the ensemble range
3.3.3 Temperatures over the regions with less multi-model consistency In the regions of Alaska, Arctic, and E Siberia, simulated temperatures show poor consistency among the different models (Fig. 3.5), especially in winter, when both positive and negative early-Holocene anomalies are suggested by the different models. In Alaska, the spread of the simulated winter temperatures at 11.5 ka ranges from 2 °C warmer in LOVECLIM to 4–6 °C cooler in FAMOUS and HadCM3 in comparison with 0 ka. This distinct multi-model variation in winter is thus up to 8 °C, which is considerably larger than in summer when the inter-model variation is only 1 °C. Over the Arctic, the discrepancies between the simulations also primarily exist in winter when they are up to
CHAPTER 3 8 °C. At 11.5 ka, the winter temperature anomaly is slightly above 0 °C in LOVECLIM while more than 8 °C cooling is produced in FAMOUS. Nevertheless, the ensemble mean temperature suggests 1 °C cooling in summer and 4 °C warming in winter throughout the Holocene. Relatively large multi-simulation differences are found over E Siberia in both summer and winter, reaching up to 3 °C at the onset of the Holocene. The simulations show decreases in summer temperatures over E Siberia throughout the Holocene, with the largest decrease (more than 4 °C) in CCSM3. This large variation in summer is primarily contributed by exceedingly warm early-Holocene conditions in CCSM3. In winter, over 2 °C cooling is simulated by LOVECLIM during the Holocene contrasting with up to 5 °C warming in FAMOUS. The ensemble mean of simulated temperatures declines by almost 4 °C in summer, but generally rises by 2 °C in winter despite a small drop at ~8.5 ka.
Figure 3.5 Simulated temperatures over the regions where temperatures are less consistent across the simulations
In general, although the simulations generally agree on temperatures over the large-scale NH extratropics, some regions show better inter-model agreements than others when zooming into the regional scale. The mismatches among the simulations are large during the early Holocene, and can be outlined as follows: 1) warm winter climate over Alaska in LOVECLIM in comparison with various degrees of cooling in other models; 2) large negative temperature deviations (from 0 ka) over the Arctic in FAMOUS contrasting with slightly positive values in LOVECLIM; 3) a stronger summer warming over E Siberia in CCSM3 than in other simulations, and a warm winter in LOVECLIM over E Siberia contrasting with the cool climate in HadCM3.
3.4 Discussion Comparisons of multi-model simulations provide an opportunity to evaluate the performance of climate models in simulating climate response to radiative forcings and other boundary conditions. The following discussion will start from the above results with a focus on the regions where the simulated temperatures are different, and the causes of these mismatches will be investigated at two levels. We will firstly try to identify the direct causes of these inter-model discrepancies via a diagnosis of various climate variables. Subsequently, the potential origin of these divergent climate variables will be further examined. 3.4.1 Divergent climate variables lead to mismatched temperatures Mismatched Alaskan winter temperature The relatively warm early Holocene in LOVECLIM results from enhanced southerly winds induced by the LIS which bring warm air from the south. This enhanced southerly winds can be diagnosed by examining the anomalous atmospheric circulation over the ice sheets at 11.5 ka, in comparison with the ice free condition at 0 ka. The atmospheric circulations are indicated by geopotential height fields, which reflect anomalous geopotential to standard gravity at mean sea level, and high values represent high pressure near the surface. Although there are largely similar responses of geopotential heights to the existence of the LIS, such as enhanced values over the LIS, the degree of these anomalies differs across individual models (Fig. 3.6). At 11.5 ka, winter geopotential heights in LOVECLIM and FAMOUS are up to 50 m higher over the center of the LIS than at 0 ka, which is larger than the enhancement of 30 m in HadCM3. The LOVECLIM and HadCM3 simulations have similar spatial patterns of Northern Annular Mode with a lower value over polar region than in the surrounding areas, but with larger anomalies of geopotential height in LOVECLIM than in HadCM3. In order to further analyse this anomalous geopotential height, the geopotential height differences between the LIS (50–75°N, 110–65°W) and the N Pacific (35–45°N, 180–120°W) were calculated. These calculated results are further standardized to changes regarding the 0 ka condition. This standardized difference of geopotential height in LOVECLIM was up to 70% at 11.5 ka, which declines with time toward 0 ka (Fig. S3.3). The FAMOUS and HadCM3 also show a similar decreasing trend in geopotential height changes during the Holocene, but with a smaller magnitude than in LOVECLIM. A similarly anomalous atmospheric circulation, with a comparable amplitude of changes as in HadCM3, is indicated by the surface pressure anomaly in CCSM3 (figure is not shown). Therefore, these different anomalies in geopotential height fields among individual models can lead to divergent climate at the edge of the ice sheet where this signal is weak and thus even a minor difference is visible. In Alaska (to the west of the LIS), the intense warm climate during the early Holocene in LOVECLIM is primarily
CHAPTER 3 caused by the relatively strong gradients of geopotential height that induce southerly winds, bring warm air from the south and increase the local temperature in Alaska. This effect lasted until the final disappearance of the LIS at 6.8 ka, after which the small decreasing trend in temperature can be associated with a sea-ice cooling effect on coastal regions of northern Alaska (Fig. 3.7). Compared with the HadCM3 simulation, lower temperatures in FAMOUS can be explained by more extended sea ice in this model (Jones et al., 2005) and a stronger anticyclone over Alaska (Fig. 3.6).
Figure 3.6 Atmospheric circulation anomaly induced by the ice sheets at 11.5 ka, shown as the anomalous geopotential height fields (11.5 ka - 0 ka, in m) in at 800 hPa in LOVECLIM, at 850 hPa in FAMOUS and HadCM3. Associated winds anomalies are indicated by the vectors.
Previous studies have analysed the effect of ice sheet on atmospheric circulation under the LGM condition. For instance, the experiment performed by the Polar MM5 atmospheric model has shown that 500-hPa geopotential height over the LIS enhanced by 260 m in January and 70 m in summer (Bromwich et al. 2004, 2005). The intensified atmospheric baroclinicity induced by the ice sheets is suggested as one of the primary mechanisms behind the atmospheric circulation changes under the LGM boundary condition (Bromwich et al., 2005). Another explanation for these effects during the LGM is that air flow can be deflected or split around an anticyclone over the LIS (e.g. Bromwich et al., 2004). Sensitivity experiments of the atmospheric circulations response to idealized circular mountains reveal that the orography effects of ice sheet is highly depended on the scale of the ice sheet (Yu and Hartmann 1995). Therefore, given the scale of the ice sheets in our experiments, the ice-sheet effects on atmospheric circulation during the early Holocene are much smaller in comparison with the LGM, as shown in our simulations.
Figure 3.7 Distribution of maximum sea ice. (a) represents sea ice concentration in FAMOUS and CCSM3, and (b) shows the thickness of sea ice (in m) in LOVECLIM and HadCM3
Mismatched winter temperature over Arctic Over the Arctic, the inter-model divergences of winter temperature during the early Holocene are associated with the different sea ice across the individual simulations. Although it is difficult to establish if the sea-ice changes are a cause or effect in relation to Arctic temperatures, sea ice plays a critical role in the climate system via sea-ice related feedbacks. Firstly, the areal extent of sea ice cover determines to which extend the albedoinvolved feedback occurs, because the albedo of sea ice is typically up to 0.5~0.6 and significantly higher than in open ocean at the high latitudes. Furthermore, the thickness of sea ice influences the amount of heat that is released from the relatively warm ocean to the cold atmosphere (Renssen et al. 2005; Holland et al. 2006). In these four simulations, the early-Holocene sea ice anomaly is overall positive compared with 0 ka (Fig. 3.7). However, the absolute sea ice area and the magnitude of anomalous sea-ice cover (between 11.5 ka and 0 ka) varies with the models. At 11.5 ka, the simulated NH sea ice in March is in the order of 1012 m2 and is decreasing by the order of FAMOUS, CCSM3, HadCM3 and LOVECLIM with the largest NH sea ice area of up to 32×1012 m2, which is two times large than the minimum (Fig. S3.4). The magnitudes of anomalous sea ice area in these models generally follow the same order as in the absolute sea ice cover, with the largest anomaly in FAMOUS (Fig. S3.4). It is well-known that accurately simulating sea ice in coupled climate models is challenging because of the highly dynamic nature of sea ice in both spatial and temporal dimensions, which can also be seen from the wide spread of simulated sea ice even under pre-industrial conditions. At 0 ka, the NH sea-ice area in these models varies from 13.5 to 24×1012 m2 (Fig. S3.4). Studies have suggested that the NH sea ice
CHAPTER 3 was overestimated in CCSM3, FAMOUS and HadCM3 with various degrees and spatial patterns (Gordon et al. 2000; Jones et al. 2005; Bryan et al. 2006). For instance, CCSM3 overestimated sea ice in the Labrador Sea, while HadCM3 and FAMOUS simulated more sea ice in the Barents Sea (Gordon et al. 2000; Jones et al. 2005; Bryan et al. 2006). Compared with these total NH sea ice areas, simulating a consistent anomalous spatial distribution of sea ice is even more difficult, as revealed by various sea ice patterns across these simulations. In particular, thickness anomalies of sea ice over the Greenland Sea in HadCM3 are larger than in LOVECLIM. Therefore, the cooler climate in HadCM3 than in LOVECLIM is partially related to the insulation of thick sea ice in HadCM3 at 11.5 ka. Simulated strong winter cooling at 11.5 ka in FAMOUS could be related to enhanced albedo feedback due to extensive sea ice coverage. It is also noticeable that considerable changes of NH sea ice area in FAMOUS occurred between 9 and 8 ka (Fig. S3.4), which explains why temperature substantially changed around that time. These changes results from the opening of the Bering Strait and short-lived FWF forcing in the simulation. A brief comparison with proxy-based sea ice reconstructions broadly supports this simulated extended sea ice extent during the early Holocene. Proxy-based sea-ice reconstructions show an overall decreased tendency in sea ice extent throughout the Holocene, for instance, in the circum Arctic areas as suggested by compiled dinocyst data. The magnitudes of sea ice extension are spatially heterogeneous (de Vernal et al. 2013). In particular, dinocyst assemblages suggest positively anomalous early-Holocene sea ice over the regions where closely connected to Atlantic, such as Labrador Sea and Greenland Sea (de Vernal et al. 2013), which roughly agrees with more extended early-Holocene sea ice in the simulations. Nevertheless, a more detailed model–data comparison of sea ice history in the Arctic Ocean is limited by spatial coverage of proxy records. AMOC changes are of great importance for the mid-to-high latitude climate because the strength of AMOC determines the amount to meridional heat transport in the Atlantic, which influences the distribution of sea ice and thus the energy balance through albedoinvolved feedbacks (Renssen et al. 2005; Roche et al. 2007; Thornalley et al. 2009). The sensitivity studies have shown that the climate in NH extratropics is sensitive to changes of AMOC strength during the early Holocene (Zhang et al. 2016). As a response to freshwater release, the AMOC in our three transient simulations shows some differences in terms of the degree of these AMOC reductions during the Holocene (Fig. 3.8). To provide a quantitative comparison of the ocean circulation in the different models, the maximum AMOC strength was calculated by taking the maximum overturning value in the box of 500–2000 m, 34°S–50°N, according to the definition of Drijfhout et al. (2012). These AMOC values during the early Holocene are in the range of 13~15 Sv, and are thus consistent among the different models. Over the course of the Holocene, the AMOC in LOVECLIM rose by about 8 Sv, and the maximum AMOC in CCSM3 increase by about 4 Sv (Fig. S3.6). In contrast, the maximum AMOC in FAMOUS is relatively stable and show
no trend change throughout the Holocene, with an exceptional bump at around 9 ka due to the opening of the Bering Strait and FWF forcing (Fig. S3.6). The inter-model discrepancies mainly result from the different AMOC strengths under the pre-industrial condition, with 22 Sv in LOVECLIM, 16 Sv in CCSM3 and 12 Sv in FAMOUS (Fig. S3.6). Previous studies have also shown these wide-spanned AMOC values under preindustrial conditions, ranging from 12 to 27 Sv (Gregory et al. 2005; Stouffer at al. 2006; Yeager et al. 2006). These divergent AMOC results are caused by multiple factors. Firstly, the absence of uniform definition and characteristics of AMOC hampers obtaining a consistent AMOC. The AMOC strength is based on the meridional overturning streamfunction, and the definitions of AMOC strength vary in the literature in terms of the range of depths and latitudes (Friedrich et al 2010; Hofer et al 2011; Drijfhout et al 2012). The meridional overturning is represented by vectors with two main directions that form a loop over the whole ocean basin, implying that the averaging procedure, supposed to obtain an overall signal, is not appropriate for the AMOC. Meanwhile, even though the general patterns of the AMOC are similar, various representations of ocean circulations regarding detailed structures across models reduce the consistency of simulated AMOC. This is for instance the case for the location of maximum flow and the depth of boundary between the mean southward and northward flows (Fig. 3.8). Furthermore, the AMOC tends to be overestimated or underestimated in some models. Multi-decadal average of maximum AMOC overturning below 500 m and north of 28°N is roughly 16 Sv in CCSM3, which is at the low end of observational estimates (18±3–5Sv) (Talley et al. 2003; Yeager et al. 2006). The AMOC in FAMOUS has a mean of 17 Sv, which is weaker than that in HadCM3 and it does not penetrate as far north (Smith et al. 2008). The maximum AMOC in LOVECLIM reaches 22 Sv, and deep convection in the model occurs both in the Greenland-Norwegian Sea and the Labrador Sea (Goosse et al. 2010a). Those values are in the upper range of the values given by other models (Ganachaud and Wunsch 2000; Gregory et al. 2005; Rahmstorf et al. 2005). Another reason for divergent AMOC responses during the early Holocene in our results is the model-dependent AMOC sensitivity to FWF and non-uniformed FWF, as presented in Sections 3.2.1 and 3.2.2. Nevertheless, the maximum AMOC is used here only to roughly represent the general ocean circulation and these various AMOC responses can indirectly contribute to the divergences of simulated Arctic temperatures. A sluggish AMOC during the early Holocene in LOVECLIM and CCSM3 reduces the heat transport from the south, facilitating more extended sea ice and thus activating albedo feedbacks. An overall weakening of the AMOC during the early Holocene has been suggested by proxy-based studies. Measured 231Pa/230Th ratio in sediment is a proxy of AMOC, as its ratios indicate amount of 231Pa exported from the Atlantic to the Southern Ocean and thus the rates of meridional overturning circulation (McManus et al. 2004). Therefore, reduced 231 Pa/230Th ratios in the N Atlantic during the early Holocene show an overall weakening of
CHAPTER 3 the AMOC. Further quantitative comparisons with these studies and precise quantification of the contribution of different ocean circulations to temperature deserve a separate article given the complexity and highly dynamic feature of ocean circulations. Mismatched temperature in east (E) Siberia As a main contributor to mismatches in multi-simulation temperatures, the intense earlyHolocene warmth in CCSM3 primarily results from its large negative albedo anomaly (more than -0.2 compared to 0 ka) in E Siberia. The climate is known to be highly sensitive to surface albedo changes (Romanova et al. 2006), and thus inter-model variations in albedo could cause discrepancies in simulated temperatures. The surface albedo anomalies between the early Holocene and the preindustrial are spatially heterogeneous among individual simulations. On the one hand, the anomalous albedos over the ice sheets are roughly similar. For instance, an enhanced surface albedo of up to 0.6 at 11.5 ka over the LIS is consistently found in all simulations (Fig. 3.8), despite a small exception over the Hudson Bay in LOVECLIM due to the fixed modern land-sea mask here. On the other hand, early-Holocene albedo anomalies are found widely different across these models over the regions where the surface albedo is primarily influenced by vegetation, snow cover and sea-ice cover. In E Siberia, summer surface albedo in CCSM3 differs from that in other models over the course of the Holocene. In CCSM3, overall Holocene albedos are higher than 0.36, and they show a rising trend during the Holocene with a rapid increase at 3 ka; whereas other models suggest a stable Holocene trend of summer albedo with absolute values ranging from 0.15 to 0.2 (Fig. S3.7). This increasing Holocene albedo in CCSM3 is anti-correlated with a decreasing temperature trend, and it is clear that the 2 °C decline around at 3 ka is related to an albedo rise. This negative anomalies and increased trend are mainly due to high albedo at 0 ka, which is up to 0.65. A further investigation on the snow cover in the simulation shows that most of E Siberia is covered by snow during the summer season, which is obviously an overestimation, as it would imply the inception of a continental ice sheet. By contrast, other simulations show low albedo (around 0.2), indicating a vegetation-covered surface, which makes more sense given the present-day landscape. In winter, the spread of inter-model temperatures over E Siberia result from multiple factors. The different albedo responses relating to snow cover is an important factor explain those spread across the simulations, with relatively warm climate in LOVECLIM corresponding to low albedo, and low temperature in FAMOUS associated with high albedo. Moreover, sea ice changes influence the temperature of the coastal Siberia. For example, the temperature bump at around 9 ka in FAMOUS and a mild increase in LOVECLIM are caused by temporal variations of sea ice (Fig. S3.5). Additionally, the warm winter in LOVECLIM is partially associated with the enhanced southerly winds mentioned previously.
Figure 4.8 Surface albedo anomalies (shown as fractions) at 11.5 ka compared to 0 ka.
3.4.2 Potential sources contributing to inter-model divergences of climate variables Uncertainty of ice-sheet-related forcing With decaying ice sheets in North America and Fennoscandia, the uncertainty in the FWF forcings during the early Holocene is mainly related to the total volume of ice melts involved, the location of discharge and the timing of discharge, which further impacts the Holocene simulations. Different FWF-forcing scenarios and associated different AMOC responses have climate impacts through adjustments in heat transport and sea-ice related feedback (Kageyama et al. 2009; Blaschek and Renssen 2013). The total amount of FWF can be constrained from both the ocean and land perspectives, as the FWF discharge serves as the link of water exchange between the ocean and continental ice. For instance, the fossil-coral-based estimates of far-field sea level change can reflect the total amount of FWF from the ocean perspective (Lambeck et al. 2014). From the land side, the total FWF amount can be roughly constrained by the geological indicators of ice-sheet retreat (Peltier 2004). For the Holocene a total amount of freshwater equivalent to 60 m of sea level was released into the ocean between 11.5 and 6 ka with a large contribution from the LIS (Peltier 2004; Lambeck et al. 2014). Sensitivity studies have further disclosed that, apart from total amount, various temporal distributions and geography locations can potentially induce different response in ocean circulation (Roche et al. 2010). Although ocean sediment data (e.g. detrital carbonate, ice rafted detritus) and geochemical tracers (e.g. δ 18O, 87Sr/ 86Sr, U/Ca) can provide certain constraints on FWF routing (Carlson et al. 2007; Jennings et al. 2015), it is uncertain how this amount of water was distributed spatially and temporally. Firstly, various magnitudes of FWF are suggested by different proxies. For
CHAPTER 3 instance, the geochemical tracer U/Ca suggests a slightly larger FWF discharge in St. Lawrence River between 12 and 11 ka than what the indicator δ18O does, probably because additional factors (such as temperature and weathering) modulate the signal of changes in FWF (Carlson et al. 2007). Furthermore, the temporal distributions of FWF in different estimates are not identical either. For instance, these proxy-based FWF estimates differ from the model-based estimates with maximum reached at different time periods (Licciardi et al. 1999; Carlson et al. 2007), which is related to the difficulty in accurately dating samples. Additionally, approaching on well-agreed geographical locations of FWF discharge (spatial distributions) is hindered by spatially sparse sites of proxy records. Overall, the FWF is hugely uncertain during the early Holocene, especially in terms of discharge rate, location and timing. This uncertainty is also reflected by FWF differences of the four experiments discussed here, which potentially induces certain degree of intermodel divergences. To avoid the FWF related influences, it will be substantially beneficial to construct FWF protocols and apply them in all participating models for next intercomparison project, as that would allow us to focus on dominant climate processes and feedbacks. Impact of inter-model differences in climate sensitivities Climate sensitivities are used to broadly indicate the sensitivity of the climate system to both radiative forcing and FWF forcing in the present study. The sensitivity for radiative forcing generally refers to the change in the global annual mean surface air temperature (in °C) in response to doubling CO2 (Knutti and Hegerl 2008), representing a global average. Yet the CSr shows temporal and spatial patterns when examined in detail. The spatial patterns are reflected by the geographical distribution of sensitivity coefficient, which are partially due to local feedback processes (Boer and Yu 2003). The temporal variations are mainly because that the CSr is a function of baseline climate and involves a series of processes on different timescales (Boer and Yu 2003). In particular, multiple studies have quantitatively examined CSr for decades and have been found it vary with different climate states (Boer and Yu 2003). The CSr generally decreases with warmer climate and increases with colder climate (Boer and Yu 2003; Knutti and Hegerl 2008). The feedback processes, such as those involving water vapor, lapse rate, surface albedo and clouds, can be activated at different degrees under the different climate states (Boer and Yu 2003; Randall et al. 2007). For simplicity, the CSr is assumed to be linear over small ranges of climate change as vital global atmospheric feedback remain close to a constant with temperatures when the threshold value is not exceeded (Randall et al. 2007). Supposedly, the CSr during the early Holocene was slightly larger than that of 0 ka within this linear assumption, owning to a slightly cooler early Holocene. By contrast, the CSr at 6 ka was slightly smaller with a similar linear assumption. However, the exact change rates of CSr in response to climate states are still controversial. For instance, recent studies suggest weaker CSr enhancements than previous findings in response to the same amount
of cooling (Kutzbach et al. 2013). The CSr varies among individual models, ranging from 2 °C in LOVECLIM to 4 °C in FAMOUS (Table 3.1). Given this high CSr in FAMOUS and positive radiative forcings during the early Holocene, the climate in FAMOUS would be expected to be warmer than in other simulations when an identical model response to the ice sheets is assumed. However, the overall cool climate in FAMOUS seems to conflict this expectation. A plausible explanation for this paradox is that the expected warmth was overwhelmed by the ice-sheet-related cooling. Moreover, the spatial heterogeneity of CSr could outweigh this expected overall warming at certain regions and thus also potentially contributes to this conflict. In addition, the FAMOUS results (at 11.5ka) were obtained from a full transient simulations since the LGM, which implies that the model still had a “memory” of the preceding cold climates. Further detailed discussions on the CSr spatial pattern and sensitivity experiments deserve more effort, which is beyond the scope of the present study. The sensitivity of the climate system to freshwater forcing (CSf) varies among these models, and associated influences on temperature are discussed in SI. Impacts of the model physics and resolution Model physics also contribute to the inter-models variations. By model physics, we refer to how climatic processes are represented in the model world, without considering the external radiative forcings. For instance, the formulation of a turbulent transfer coefficient in CCSM3 primarily contributes to multi-simulation variations in E Siberia by inducing intense warmth in CCSM3. The early-Holocene summer albedo in CCSM3 is reduced by more than 0.2 compared with 0 ka (Fig. 3.8), which primarily results from overestimated albedo at 0 ka since the value is more than 0.7. Such high albedo can be associated with the later adopted formulation of a turbulent transfer coefficient (Collins et al. 2006). According to an assessment of surface albedo (using MODIS data), this new formulation produces an extensive snow cover, because white-sky albedo in vegetated area might be insufficiently simulated and albedo increase with solar zenith angle probably is overestimated (Oleson et al. 2003). The spatial resolution of a model determines the overall level of detail in the model representation of climate processes. The detailed representation of key physical processes such as the barrier effect of topography can improve the accuracy of the simulation. The spatial resolution, however, is limited by the computer power, especially for the simulations spanning long time periods (e.g. palaeoclimate simulation in the order of kyr or even longer). Lunt et al. (2013) found that the resolution effect could partially explain multi-model differences, such as stronger cooling over African monsoon region in GCM than in EMIC, but also stated that this should be confirmed with further analysis. Some resolution-related patterns are observed when the atmosphere and ocean components are individually examined. For instance, the widely extended sea ice in FAMOUS is mainly
CHAPTER 3 caused by a relatively coarse spatial resolution in the ocean component (Gordon et al. 2000; Jones et al. 2005). Even though FAMOUS has similar physical and dynamical processes to HadCM3, this coarse resolution may lead of insufficient heat transport, such as in the Barents Sea, which ultimately leads to overestimated sea ice cover in that region (Gordon et al. 2000; Jones et al. 2005). This overestimated sea ice cover can cause cool Arctic climate through enhanced albedo-feedback (Renssen et al. 2005). In addition, the intense Alaskan warmth in winter in LOVECLIM might be related to its coarse vertical resolution with simplified levels, implying a relatively poor representation of the LIS topography. Nevertheless, to further investigate these resolution-related effects, more analyses, such as using proxy data to evaluate the simulated early-Holocene climate and applying a fully identical setup procedure, are still needed.
3.5 Conclusions Transient features of the early-Holocene climate potentially introduce large uncertainties in Holocene climate simulations. To narrow these uncertainties and analyze the temperature trends, we compared four Holocene simulations performed with different models. The main findings are outlined as following: 1) Consistently simulated Holocene temperatures in multi-model simulations Over the large scale of NH extratropics, the simulated temperatures are generally consistent among models with better agreements in summer than in winter, with overall temporal patterns of an early-Holocene warming, mid-Holocene maximum and gradual decrease toward 0 ka. On a regional scale, reasonably consistent temperature trends are found where climate is strongly influenced by the ice sheets, including Greenland, N Canada, N Europe and central-west Siberia. These simulated temperatures generally follow a similar pattern stated above. Within these general patterns, the magnitude of earlyHolocene warming slightly varies with regions. The strongest early-Holocene warming, up to 5 °C in summer and 10 °C in winter, is found in N Canada; whereas NE Europe and central-west Siberia show the least warming magnitude, with 4 °C warming in winter and 1–3 °C cooling in summer. An intermediate degree of warming is found in Greenland and NW Europe, with about 2 °C in summer and 8 °C in winter. Overall, these generally consistent temperature trends illustrate that forced climate change overwhelms the structural and parametric uncertainties, implying temperature trend are relatively well established in these regions. 2) Differences of the multi-model simulations and their direct causes Large inter-model variations exist in Alaska, the Arctic, and E Siberia. In particular, the signals of individual model simulations are incompatible during the early Holocene. On the one hand, the strong southerly winds induced by the LIS over Alaska and part of E Siberia result in an anomalous warm climate in LOVECLIM. Higher summer temperatures (1–
2 °C) over E Siberia in CCSM3 than in other models are caused by strongly negative albedo anomaly (over -0.2) between 11.5 ka and 0 ka, which ultimately is associated with a high albedo at 0 ka. On the other hand, the wide spread of simulated winter temperatures over the Arctic can be partially attributed to cold climate in FAMOUS due to its extensive sea ice cover. This extended early-Holocene sea ice cover influences the strength of the albedo-related feedback and could explain why the winter temperature in FAMOUS is 2– 3 °C lower than the ensemble mean. 3) Possible sources contributing to the different responses of climate variables The multi-model comparisons reveal that varied responses in the models can be caused by the model physics, model resolution and model-dependent sensitivities. For instance, the later adopted formulation of turbulent transfer coefficient in CCSM3 may cause an overestimated albedo over Siberia at 0 ka. Moreover, relatively simplified sea ice representation in FAMOUS may leads to overestimated sea ice cover. Also, the coarse vertical resolution in LOVECLIM might result in overestimated responses of atmospheric circulation to the LIS over Alaska. This inter-model comparison is partially hampered by the differences between the experimental setups and forcings, especially concerning the FWF, which has a major impact on the early-Holocene climate. Hence, we suggest that constructing a standardized FWF for the early Holocene can be advantageous for further model inter-comparisons.
Chapter 4 Holocene temperature evolution in the Northern Hemisphere high latitudes— model–data comparisons ____________________________________________________________________________________________________
Based on: Zhang, Y., Renssen, H., Seppä, H., Valdes, PJ. Holocene temperature evolution in the Northern Hemisphere high latitudes—data–model comparisons. Quat. Sci. Rev., 173,101–113, 2017.
Abstract Heterogeneous Holocene climate evolutions in the Northern Hemisphere high latitudes are primarily determined by orbital-scale insolation variations and melting ice sheets. Previous inter-model comparisons have revealed that multi-simulation consistencies vary spatially. We, therefore, compared multiple model results with proxy-based reconstructions in Fennoscandia, Greenland, north Canada, Alaska and Siberia. Our model–data comparisons reveal that data and models generally agree in Fennoscandia, Greenland and Canada, with the early-Holocene warming and subsequent gradual decrease to 0 ka BP (hereinafter referred as ka). In Fennoscandia, simulations and pollen data suggest a 2 °C warming by 8 ka, but this is less expressed in chironomid data. In Canada, a strong early-Holocene warming is suggested by both the simulations and pollen results. In Greenland, the magnitude of early-Holocene warming ranges from 6 °C in simulations to 8 °C in δ18O-based temperatures. Simulated and reconstructed temperatures are mismatched in Alaska. Pollen data suggest strong early-Holocene warming, while the simulations indicate constant Holocene cooling, and chironomid data show a stable trend. Meanwhile, a high frequency of Alaskan peatland initiation before 9 ka can reflect either high temperature, high soil moisture or large seasonality. In high-latitude Siberia, although simulations and proxy data depict high Holocene temperatures, these signals are noisy owing to a large spread in the simulations and between pollen and chironomid results. On the whole, the Holocene climate evolutions
CHAPTER 4 in most of regions (Fennoscandia, Greenland and Canada) are well established and understood, but important questions regarding the Holocene temperature trend and mechanisms remain for Alaska and Siberia.
4.1 Introduction The Holocene, the most recent geological epoch, experienced detectable climate change. Generally, the Holocene climate evolution can be characterized by an early cool phase followed by substantial warming towards the well-known Holocene Thermal Maximum (HTM) and finally a long-term cooling that ended in the preindustrial era (Marcott et al., 2013; Renssen et al., 2009). The main long-term cooling primarily resulted from solar insolation variations due to changing astronomical parameters (Berger, 1988; Denton et al., 2010; Abe-Ouchi et al., 2013; Buizert et al., 2014). These parameters determine the incoming solar radiation at the top of atmosphere, and lead to latitudinal climate patterns (Berger & Loutre, 1991; Berger, 1978). Retreating ice sheets, including the Laurentide Ice Sheet (LIS) and Fennoscandian Ice Sheet (FIS), add spatial irregularities to this latitudinal pattern, resulting in heterogeneous spatial distributions of simulated temperatures. This spatial heterogeneity was characterized by relatively cool conditions in the early Holocene in some regions, while other areas were relatively warm, as revealed in palaeoclimate modelling studies (Renssen et al., 2009; Blaschek & Renssen, 2013; Zhang et al., 2016). However, the spatio-temporal details of climate during the early Holocene are still uncertain, and inter-model comparisons have been conducted to identify consistently simulated climate patterns among independent model results and to detect inconsistent features (Bothe et al., 2013; Eby et al., 2013; Bakker et al., 2014;). For example, Zhang et al. (accepted) have compared Holocene simulations performed with four different models (LOVECLIM, CCSM3, FAMOUS and HadCM3) and found good multi-model agreements over places directly influenced by strong ice-sheet cooling, such as in northern Canada, northwest Europe and Greenland. Yet, divergent early-Holocene temperatures across models have been identified in regions where the climate was indirectly affected by the ice sheets, such as Alaska and Siberia. Even though climate models are useful tools for linking proxy records and understanding the impact of forcings on climate, proxy data are required to validate climate models at an early development stage (Braconnot et al., 2012) and to evaluate the simulations when multiple models perform differently. Climate proxy records are relatively abundant for the Holocene (e.g. Marcott et al., 2013; Sundqvist et al., 2014). To investigate the general patterns of climate evolution, Marcott et al. (2013) for instance have stacked the proxy records over the latitude bands of 30–90°N and 30°S–30°N, and found that the highlatitude cooling trend is opposite to a warming trend in low latitudes during the last 11 kyr. Eldevik et al. (2014) also have compiled climate records on a regional scale to shed light
on the climate history of Norway and the Norwegian Sea. Recent progress in proxy-based reconstructions and newly established databases provides ground for a systematical spatiotemporal investigation of Holocene temperature evolutions. For instance, based on the Holocene database of Sundqvist et al. (2014), temperature changes in the north Atlantic region and Fennoscandia (Sejrup et al., 2016), Alaska (Kaufman et al., 2016), the Canadian Arctic and Greenland (Briner et al., 2016) have been recently examined. Although considerable improvements have been achieved in proxy-based reconstructions, proxy data still contain inherent uncertainties. Firstly, climate proxies archive a matrix of environmental variables rather than only a climate signal of interest, as they are influenced by confounding effects (Brooks & Birks, 2001; Birks et al., 2010, Velle et al., 2010). For instance, a summer temperature reconstruction derived from pollen can include a signal related to other variables, such as winter temperature, precipitation, or even non-climatic factors (Seppä et al., 2004; Birks et al., 2010; Li et al., 2015). Moreover, the interpretations of proxy results are primarily based on observed contemporary relationships, implying potential uncertainties in reconstructions as these relationships may change slightly over time (e.g. Jackson et al., 2009). In addition, many processes, such as sediment disturbance and contamination, affect the translation of climate signals to depositional proxy signals, some of which may bring uncertainty into the interpretation of proxy-based results. Consequently, comprehensive comparisons of proxy data with model simulations may shed light on a better climatic interpretation of proxy-based results. Combining proxy and model results provides opportunities to improve our understanding of climate mechanisms in addition to the interpretation of proxy results and evaluating models. Owing to recent progress in proxy-based reconstructions and model simulations, it is possible to conduct comprehensive model–data comparisons by identifying consistent features and analyzing discrepancies. Indeed, numerous model–data comparisons have been conducted. For instance, model results (in 30–90°N and globally) were recently compared with proxy-based reconstructions to investigate the contradiction of Holocene temperature trends between the reconstructed cooling and the simulated warming, although some of simulations did not include the FWF forcing (Liu et al., 2014). Another model– data comparison revealed that increasing CO2 precedes global warming during the last deglaciation (Shakun et al., 2012). These model–data comparisons, however, have only used one or two model results to compare with stacked reconstructions and primarily focused on large-scale climate change, such as over 30° latitude bands (i.e. 30–60°N, 60– 90°N). The Palaeoclimate Modeling Inter-comparison Project (PMIP) also has conducted several model–data comparisons of Holocene climate, but focused mainly on the midHolocene (e.g. Masson et al., 1999; Bonfils et al., 2004; Brewer et al., 2007; Zhang et al., 2010; Jiang et al., 2013). Therefore, comparisons between transient multi-model simulations and proxy-based datasets on a detailed sub-continental scale remain unexamined.
CHAPTER 4 In order to evaluate Holocene simulations and to improve our understanding of the transient early-Holocene climate, we compare the four Holocene climate simulations performed with the LOVECLIM, CCSM3, FAMOUS and HadCM3 models that have been discussed by Zhang et al., (submitted), with proxy-based reconstructions of terrestrial temperatures from the Northern Hemisphere high latitudes. In particular, the present study aims to: 1) evaluate model results by identifying consistencies and mismatches between the model results and proxy data over regions on a sub-continental scale; 2) analyze the uncertainty sources of simulations and of quantitative proxy records to illustrate what we can learn about validation of simulations and the interpretation of proxy results; and 3) identify the most probable temperature trends during the Holocene on a sub-continental scale with the aid of additional available evidence.
4.2 Methods 4.2.1 Data and analysis Proxy data were mainly derived from the Arctic Holocene database of Sundqvist et al. (2014). Sundqvist et al. (2014) collected as many published records as possible, with the selection criteria of: 1) latitude: sites north of 58 °N; 2) time-frame: proxy time series extending back at least to 6 ka; 3) temporal resolution: higher than 400 ±200 yr; and 4) dating frequency: interval in age models smaller than 3000 yr. We picked terrestrial records providing quantitative reconstructions of temperature and conducted a further selection based on the time-frame of the records. As we are interested in climate evolutions of the entire Holocene, the records shorter than 9.5 ka were excluded in order to obtain records that also cover the early Holocene. The exception of this further selection was north Canada where long records are limited by coverage of the LIS before the final melting at ~6.8 ka. In order to obtain a comparable record density in north Canada, all records in the database were collected despite some being shorter than 9.5 kyr. With these extended criteria, 8 pollen records were obtained from the database and used in our analysis, together with additional 4 pollen records from Kerwin et al. (2004). Only one chironomid record was available from north Canada that is not included in our dataset, as we aim to compile multiple records to obtain a regional reconstruction. All together, we selected 61 records from 54 sites that are unevenly distributed over the study area (Fig. 4.1). High data density is represented in Alaska and Fennoscandia, whereas high-latitude Siberia has a low density. Site information on these proxy records is available from the supplementary information (Table S4.1). The temperature reconstructions are mainly based on pollen and chironomid assemblages, since these proxy data have been quantitatively interpreted as representing summer temperature, which is our target climatic variable. This proxy-based climate reconstruction
conducted by the original authors of individual records involves three main steps: 1) establishing modern training sets; 2) constructing a numerical (transfer function) model
Figure 4.1 Map showing the locations of 61 proxy records (from 54 sites) used in the study and the domains of the five regions applied in the analysis of the model simulations.
based on the relationship between the climate and biological datasets; 3) applying the transfer function model on fossil stratigraphical records and evaluating the resulting reconstruction before finally obtaining the quantitative climate record (Juggins & Birks, 2012). The transfer-function-based reconstruction facilitates conversion of the past fossil assemblages to quantitative temperature, precipitation and other climate variables, assisting direct comparison with model results. These quantitative reconstructions also allow us to statistically estimate their performance and sample-specific uncertainty. For instance, empirical tests have shown that in pollen-based Holocene climate reconstructions the sample-specific uncertainty generally varies from 0.9 to 1.3 °C (Seppä & Bennett, 2003). This uncertainty of individual reconstructions, however, was not taken into account in the present study, because we compiled the individual records to regional reconstructions. In Greenland oxygen isotopic data from ice cores, used here as a paleo-thermometer, were collected and calibrated into temperatures based on the published relationship between δ18O and temperature (Cuffey et al., 1995). As suggested by Cuffey et al. (1995), the
CHAPTER 4 deglacial isotopic sensitivity value of 0.33 ‰/°C was applied before 8 ka, and 0.25 ‰/°C was used for the rest of the Holocene. In addition, borehole-based temperature measurements from the GRIP ice core (72.6°N, 37.6°W) in Greenland were also included as these measurements directly relate to the past temperature changes (Dahl-Jensen et al., 1998). We applied three steps to compile the original individual records into one single composite reconstruction for a given region. Firstly, the anomalies from the present day (average of the last 200 yr) were calculated for each individual record of our data collection. Secondly, a binning procedure at 500-yr interval was applied for each individual record to filter out the high frequency variability and to obtain a consistent temporal interval. We took the median of these anomalies within the same bin to represent the individual record for corresponding 500-yr intervals. Thirdly, according to the location, these binned individual records were grouped into the five regions, including Fennoscandia, Greenland, north Canada, Alaska and Siberia (Fig. 4.1), and the final reconstruction for each region was compiled from these temporally-equally distributed proxy data. For a given region, we used the median of the proxy data values within the same bin as the reconstruction with the range of variability, as indicated by their lower and upper quartiles of these values. For the sake of clarity, we use the term “reconstruction” together with a proxy name (e.g. the pollen-based reconstruction) to refer to the composite regional reconstructions, and the term “record” to refer to individual site-based datasets. 4.2.2 Forcings and simulations The orbital parameters determine the seasonal and latitudinal variation of incoming solar radiation at the top of the atmosphere. During the Holocene, summer (JJA) insolation at 65°N decreased by 30 W m-2, as shown in Figure 4.2 (Berger, 1978). According to ice core measurements of CO2, CH4 and N2O, greenhouse gases caused a total radiative forcing variability of 1W m-2 during the Holocene (Joos & Spahni, 2008; Schilt et al., 2010). The GHG forcing peaked at 10 ka before reaching a low value at 8 ka and then increased again toward the preindustrial level. Ice-sheet forcing includes the orography, spatial extent and meltwater flux (FWF), which were constrained by geological evidence. The presence of the LIS and FIS enhanced the surface albedo, which however declined over time as the thickness and extent of these ice sheets generally decreased before their final vanishing at around 6.8 and 10 ka (Peltier, 2004; Ganopolski et al., 2010). The total freshwater release during the Holocene was the equivalent of a 60 m sea level rise (Lambeck et al., 2014), with slightly varying estimations of temporal and spatial distribution (Licciardi et al., 1999; Carlson et al., 2007; Jennings et al., 2015). Although ocean sediment data (e.g. detrital carbonate, ice rafted detritus) and geochemical tracers (e.g. δ18O, 87Sr/86Sr, U/Ca) can provide some constraints on FWF routing (Carlson et al., 2007; Jennings et al., 2015), FWF forcing is still uncertain in terms of exactly spatial and temporal distribution of this
total freshwater release. This uncertainty in the spatial-temporal distributions of FWF stems mainly from the various FWF magnitudes suggested by different proxies (Carlson et al., 2007). Defining well-agreed geographical locations of FWF discharges is hindered by the sparse distribution of proxy records. We employed four Holocene simulations that were performed with different models, namely LOVECLIM, CCSM3, FAMOUS and HadCM3. As these simulations have been discussed in detail by Zhang et al. (submitted), we give here only a brief description of the models and the experimental setup. The LOVECLIM simulation was performed with the LOVECLIM model, which explicitly represents the components of the atmosphere, ocean and sea ice, and vegetation with intermediate complexity (Goosse et al., 2010a). Despite its intermediate complexity, the model simulates synoptic variability associated with weather patterns. The simulation is an 11.5 kyr long transient run, which was named OGIS_FWFv2 in Zhang et al. (2016). The simulation was initialized from an equilibrium experiment for 11.5 ka and run with annually-based transient ORB and GHG forcings. Additional icesheet configurations were prescribed at a time step of 250 yr, and associated FWF (the most plausible version2, Zhang et al., 2016) was applied at irregular time intervals (Fig. 4.2b). The CCSM3 simulation was conducted with the CCSM3 model, which is a coupled ocean–atmosphere–sea-ice–land surface general circulation model (GCM), with a T31 resolution in the atmospheric component (Collins et al., 2006; Yeager et al., 2006). The simulation was truncated from a transient simulation covered the whole period since the LGM, with transient ORB and GHG forcing (He, 2011). The ice-sheet configuration (derived from ICE-5G reconstruction and updated at every 250 yr) and freshwater fluxes were prescribed at the stepwise time intervals (Fig. 4.2b) discussed by He (2011). The HadCM3 simulation was carried out with the HadCM3 GCM, which consists of coupled components for the atmosphere-ocean-sea-ice system, and has a resolution of 2.5° × 3.75° × L19 (lat × lon × vertical layers) in the atmospheric component (Gordon et al., 2000; Pope et al., 2000). The simulation consists of a set of snap-shot experiments performed at every 1 kyr, and the last 30 yr of each 300-yr run was used for the analysis. Apart from the GHG and ORB radiative forcings, the topography and spatial extent of the ice sheets were updated in each snap-shot experiment (according to ICE-5G reconstruction), but no exclusive FWF was applied into the oceans. The high spatial resolution of the HadCM3 model is one main consideration of including these experiments. The FAMOUS simulation is the Holocene part of a 22 kyr-long simulation performed with the FAMOUS model. The FAMOUS is a low-resolution version (approximately half) of HadCM3 with almost identical parameterizations of physical and dynamical processes to those of HadCM3, and can run around ten times quicker (Smith et al., 2008). This simulation involved forcings including GHG, ORB and prescribed 3-dimension ice sheets (derived from ICE-5G reconstruction and updated at every 1 kyr together with associated weak FWF (Fig. 4.2b), and the Bering Strait was opened at around 9 ka. The extents and topography of ice sheets
CHAPTER 4 Table 4.1 Summary of involved climate models and simulations Model
Components Ocean, Sea ice, Atm, Veg Resolution of atmospheric 5.6°×5.6°×L3 component(lat*lon) Dynamic-thermodynamic Sea ice model (CLIO) Resolution of oceanic 3°×3°×L20 component ORB Berger 1978 Prescribed Loulergue et al. 2008; GHG Schilt et al. 2010 forcing FWF Icesheet, FWF Equilibrium expt. Initial condition at 11.5 ka (1.2 kyr) Length of expt. 11.5 kyr
Ocean, Sea ice, Atm, Veg
Ocean, Sea ice, Atm, Veg
Ocean, Sea ice, Atm
Berger & Loutre (1991) Spahni et al. 2005; Loulergue et al. 2008 Icesheet Pre-industrial Snapshot Multiple snapshots
Berger & Loutre (1991) Spahni et al. 2005; Loulergue et al. 2008 Icesheet, FWF Transient expt. of 21 kyr 21 kyr
in the CCSM3, FAMOUS and HadCM3 simulation were based on the ICE-5G reconstructions (Peltier, 2004), which are comparable with the ice sheet configurations in LOVECLIM, except for slightly larger ice-sheet extents than in LOVECLIM (based on existing moraine dating results). According to these data, the retreating ice sheets were updated every 250 yr in LOVECLIM and CCSM3 and every 1 kyr in FAMOUS and HadCM3. Brief information on the involved climate models and simulations is summarized in Table 4.1. The domains of the five selected regions, including Fennoscandia, Greenland, Canada, Alaska and high-latitude Siberia, are indicated in Figure 4.1. We obtained the simulated areal average summer (JJA) temperatures for these regions to compare with proxy data that reflect summer or July conditions, with the exception of Greenland where we obtained the annual mean temperature to match the annual temperatures from δ18O and borehole ice core data. We applied a running mean of 500 yr to these obtained temperatures to filter out high-frequency variability of simulated temperatures. The results are presented as anomalies from the preindustrial era (i.e. negative value means cooler and positive represents warmer than the preindustrial). To obtain the overall temperature trend throughout the Holocene, we calculated the ensemble mean by averaging all transient simulations, as in other paleoclimate simulation studies (e.g. Lunt et al. (2013) & Bakker et al. (2014)). The HadCM3 results are based on snapshots and therefore shown separately. The name of the climate models hereinafter will be used equivalently to the corresponding simulation to remove redundancy.
Figure 4.2 Climate forcings in the simulations. (A) GHG forcings and summer (JJA) insolation at 65°N (both in W m-2) during the Holocene. (B) Change of ice sheet areas (km2) and FWF release into the oceans (mSv = Sverdrup×10-3 = 1×103 ms-1) during the early Holocene.
4.3 Results and discussion Given the spatially heterogeneous climate during the early Holocene, we compare the simulated temperatures with the proxy-based reconstructions on a sub-continental scale in the NH extratropics. Correspondingly, the following sections will firstly discuss separately the results for Fennoscandia, Greenland, Canada, Alaska and high-latitude Siberia. Together with the model–data comparisons, we investigate the uncertainty sources from both the simulation and proxy-based reconstruction perspectives. Additional evidence of climate change that is independent of δ18O, pollen- and chironomid data (e.g. glacier frequency and peatland initiation data) is used when available to further demonstrate how the climate most likely evolved in a given region. The closing section summarizes the overall Holocene climate history of these regions and discusses the implications of these model–data comparisons. 4.3.1 Temperatures in Fennoscandia Model–data comparisons The ensemble mean summer temperature of the simulations is consistent with the pollenbased reconstruction in Fennoscandia. Both simulations and proxy data suggest that the summer temperature rises from -2 °C at the onset of the Holocene to 1 °C by 8 ka, after which the temperature gradually decreases toward the preindustrial value (Fig. 4.3). The spread of the composite pollen-based reconstruction, statistically represented by its upper and lower quartiles of 13 records, stays at about ±0.5 °C during most of the Holocene. Multi-model differences are large, up to 2 °C before 8 ka, and are mainly caused by lower values in FAMOUS and HadCM3. Compared to this good agreement with pollen data, less consistency is found in the comparisons of chironomid-based reconstruction and simulated temperatures. In particular, the chironomid-based data indicate a relatively stable climate during the Holocene, with about 1 °C decrease in temperature throughout the Holocene. The range of variability in chironomid data, however, is up to 2 °C, which is larger than in the pollen-based reconstruction. In general, both composite pollen-based temperature reconstruction and simulations reveal an early-Holocene warming trend until 8 ka, differing from almost stable temperatures in the chironomid data. Uncertainty sources From the simulation-perspective, paleo-topographical changes related to the melting of ice sheets during the early Holocene are critical issues that influence the simulated temperature, as temperatures will go down by 0.65 °C with every 100 m rise in altitude according to the environmental lapse rate. At the onset of the Holocene, the ice sheet enhanced the surface elevation by more than 200 m over the center of the FIS that existed until ~10 ka (Peltier, 2004; Ganopolski et al., 2010; Cuzzone et al., 2016). However,
isostatically depressed ground rose during the Holocene with the ice-sheet load being removed, which adjusted the topography in an opposite direction to what the thick ice sheet did. In response to the FIS thickness of about 2.5 km during the LGM (Ehlers, 1990), the maximum of uplift was about 250 m since the last deglaciation (Vorren et al., 2008), which is comparable with the elevation effects of the LIS during the early Holocene. Consequently, the net paleotopography effect due to icesheet thickness and post-glacial rebound is relatively small since these two processes partially balance each other out. Considering that corrections on this small change would bring extra uncertainty, we therefore have not applied such correction in the simulated temperatures. Figure
4.3 Comparisons of simulated summer (JJA) temperatures with pollen- and chironomid-based temperature reconstructions in Fennoscandia. The grey and light blue shading indicate the range between lower and upper quartiles of 13 pollen and 11 chironomid records, respectively.
The post-glacial rebound might also influence the reconstructed temperatures. A warmer bias would be induced when the reconstructed temperature is strictly defined as the temperature at the same elevation without considering relative sea level changes. During the early Holocene, the maximum warm bias in Scandinavia is estimated to be up to 1 °C (Mauri et al., 2015). Global sea level, however, rose by almost 60 m during the Holocene (Lambeck et al., 2014), which partially compensated for the influence of this post-glacial rebound. In addition, the amplitudes of the rebound at proxy sites that are typically located near the margin of the ice sheet have been smaller than the estimated regional average. Therefore, considering the relatively small effects of post-glacial rebound on reconstructions and the potential uncertainty resulting from a correction, we applied no correction in the reconstructed Fennoscandian temperatures, despite our awareness of this effect. 97
CHAPTER 4 Additional evidence of climate evolution
The glacier record, used as a geophysical proxy of climate, also suggests an earlyHolocene warming trend (Nesje et al., 2009). Glacier growth and retreat are a response to changes in ambient environment (e.g. summer temperature and snow
Figure 4.4 (A) A frequency-distribution histogram of glacier-size variations in Fennoscandia (based on Nesje et al., 2009). (B) Composited geochemical proxy (average of records from Lake Hvítárvatn and Haukadalsvatn) in Iceland (derived from Geirsdottir et al., 2013). (C) Frequency of peatland initiation in Alaska (based on Jones & Yu, 2010) and (D) simulated soil moisture in LOVECLIM.
accumulation in winter), and thus the temporal glacier variations can reflect climate history (Nesje, 2005; Nesje et al., 2009). The extensive glacier during the early Holocene (Fig. 4.4a) primarily illustrate relatively cool summer temperatures that were followed by a distinct warm peak at 7~6 ka (Nesje et al., 2009). Therefore, multiple pieces of evidence indicate that the Fennoscandian climate was cool at 11.5 ka followed by warming trend until around 6 ka, implying that the relatively stable temperature suggested by chironomid data is probably arguable. Although it is well agreed that chironomid-based temperature records can provide reliable reconstructions on climatic variability during the Late-glacial period (Brooks et al., 2012; Heiri et al., 2014), there has been a discussion on how to interpret earlyHolocene chironomid assemblage data obtained from Fennoscandia (Velle et al., 2010; Brooks et al., 2012; Velle et al., 2012). One argument mentioned in this discussion is the influence of non-climatic processes following the last deglaciation 98
on chironomid data. These non-climatic factors included, for example, the nutrient availability, trophic state, and dissolved and total organic carbon in the lake, which may have changed following deglaciation process in the lake catchment. Apart from temperature, these non-climatic factors also influenced chironomid distribution and abundance, and thus potentially biased the assumed relationship between chironomids and temperature (Brooks & Birks, 2001; Velle et al., 2010). 4.3.2 Temperatures in Greenland Model–data comparisons Simulated annual temperatures and δ18O-based climate data in Greenland are generally consistent for Holocene trends (Fig. 4.5). One exception is the slightly stronger magnitude of the early-Holocene warming in δ18O data, leading to a higher temperature peak than in the simulations. In general, the temperature at 11.5 ka is 6 °C lower than 0 ka, which is followed by a warming, reaching a 2 °C warmer condition at ~7 ka. The spread in the δ18O-based temperatures is large (about 3 °C) in the early Holocene, but reduces to 2 °C by 7 ka and stays within 1 °C after 3 ka. The median values in the early Holocene are very close to the upper quartiles, and may thus seem implausible at the first sight. However, an inspection of a boxplot of the six temperature records (Fig. S4.1) reveals that the median of reconstructed temperatures is indeed similar to the upper quartile before 10 ka because the reconstructions primarily fall into two groups and the distribution is dominated by the upper one. The largest inter-model difference, up to 4 °C, was found at 8.5–8 ka since the FAMOUS simulation gives a drop of temperature that contrasts with the continuous temperature rise in the LOVECLIM simulation. Figure 4.5 Comparisons of Greenland temperatures between simulations and δ18O-based reconstruction (based on Cuffey et al., 1995) and borehole measurements at GRIP (Dahl-Jensen et al., 1998). The grey shading represents the range between the lower and upper quartiles of 6 δ18O records. No binning procedure was applied to process the borehole data, as indicated by the symbol of the dotted line.
CHAPTER 4 The borehole temperature record suggests a higher temperature than the simulation, especially around the mid-Holocene. At the onset of the Holocene the measured GRIP borehole temperatures show a less negative anomaly (compared with 0 ka) and matches better with simulated temperatures in comparison with the δ18O data. From 10 to 9 ka the borehole data is closer to the simulations than the δ18O does. However, the positive anomaly of more than 2 °C during the mid-Holocene in the borehole data is higher than in the simulations, thus diminishing its agreement with the model results. When comparing these borehole measurements with individual simulations, a better consistency is found with the temperature of LOVECLIM than with others. Even though the present study mainly focuses on long term climate change, it is worth noticing that the borehole data also suggest a temperature contrast between the Medieval Climate Anomaly and the Little Ice Age, which is absent in the simulations, except for CCSM3. Uncertainty sources The conversion of δ18O measurements to paleo-temperature estimates was obtained through a simple relation called the isotopic paleo-thermometer (Cuffey et al., 1995). However, two sources of uncertainty are involved in establishing this paleo-thermometer. First, the true coefficients are unknown because many factors in addition to local environmental temperature affect the isotopic composition, such as changes in sea-surface composition (Fairbanks, 1989), atmospheric circulation (Charles et al., 1994), and the seasonality of precipitation (Fisher et al., 1983). Second, all these factors may vary with time (Cuffey et al., 1995). Therefore, the complex δ18O response and its uncertain sensitivity to temperature might explain some of this mismatch in early-Holocene climate. The borehole temperatures are down-core measurements of the GRIP ice core, and thus might include a site-specific signal of this individual record (Dahl-Jensen et al., 1998), which likely could explain its intense warmth in the mid-Holocene. This site-specific signal is also reflected by considerable differences between the GRIP and Dye-3 borehole temperatures, with the latter indicating a peak warming between 5–4 ka, which is absent in the GRIP record (Dahl-Jensen et al., 1998). From the simulation-perspective, the relatively low temperature in the simulation (compared to proxy data) may be due to the difficulty of simulating a correct climate over the Greenland ice sheet. One of the challenges in simulating temperatures over ice sheets is that the accuracy of climate simulation highly depends on the model resolution, as high resolution allows detailed representation of topography and precise description of thermodynamics (e.g., turbulence and convection) (Genthon et al., 1994; Ettema et al., 2009). Meanwhile, the Greenland region in simulations is simplified and represented as a rectangular box, which may bring some uncertainties, especially over places where a strong gradient can be expected, such as near the southeastern coastal area. In addition, if the models do not simulate a correct magnitude of the reduction in Atlantic Meridional
Overturning Circulation (AMOC) in the early Holocene, the Greenland climate is likely to be biased as well. The climate over southern Greenland is highly influenced by the AMOC strength through the heat transport from the south and associated sea-ice feedbacks (Bond et al., 1993; Barlow et al., 1997; Rahmstorf, 2002). A weak early-Holocene AMOC is evidenced by 231Pa/230Th measurements in sediment cores from the North Atlantic (McManus et al., 2004) and geochemical proxies from Iceland (Fig. 4.4b) (Geirsdóttir et al., 2013), which is roughly consistent with the slowdown of AMOC in the simulations (e.g. LOVECLIM discussed by Zhang et al. (2016)). However, it is not straightforward to accurately evaluate the magnitudes of AMOC weakening in simulations with spatiallyscattered proxy records, implying uncertainty in the absolute values of the AMOC reduction and hence in the estimations of early-Holocene warming in coastal Greenland and regions influenced by AMOC. 4.3.3 Temperatures in north Canada Model–data comparisons Simulations and proxy data show similar Holocene temperature trends in north Canada, with a cooler early Holocene in both pollen data and the ensemble mean of the simulations (Fig. 4.6). The ensemble mean indicates a 5 °C warming from 11.5 to 8 ka with a ~3 °C inter-model spread, and an even stronger warming is shown in HadCM3 (up to 8 °C). The pollen-based reconstruction is compiled from 12 pollen records with a varied number of available records (as shown in Fig. 4.6), which extends only back to 10 ka. This composite pollen-based reconstruction indicates a 1~2 °C warming until 7 ka with a 3 °C spread, despite the overall low number of records before 8 ka. From 7 ka onwards, simulations and pollen data are consistent and indicate a ~1 °C cooling trend toward 0 ka.
Figure 4.6 Comparisons of simulated temperatures with pollen-based reconstruction in north Canada. In total, 12 records are available and the temporal variations of number of record are illustrated by the green circles at the top, and the range between lower and upper quartiles of individual records are represented by the grey shading.
CHAPTER 4 Uncertainty sources Similar to Fennoscandia, simulated temperatures in north Canada might be influenced by paleo-topography changes due to the reducing thickness of the LIS and the post-glacial rebound. The maximum thickness of the LIS during the LGM was up to 3–4 km (Peltier, 2004). As a rough estimate, the ice load will produce an isostatic depression of one-third of the ice-sheet thickness, since the specific gravity of ice is one third of that of rock (Vorren et al., 2008). Accordingly, the rough estimation of this total isostatic rebound is about 1– 1.3 km, although the estimation of rebound is influenced by multiple factors, such as different ice-sheet loads, usage of linear or non-linear rheology, the delay between load release and uplift (Wu & Wang, 2008; der Wal et al., 2010). Therefore, the effects of ice thickness (e.g., about 1.6–2 km at 11.5 ka, Peltier, 2004) and post-glacial rebound are roughly assumed to compensate each other when the uncertainty of different estimations was taken into account. This simplified assumption is plausible to some extent, but uncertainty could be induced in simulated temperatures by simply leaving out corrections on the associated paleo-topography. From the proxy-perspective, no correction associated with paleo-topography is applied in reconstructed temperatures in north Canada, as the correction of relatively small effects would induce considerable uncertainty. Moreover, uncertainty due to the coarse time resolution and poor dating control might be induced in some records, as the expanded selection criteria were applied in selecting proxy records. These expanded selection criteria implied that all records, as long as being interpreted as a temperature proxy by the original author(s), were included without further considerations on resolution and dating interval, as explained in Section 4.2.1. Additionally, the low number of records together with large spread during the early Holocene adds a further uncertainty to the early part of reconstruction (before 7 ka), as a potential site-specific signal of the individual records might be induced. In particular, some of those records are located on different climatic background, but no clear pattern was found in further inspection on individual records. Lastly, disequilibrium vegetation dynamics during the transient early Holocene in N Canada may influence the accuracy of pollen-based Canadian temperature records. It has been suggested that the post-glacial migration of trees to deglaciated regions such as Canada may have been constrained by their limited speed dispersal and population growth rates (Ritchie 1986; Birks 1986; Webb 1986). Such a time lag in their migration history would prevent the pollen records from tracking rapid climate changes, especially in the early Holocene. Thus it is possible that the considerable temperature rise by 7 ka in N Canada, as indicated by the models, was not fully reflected in pollen-based climate reconstruction, leading to an underestimation of the early-Holocene warming in pollen data. Nevertheless, other evidence suggest a compatible climate signal as the above model–data results. For instance, a chironomid-based temperature record (extending back to around 7 ka) from the lake K2 in Arctic Quebec suggests a slightly cool climate till 6~5 ka (Fallu et
al., 2005), despite a potential influences of non-climatic process following the deglaciation, as discussed in the section 4.3.1. The low δ18O values of the ice core of from the Penny and Agassiz ice caps before 10 ka also reflect a relatively cool early-Holocene climate (Fisher et al., 1995, 1998), although the quantitative temperature reconstruction based on these δ18O measurements would be more complex than in Greenland. Overall, proxy data and simulations agree reasonably well in the general trend of Holocene temperature in northern Canada. 4.3.4 Temperatures in Alaska Model–data comparisons Discrepancies between simulated summer temperatures and proxy-based reconstructions are found in Alaska, particularly in comparisons with the pollen-based temperature reconstruction (Fig. 4.7). The ensemble mean of the simulations shows a 2 °C cooling trend during the Holocene with a 1–2 °C range across individual models, while pollen data suggest a 4 °C lower temperature at 11.5 ka, followed by a warming trend until 6 ka, with a spread of 2 °C. At the same time, the chironomid-based reconstruction shows almost stable Holocene values with a 2 °C spread by 9 ka. Overall, the signal of the composite pollenbased reconstruction regarding the sign of early-Holocene temperature is incompatible with the model results, which is also different from the chironomidbased reconstruction. Large differences in pollen-based and chironomid-based temperature reconstructions during the early Holocene reflect the complexity and large uncertainty of the earlyHolocene climate change in Alaska. Figure
4.7 Comparisons of simulated temperatures with pollenand chironomid-based reconstructions in Alaska. The grey and light blue shading indicates the range between lower and upper quartiles of 9 pollen and 5 chironomid records, respectively.
Uncertainty sources Apart from the influences of non-climatic factors discussed in section 4.3.3, the different
CHAPTER 4 responses of the pollen and chironomid proxies to seasonality might also contribute to the incompatible climate signals between these proxies-based reconstructions. The occurrence and abundance of chironomids in lakes are strongly determined by summer air and water temperatures, as the ice-cover in lake decouples their habitat from the direct influence of temperature in winter (Heiri et al., 2014). This implies that chironomid-based temperature estimations predominantly reflect summer temperature and are less influenced by winter temperature than many other biotic temperature indicators. Pollen data, however, could be also influenced by other factors rather than only summer temperature, such as precipitation, water availability and winter temperature. For example, it is known that the low cold tolerances of some plant species (e.g., Atlantic element, as suggested by Dahl 1998) limit their distribution and growth, which also depends on the latitudinal location of proxy record. Commonly, the records from high latitudes, especially within the Arctic are interpreted as summer temperature signal, as the ground is covered by the snow during the winter (e.g. Woodward, 1987). Therefore, the chironomid assemblage would be a more suitable proxy for reconstructing summer temperature in this case than pollen. Accordingly, the chironomid-based reconstruction in Alaska would be more reliable than pollen records given a relatively large seasonality with cold winters and warm summers in Alaska. However, this benefit of chironomids would be no longer existed when the climate conditions changed. In Fennoscandia and Canada, where the early-Holocene temperatures were low in both summer and winter (Zhang et al., submitted), the advantage of chironomid proxy related to strictly constraining of summer temperature is no longer obvious. A further investigation of this issue requires a species by species examination on detailed pollen diagram data, which is outside the scope of the present study. Additional evidence of climate evolution Additional evidence of Holocene climate history is the reconstructed frequency of peatland initiation, which primarily responds to high temperature, soil moisture, or large seasonality (Jones & Yu, 2010; Korhola et al., 2010). Correspondingly, the high frequency of peatland initiation during the early Holocene in Alaska (Fig. 4.4C) can result from either one of these factors. If the frequency value of peatland initiation is mainly determined by temperature (Kaufman et al., 2004), the simulated decreasing temperature in the course of Holocene would be consistent with the diminishing frequency of peatland initiation. If the peatland initiation process is predominantly controlled by the availability of soil moisture (Strack et al., 2009; Zona et al., 2009; Jones & Yu, 2010), the peatland data would agree reasonably well with the LOVECLIM simulation on the high soil moisture during the early Holocene (Figs. 4C & D). If the seasonality was the main contributor to the initiation of peatlands in Alaska (Jones & Yu, 2010; Kaufman et al., 2016), the LOVECLIM simulation might overestimate the LIS induced winter warmth, leading to reduced early-Holocene seasonality compared with the pre-industrial (Zhang et al., 2016). Therefore, the Holocene temperature trend in Alaska is still inconclusive despite of the peatland-initiation data, as
these data can reflect either temperature, soil moisture or seasonality. The inconclusive Holocene climate in Alaska has also been illustrated by the divergent winter temperatures among different simulations (Zhang et al., submitted). From proxy viewpoint, this inconclusive Alaskan temperature trend during the Holocene has been reflected by the multiple timings of the HTM in different studies. A recent study has suggested that the HTM occurred as late as 8–6 ka in Alaska (Kaufman et al., 2016), which is much later than the earlier finding (~11–9 ka) of Kaufman et al. (2004). 4.3.5 Temperatures in high-latitude Siberia Model–data comparisons The Holocene-temperature signals in the comparisons between simulations and proxybased reconstructions are overall noisy in high-latitude Siberia, despite the simulations matching better with chironomid results than with pollen data (Fig. 4.8). The ensemble mean of simulations shows a 2.5 °C cooling trend with a range of more than 4 °C in CCSM3 to 0.5 °C in LOVECLIM. Two of three chironomid records show 2–3 °C Holocene cooling, while Temje record suggests no clear trend but flucated Holocene temperature. This lead to the large spread in the chironomid records with a cooling Holocene trend of 1 °C in the median of the dataset suggests, which is slightly smaller than in the ensemble mean of the simulations. Pollen data suggest an increasing temperature until 7 ka, reaching a period with distinct warmth of 2 °C above the preindustrial level that lasts until 4 ka, after which the temperatures decreases to the 0 ka level. Nevertheless, both simulations and proxy data suggest that Holocene temperatures in high-latitude Siberia were overall higher than at 0 ka. Figure 4.8 Comparisons of simulated temperatures with pollen- and chironomid-based reconstructions in high-latitude Siberia. The dots indicate median of 2 pollen and 3 chironomid records with different line styles represent individual records identified with ID numbers.
CHAPTER 4 Uncertainty sources In high-latitude Siberia, only two pollen and three chironomid records were used in the reconstructions due to limited data availability (Sundqvist et al., 2014). This low number of available records may lead to a decreased reliability in the Siberian reconstruction, as these limited records may induce some site-specific signals in the complied regional temperature reconstruction. Although four sites are under climatically different environments (e.g. continentally, climate type), no obviously different was found. From the simulationviewpoint, more than 4 °C early-Holocene warmth in CCSM3 is far higher than suggested by pollen and chironomid-based temperature reconstructions, implying that the warmth in CCSM3 might be overestimated. The overestimated early-Holocene warmth in CCSM3 is caused by its substantially positive albedo anomaly resulting from an overestimated snow cover at 0 ka (Zhang et al. submitted), which is ultimately due to the newly adopted formulation of the turbulence coefficient (Oleson et al., 2003; Collins et al., 2006). In addition, biomarker records, such as the records from Kyutyunda and Billyakh along the Lena River in Siberia, suggest a slightly increasing temperature during the early Holocene that is followed by the relatively warm mid-Holocene (Biskaborn et al., 2016). Overall, a higher Holocene temperature than at the preindustrial is suggested by simulations and available proxy data, although the temperature signal is weak because of the large multimodel spread and the very limited number of records. 4.3.6 Summarized discussion on model–data comparisons Implications of model–data comparisons on the Holocene climate history Holocene climate shows regional differences due to different mechanisms over various regions. Based on the current studies, insolation variations on the orbital timescale and the presence of the LIS and FIS during the early Holocene are the main drivers of Holocene climate change (Berger, 1978; Renssen et al., 2009; Denton et al., 2010; Blaschek & Renssen, 2013; Zhang et al., 2016). Our earlier simulation study (Zhang et al., 2016) has revealed that the cooling effects of the ice sheets overwhelms the higher summer insolation in some regions, such as N Canada and Fennoscandia, which leads to low temperature during the early Holocene. This cool climate was followed by an early-Holocene warming with shrinking of the FIS and LIS. Meanwhile, the retreating ice sheets caused a freshwater release to the oceans, weakening the AMOC, which together with the anomalous atmospheric circulation influenced the climate beyond the ice-sheet boundaries, such as in Alaska, Siberia. To further investigate these effects and evaluate the simulations, an intermodel comparison was conducted by Zhang et al. (submitted), in which they have found that multi-model consistencies varied spatially. A relatively consistent climate was suggested by multiple simulations over regions where the climate was directly influenced by the ice sheets, such as north Canada and Fennoscandia. However, in other regions, such as Alaska, Arctic and E Siberia, the multiple models indicate a divergent climate, implying
that model-dependences exist (Zhang et al., submitted). By systematically comparing with the proxy-based temperature reconstructions, this study further confirmed the Holocene climate trends in the regions of Fennoscandia, Greenland and north Canada discussed by Zhang et al. (2016). Since the model simulations and proxy data are independent methods to investigate the climate history, we consider the Holocene climate in the above three regions to be relatively well established. However, the Holocene climate evolutions in Alaska and Siberia are, unfortunately, still inconclusive. The large uncertainty of proxy reconstructions hinders us to draw conclusions on the climate of these two regions. In Alaska, the simulations, pollen- and chironomid-based reconstructions all show different temperature trends. In Siberia, the pollen- and chironomid-based reconstructions are also different, which together with the low number of records impair the reliability of reconstructions. Implications of model–data comparisons for the interpretation of proxy results Proxy data can provide evidence of past climates, but multiple factors potentially induce an uncertainty in the interpretation of proxy results, thus impairing the quality of the proxy reconstructions. Combining the proxy data with simulations could shed light on the more plausible interpretation of proxy data. The above model–data comparisons suggest that the uncertainty sources of proxy reconstructions may vary over different regions, implying that different factors should be taken into consideration when compiling the proxy-based results. In newly deglaciated regions, the influence of non-climatic limnological factors preceded by deglaciation process probably impair the accuracy of proxy-based reconstructions. In Fennoscandia, the influence of non-climatic factors following the last deglaciation potentially contributes to a weak early-Holocene warming in chironomid data. Over the areas with δ18O data of ice cores, such as Greenland, converting measured δ18O to temperature probably induces some degree of uncertainty to δ18O-based temperature reconstruction, as the relationship between the δ18O and temperature was simplified in this conversion (Cuffey et al., 1995). In the regions experiencing relatively fast post-glacial temperature changes, the pollen data potentially underestimate this changes due to disequilibrium vegetation dynamics with local climate conditions. In Canada, for instance, the early-Holocene temperature rise was probably underestimated due to the vegetation establishing following the last deglaciation. Meanwhile, the uncertainty of composite pollen-based reconstructions in Canada may also partially result from the uncertainty of individual records due to extended selection criteria and a low number of proxy records in the earlier part of the reconstruction (before 7 ka). In Alaska, on the other hand, the pollenbased reconstruction may be biased by the winter temperature to some degree due to the large seasonality. Finally, in high-latitude Siberia, the low number of records is the main obstacle to make conclusions on Siberian climate history during the Holocene. From a methodological point of view, our model-data comparisons also suggest
CHAPTER 4 uncertainties associated with the choice of quantitative reconstruction method. The weighted averaging partial least square regression and calibration (WAPLS) and modern analog technique (MAT) are the commonly used methods in the proxy-based temperature reconstruction (Table S4.1). MAT is based on a direct space-for-time substitution by assuming that similar biological assemblages are deposited under similar environmental conditions (Birks et al., 2010). Thus, the similarity, in terms of species composition and abundances, of the fossil pollen and the modern training-set is firstly measured to find the sample(s) of the highest similarity from a training-set and their climate condition is assigned to the fossil sample. WAPLS, which combines weighted-averaging regression (WA) and partial least squares (PLS), has the attractive features of these both methods, such as the ability to model unimodal response (of WA) and efficient of components (of PLS) (Juggins & Birks, 2012). MAT and WAPLS have their own merits and weaknesses. MAT is a direct and efficient method to reconstruct the environmental condition through identifying analogous modern samples, but it could yield unreliable reconstructions when good analogues do not exist (Juggins & Birks, 2012). WAPLS has been shown to be robust for different response models and it enables extrapolation, but its efficiency is impacted by its assumption of a unimodal species-environment response model (Juggins & Birks, 2012). Moreover, the edge effect is one inherently limitation of WAPLS since that results in distortions at the ends of the environmental gradient (ter Braak and Juggins, 1993). Therefore, one method can outperform another for particular datasets, depending on differences in training-set size, taxonomic diversity, and complexity of the speciesenvironment relationship. Telford & Birks (2005; 2009) have compared the performance of a range of methods and found that MAT is particularly sensitive to spatial autocorrelation. In some of the pollen-based temperature records from Alaska and Canada, the MAT method rather than WAPLS was applied to build the relationship between the climate and biological data-sets (Table. S4.1), which might explain some of the above model–data discrepancies as well. Nevertheless, it is too complex and also out of the scope of the present study to draw conclusions on whether MAT gives higher or lower temperature than the WAPLS does.
4.4 Conclusions In this study, we compared four simulations with composite proxy reconstructions over Fennoscandia, Greenland, north Canada, Alaska and high-latitudes Siberia. Related uncertainty sources were also examined and additional evidence was employed to identify the most possible Holocene temperature trends. The main findings are outlined below: Simulated and reconstructed temperatures in Fennoscandia, Greenland and north Canada are generally consistent. During the last 11.5 ka, overall temperature evolution patterns are the early-Holocene warming until the mid-Holocene warmth at around 8–6 ka and subsequent decrease to 0 ka. Within this generally consistent frame, the degrees of
consistency, however, vary among these regions due to various sources of uncertainty. In Fennoscandia, pollen data and the ensemble mean of simulations consistently indicate ~2 °C early-Holocene warming, while this warming is small in the composite chironomidbased reconstruction. Glacier frequency data from Fennoscandia show a high value during the early Holocene, and this value decreases until 7 ka, implying an early-Holocene warming trend that is probably underestimated in chironomid data. The influences of nonclimatic factors following the last deglaciation provide potential explanation of this underestimation. In north Canada, the early-Holocene warming is slightly stronger in simulations than in pollen data, despite a potentially large uncertainty in the composite proxy-based reconstruction induced by the low number of long proxy records. In addition, paleo-topography changes due to the thick ice sheets and post-glacial rebound following the last deglaciation also exert uncertainty in model–data comparisons in Fennoscandia and north Canada. In Greenland, minor differences between the simulations and proxy records can either result from underestimation of the warm peak in the model results or from overestimation of δ18O-based temperatures due to simplified calibration of δ18O to temperatures. The ensemble mean of simulations mismatches with the proxy-based reconstructions in Alaska. Pollen-based reconstruction suggests a 4 °C cooler early Holocene with ±1 °C spread, while the ensemble mean of simulations indicates a constant cooling throughout the Holocene with a 1–2 °C range across individual models. Therefore, the signal in the pollen-based temperature reconstruction is incompatible with the model results, which is also different from the signal recorded by chironomid data. Basal peat dates depict highfrequencies of peatland initiation in Alaska during the early Holocene, but these frequent initiations can reflect either a high summer temperature, increased soil moisture or large seasonality. Consequently, the temperature evolution in Alaska remains inconclusive despite additional available peatland initiation data. In Siberia, temperatures during the early-to-mid Holocene were probably higher than during the preindustrial, but the signals of specific temperature trends are noisy. In particular, the spread of the simulations is wideranging, with the constant cooling in CCSM3 and FAMOUS contrasting with the result of LOVECLIM that shows a much smaller temperature decrease following a minor increase. The pollen- and chironomid-based reconstructions are compiled from a low number of available proxy records and are divergent as well. These comparisons of multi-model simulations with proxy reconstructions further confirm the Holocene climate evolution patterns in Fennoscandia, Greenland and North Canada. This implies that the mechanisms behind these changes (Zhang et al., 2016) would be plausible, and that the multiple simulations provide a reasonable representation of Holocene climate (Zhang et al., submitted). However, the Holocene climate history and their underlying mechanisms in the regions of Siberia and Alaska remain inconclusive. Thus, more work is still needed to pin down the uncertainties in climate estimations. For
CHAPTER 4 instance, more records from Siberia with a broad spatial coverage are highly demanded. Further examinations on the quantitative contribution of these interpreted variables would be helpful to identify the climate signal, which can also improve our understanding of Holocene climate variability and underling mechanisms. Meanwhile, with detailed regional models, conducting sensitivity test on different aspects regarding the climatic interpretations of Alaskan peatland would also provide evidence to narrow down this inconclusive Holocene temperature.
Chapter 5 The role of climate, forest fires and human population size in Holocene vegetation dynamics in Fennoscandia ________________________________________________________________________ Based on: Kuosmanen, N., Marquer, L., Tallavaara, M., Molinari, C., Zhang,Y., Alenius, T., Trondman, A-K., Edinborough, K., Pesonen, P., Reitalu, T., Renssen, H., and Seppä, H. The role of climate, forest fires and human population size in Holocene vegetation dynamics in Fennoscandia. In preparation for Vegetation Science. Contribution of Zhang Y: The LOVECLIM simulation employed in this chapter was performed and analysed by Zhang Y, who also contribute to the interpretation and discussion of these results.
Abstract Questions Together with natural drivers, increasing human population size and anthropogenic activity influence the broad-scale composition of the boreal forest. Here we investigate the changing role of climate, forest fires and human population size in Holocene vegetation dynamics. We aim to quantify the relative importance of these three different drivers on variation in vegetation composition before and after the onset of farming in Sweden (at 6000 cal yr BP) and in Finland (at 4000 cal yr BP). Location Four regions in Fennoscandia: southern Sweden, central Sweden, southwestern Finland and southeastern Finland. Methods Holocene regional plant abundance was reconstructed by applying the REVEALS model to fossil pollen records from 33 lakes. REVEALS estimates of regional plant abundances were grouped into plant functional types (PFTs) to assess the Holocene vegetation dynamics in the light of functional changes in vegetation. Variation partitioning was applied to assess the relative importance of climate, forest fires and human population size on changes in PFTs. Climate variable include mean winter (DJF) and summer (JJA) temperatures that were derived from the LOVECLIM climate model. Fire variable was reconstructed based on sedimentary charcoal records from 19 lakes. Estimated trend in
CHAPTER 5 human population size is based on the temporal distribution of archaeological radiocarbon dates. Results Climate explains the highest variation in vegetation dynamics when the whole study period in Sweden (4000–10 000 cal yr BP) and in Finland (1000–10 000 cal yr BP) is considered, and during the pre-agricultural period in each region. In general, fires explain low proportions of the variation, although fires significantly affect vegetation changes in south Sweden and in Finland (both southeast and southwest). Human population size has significant effect on vegetation dynamics after the onset of farming and explains the highest variation in vegetation in south Sweden and southwestern Finland. Main conclusions Mesolithic hunter-gatherer populations did not significantly affect vegetation dynamics in Fennoscandia and climate was the main driver of the change at that time. Agricultural populations, however, had greater effect on vegetation dynamics and the role of human population size became more important during the late Holocene. There is a clear regional difference in the importance of human population size, i.e. human impact on vegetation dynamics was notably higher in southern Sweden and southwestern Finland (where land use was more intensive) than in central Sweden and southeastern Finland. Our results demonstrates that climate can be considered as the main driver of long-term vegetation dynamics, but in some regions in Fennoscandia the influence of human population size in vegetation dynamics may exceed that of climate and has longevity dating to early Neolithic.
5.1 Introduction Several reconstructions of Holocene vegetation dynamics based on pollen records have demonstrated the influence of anthropogenic activity on forest composition and landscape openness from the mid-Holocene in central and northern Europe (Behre 1988; Holst 2010; Nielsen et al. 2012; Marquer et al. 2014; Davis et al. 2015; Fyfe et al. 2015). In Fennoscandia, most probably due to its more remote and northern location, vegetation remained under natural drivers longer than in other regions in Europe. Pollen records derived from Fennoscandia demonstrates regional differences in the forest composition throughout the Holocene. Although, the variation in long-term vegetation dynamics has been generally connected to the changes in climate (e.g. Birks 1986b; Heikkilä & Seppä 2003; Soja et al. 2007; Miller et al. 2008) archaeological and palynological evidence suggests that already Mesolithic hunter-gatherers may have impacted on local scale vegetation dynamics through burnings and favoring food-plants in Fennoscandia (Regnell et al. 1995; Hörnberg et al. 2005). Presently, the intensive anthropogenic influence and the fact that boreal biome is predicted to experience accelerated warming (Christensen et al. 2013) places vegetation under pressure and susceptible for changes. Even though climate is considered as the main natural driver behind long-term vegetation
changes, biotic factors such as competition by other species and insect outbreaks (Mitchell et al. 2006; Tudoran et al. 2016) and abiotic factors such as soil properties, light availability, natural disturbances such as windstorms and forest fires (Selikhovin 2005; Miller et al. 2008; Kuneš et al. 2011; Post 2013) can regulate the effect of climate on the regional long-term vegetation dynamics (Lindner et al. 2010; Kröel-Dulay et al. 2015; Kuosmanen et al. 2016b). In boreal forests, fire is considered an important natural disturbance factor that profoundly affects forest age and structure, species composition and succession dynamics (Kuuluvainen et al. 1998; Lehtonen & Kolström 2000; Granström 2001; Bradshaw et al. 2010; Kuosmanen et al. 2014). In addition to natural drivers, the intensified anthropogenic impact has affected the natural vegetation dynamics. Therefore, understanding the complex interactions between natural- and human-induced changes in the past regional vegetation dynamics, can shed light on the future effects of a changing climate on ecosystems that are heavily influenced by human activity. Evident effect of human impact on vegetation in Fennoscandia is connected to the onset of agriculture, which was often accompanied by increased forest fires (Huttunen 1980; Granström & Niklasson 2008). The archaeological, macrofossil and pollen evidence suggest that the change from hunter-gatherer practices to farming was gradual, spreading from south to north in Sweden and from southwest to east and north in Finland. The earliest signs of agriculture are found at ca. 6000 calibrated years before present (cal yr BP) in south Sweden (Sørensen & Karg 2012; Welinder 2011). In southwestern Finland first firm signs of cultivation are dated around 4000 cal yr BP (Vuorela & Lempiäinen 1988; Taavitsainen et al. 1998; Lahtinen et al. 2017), and slightly after 3200 cal yr BP in eastern Finland (Taavitsainen et al. 1998; Lavento 2001). Presently, management practices have altered the natural forest structure and disturbance dynamics in Fennoscandian boreal forests (Halme et al. 2013). The role of human impact and climate on vegetation changes during the mid- and lateHolocene in Northern Europe has been widely studied. However, the assessment of human impact has been mainly qualitative and based on interpretation of pollen data or archaeological findings. Recently, Reitalu et al. (2013), Kuosmanen et al. (2016a) and Marquer et al. (in review) have addressed this question quantitatively by means of variation partitioning. However, all these studies used different proxy for reconstructing the human impact. In Reitalu et al. (2013) indicators of human impact were derived from pollen records, nevertheless reconstructing both human impact and vegetation from pollen records may cause a problem of circularity. In Marquer et al. (in review) anthropogenic impact on vegetation was based on simulated data from the anthropogenic land cover change scenario (ALCC) (Kaplan et al. 2010). Following the approach of Lechterbeck et al. (2014) and Woodbridge et al. (2014), Kuosmanen et al. (2016) utilized human population size proxy, based on the temporal frequency distribution of radiocarbon dated archaeological findings, to assess the influence of human population size on long-term
CHAPTER 5 vegetation changes in Finland. This approach avoids the problem of circularity related to the indicators of human impact derived from pollen records. Such proxy is based on the assumption that the amount of archaeological material deriving from any particular time point is correlated with the human population size at that time point. Large populations produced more archaeological remains for archaeologists to find and date than smaller populations (e.g., Gamble et al. 2005; Shennan & Edinborough 2007; Surovell et al. 2009; Tallavaara et al. 2010; Shennan et al. 2013; Brown 2015; French and Collins 2015). Considering the climate variables, all of the three studies (Reitalu et al. 2013; Kuosmanen et al. 2016a; Marquer et al, in review) used simulated climate datasets, although from different models. Reitalu et al. (2013) and Marquer et al. (in review) used regional vegetation abundance reconstructed by the REVEALS model, which provides more realistic vegetation estimates than the ones based on untransformed pollen data (Sugita 2007). Kuosmanen et al. (2016) did not use any pollen-based modelling approaches and – due to the difference in spatial representativeness of pollen and human population size data – their results suggesting relatively low importance of human population size on vegetation change can be biased. Here we employ larger and spatially more comparable dataset of response and explanatory variables to investigate the changing role of climate, forest fires and human population size in Holocene vegetation dynamics in Fennoscandia. We statistically assess the relative importance of the three environmental factors on Holocene vegetation dynamics 1) during the whole study periods from Sweden (10 000–4000 cal yr BP) and Finland (10 000–1000 cal yr BP), 2) before the onset of farming in Sweden (at 10 000–6000 cal yr BP) and in Finland (at 10 000–4000 cal yr BP), and 3) after the onset of farming in Sweden (at 6000– 4000 cal yr BP) and in Finland (at 4000–1000 cal yr BP). The length of the study periods between Sweden and Finland differ due to the limitations in human population size data. Pollen based REVEALS corrections grouped into plant functional types (PFTs) are used to reconstruct the regional vegetation dynamics in order to assess the functional changes in Holocene vegetation dynamics. Climate variables are simulated with the latest version of the LOVECLIM climate model (Zhang et al. 2016) and changes in regional fire regime are reconstructed based on sedimentary charcoal records. Human impact is assessed based on independent proxy of human population size derived from radiocarbon dated archaeological findings from Fennoscandia (Oinonen et al. 2010; Tallavaara 2010; Manning et al. 2016). Compared to the previous studies (Reitalu et al. 2013; Kuosmanen et al. 2016a; Marquer et al. in review), the use of these larger and spatially more comparable datasets will provide more accurate insights into the Holocene vegetation dynamics and its potential drivers in Fennoscandia.
5.2 Materials and methods 5.2.1 Study area The study area is located in Fennoscandia and is divided into four regions: 1) South (S) Sweden, 2) Central (C) Sweden, 3) Southwest (SW) Finland, and 4) Southeast (SE) Finland (Fig 5.1). The northern limit of the study region in Sweden is located at 61.5°N, while the border between S and C Sweden is at 58.3°N. In Finland, the northern limit of the study region is at 62.5°N, and the border between SW and SE Finland is at 25.5°E (along Lake Päijänne). These study regions were chosen due to the high number of data for pollen-based REVEALS estimates, human population size data and charcoal data were available in these areas. C Sweden, SW and SE Finland are characterized by boreal coniferous and deciduous tree species, with Norway spruce (Picea abies), Scots pine (Pinus sylvestris), silver birch (Betula pendula) and downy birch (Betula pubescens) as the dominant species, and aspen (Populus tremula), grey alder (Alnus incana) and black alder (Alnus glutinosa) as other common tree species. S Sweden is located in the transition zone between the boreal and temperate vegetation and, in addition to boreal taxa, forests includes temperate tree species such as Oak (Quercus robur), ash (Fraxinus excelsior), linden (Tilia cordata), maple (Acer planatoides) and hazel (Corylus avellana).
Figure 5.1 Location of the study area. The four regions are shown, as well as the pollen and
CHAPTER 5 charcoal records that have been used in the present study.
5.2.2 Regional plant abundances and plant functional types (PFTs) Pollen data and the REVEALS model (Sugita 2007) were used to reconstruct the regional plant abundance in the study regions. The REVEALS model reduces biases in pollen analysis (when expressed as percentages) caused by inter-taxonomic differences in pollen productivity, dispersal and depositional characteristics, as well as basin size. The REVEALS model has been largely used within the LANDCLIM project (Gaillard et al. 2010) for Europe north of the Alps (Nielsen et al. 2012; Fyfe et al. 2013; Marquer et al. 2014, in review; Trondman et al. 2015, 2016). We used 33 pollen records selected from the European Pollen Database (Fyfe et al. 2009; Giesecke et al. 2014), collected from publications or provided directly by data contributors (S5.1). The REVEALS model was run separately for each region using 5 pollen records for C Sweden, 17 for S Sweden, and 5 and 6 pollen records for SE and SW Finland, respectively (Fig. 5.1). We only used pollen records from lakes (both large and small sizes) because lake data, in particular from large lakes (>50 ha), are considered more appropriate than data from bogs. REVEALS data from multiple sites, small and large sizes, are generally reliable to reconstruct land cover over large areas, although they have larger error estimates than using one large lake alone (Trondman et al. 2016). An important criterion for the selection was to choose pollen records with robust chronological control and satisfactory Holocene time resolution. Table 5.1 Correspondence between the 25 pollen taxa and the 10 plant functional types (PFTs). TBE1 TBE2 IBE TSE IBS
Plant functional types Shade-tolerant evergreen trees Shade-tolerant evergreen trees Shade-intolerant evergreen trees Tall shrub evergreen trees Shade-intolerant summergreen trees
The methodological protocol for running REVEALS follows the LANDCLIM Project (Trondman et al. 2015). Means of REVEALS estimates (in percentage cover) of 25 plant taxa (pollen-type equivalent groups) and associated standard errors were calculated for all pollen records distributed within each region for consecutive time windows of 200 years.
The two most recent windows were 50 years. The study period covers the entire Holocene (i.e. 0–11700 cal. yr BP). Means of REVEALS estimates of 10 plant functional types (PFTs) after Trondmand et al. (2015) and associated standard errors were calculated (Table 5.1). PFTs were employed as response variable in order to focus on functional changes in vegetation rather than changes in community taxonomical composition. Compared to taxa specific REVEALS estimates PFTs provide larger scale trends in Holocene change in different forest and plant ecosystems and are potentially more comparable with the general climatic trends provided by climate simulations. 5.2.3 Explanatory variables of Holocene vegetation changes Climate The climate variables for the study regions were obtained from a simulation performed with the LOVECLIM climate model. A version of LOVECLIM was applied that explicitly represents components of atmosphere, ocean (including sea ice) and vegetation in the climate system with intermediate complexity (Goosse et al. 2010a). As presented in Zhang et al. (2016), the forcings included in the simulation were orbital forcing and concentration of greenhouse gases CO2, N2O and CH4 in the atmosphere. The prescribed ice-sheet forcing includes 3-dimensional configuration of the ice sheets and associated freshwater flux (FWF) with constraints of geological data. The vegetation module (VECODE) is forced by climate parameters (e.g. temperature and precipitation) and simulates a fractional distribution of deserts, forests, and grassland in each land grid cell, reflecting the main regularities of climatic and geographical spaces in vegetation distribution (Brovkin et al. 1997). The simulation used in the present study is forced by the version 2 of FWF forcing (FWF-v2, in Zhang et al. 2016), which is a more realistic scenario in terms of overall Holocene temperature trend. More detailed information on the simulation setup can be found in Zhang et al. (2016). The spatial resolution of one grid cell of the model is 5.6° x 5.6° and three grid cells were used to provide regional temperature data for the study regions. The S and C Sweden fell into the same latitudinal band in the simulation, and thus the climate data are the same for these regions. Climate variables include winter (December-January-February or DJF) and summer (June-July-August or JJA) temperature. Precipitation data was excluded due to large uncertainty in precipitation simulations. The temperature data are provided as a 30-yr average and expressed as anomalies from the preindustrial mean (550–250 cal yr BP). Regional fires The fire variable was reconstructed based on sedimentary charcoal records from 19 lakes. Charcoal data covering partly or fully the last 10 000 years were acquired from the latest version of the Global Charcoal Database (GCD v3, Marlon et al. 2016), from previously published syntheses (Molinari et al. 2013; Clear et al. 2014). To retrieve regional composites of changes in fire activity over the Holocene, charcoal accumulation in
CHAPTER 5 sediments was transformed and standardised according to a protocol previously described by Power et al. (2010). This procedure includes: 1) transforming non-influx values (e.g., concentration expressed as particles cm-3) to influx values (e.g., particles cm-2 yr-1) by dividing the concentration values by sediment deposition rates (yr cm-1), 2) rescaling the values using a minimax transformation to allow comparisons among sites, 3) homogenizing the within-records variance using the Box-Cox power transformation, and 4) rescaling values once more to Z-scores using a base period most representative of the entire database (in this study the interval 700–6000 cal yr BP), so that all records have a common mean and variance. The most important step of the transformation is the homogenization of the variance, which serves to make small-scale variations visible while also reducing the importance of high-value outliers. Transformed charcoal records of each sub-region were then re-sampled by bootstrapping 1000 times with a moving window procedure using non-overlapping bins of 100-years. Re-sampled series were then aggregated and smoothed using a locally weighted scatterplot smoother with a window half width of 500 years. Regional charcoal composite series mean and 90% confidence intervals were calculated by averaging the smoothed and bootstrapped series (Daniau et al. 2012). Composite curves in this paper were produced following these methods as implemented in the R paleo-fire package (Blarquez et al. 2014). Human population size Changes in prehistoric human population size were reconstructed using the temporal frequency distribution of archaeological radiocarbon dates. The main part of the Swedish data was extracted from the EUROEVOL dataset that contains over 14 000 archaeological radiocarbon dates from the Neolithic in northwestern Europe, dating between the Late Mesolithic and Early Bronze Age (Shennan et al. 2013; Timpson et al. 2014; Manning et al. 2016). The Swedish samples were supplemented by Mesolithic dates (Edinborough 2005) and divided into southern (N=414) and central (N=352) subsets according to the borders previously defined. Due to the Neolithic focus of the EUROEVOL data, the Swedish samples were restricted to cover the period between 10 000 and 4000 cal yr BP. The Finnish dataset (Oinonen et al. 2010; Tallavaara et al. 2010) covers the period between 10 000 and 1000 cal yr BP both in southwestern (N=692) and southeastern (N=315) areas. Summed probability distributions of calibrated radiocarbon dates (SPDs) were created for each area, and these SPDs were evaluated for the biasing effects of random sampling and calibration effects using a Monte-Carlo simulation approach (Shennan et al. 2013; Timpson et al. 2014; Brown 2015; Crema et al. 2016). To reduce the potential biasing effect of oversampling within sites, multiple dates from individual sites were grouped into nonoverlapping bins, such that after ordering the dates at a site, the next date was only assigned to a new bin if there were more than 200 radiocarbon years between it and the previous date (Tallavaara et al. 2010; Shennan et al. 2013). Binning, calibration, summing
of calibrated dates, and Monte-Carlo tests with exponential null-models and 1000 simulation runs were performed for each geographical sample using the implementation of the approach by Crema et al. (2016) in statistical software R (R Development Core Team 2016). Monte-Carlo tests demonstrate that the main features in all four regions are not artefacts of calibration or random sampling (Table S5.3, Fig. 5.1). Finally, SPDs were binned to 200 yr bins, by averaging summed probability value within each bin to match the resolution of vegetation reconstructions and facilitate comparisons between the human population proxy and other variables. This binning further reduces radiocarbon date calibration-induced artificial high frequency variation in the distribution. 5.2.4 Statistical analysis Variation partitioning (e.g. Borcard et al. 1992) was performed to assess the relative importance of climate, forest fires and human population size on variation in the Holocene vegetation dynamics. In the analysis, plant functional types (PFTs), derived from the REVEALS reconstruction, were used as response matrix, while climate, forest fires and human population size were used as explanatory variables. For the statistical analyses, PFTs based on pollen data were Hellinger transformed (Legendre & Gallagher 2001) and all variables were binned into 200 year bins. To assess the significance of each constraining variable on the PFT variation, an ANOVA permutation test was run with 999 randomizations. Variation partitioning analysis with climate, human population size and fire as explanatory variables was first performed for the whole study period in all regions, namely at 4000–10 000 cal yr BP in S and C Sweden, and at 1000–10 000 BP in SW and SE Finland. In order to assess the effect of the explanatory variables on vegetation before the onset of farming, variation partitioning was performed for the period of 6000–10 000 cal yr BP for C and S Sweden, and for the period of 6000–10 000 BP and 4000–10 000 BP for SW and SE Finland. To assess the relative importance of the explanatory variables on vegetation after the beginning of agriculture, variation partitioning was performed for the period of 4000–6000 cal yr BP for S and C Sweden, and for the period of 1000–4000 cal yr BP for SW and SE Finland. To separately evaluate the relative importance of summer and winter temperatures on vegetation, variation partitioning with summer and winter temperature as individual explanatory variables was performed for all three study periods, for the whole Holocene, the pre-agriculture and the agricultural periods, for all the study regions. In order to more closely assess the change in the relative importance of climate and human population size during the Holocene, a moving window approach (Reitalu et al. 2013) was employed. This approach enables performing variation partitioning for subsets of data in shorter time periods, hence providing information on how the relative roles of climate and human population size change over time. Moving window analysis was performed for all
CHAPTER 5 study regions in 2000 year time windows, with ten 200-year bins in each time window.
5.3 Results 5.3.1 Holocene change in PFTs and explanatory variables Sweden In general, changes in vegetation dynamics (PFTs) are similar in both study regions of Sweden. Picea, a shade-tolerant conifer (TBE1) migrated to Sweden only after 4000 cal yr BP and is not present during our study period. Values of Pinus, a shade-intolerant conifer (IBE) decrease since 9500 cal yr BP onwards in S Sweden, whereas in C Sweden its proportion stays constant through the study period (Fig. 5.2). In both regions, the proportion of shade-intolerant summer green trees (IBS, i.e. Alnus, Betula, Corylus, Fraxinus, Quercus) rises from 40% to 60% during the early Holocene and stays constant through the mid-Holocene, with some variation in S Sweden. The proportion of shadetolerant summer green trees (TBS, i.e. Tilia, Ulmus) increases from the early Holocene until 5700 cal yr BP, after which there is a decrease in TBS values in S Sweden. A clear difference between the two regions is the increase of Juniperus, tall evergreen shrub (TSE), in C Sweden during 5500-4000 cal yr BP. Proportion of shrubs (TSD, i.e. Salix and LSE, i.e. Calluna vulgaris) and grassland taxa (GL, i.e. all herbs) is higher in the early Holocene and decreases towards the mid-Holocene in both regions in Sweden. In S Sweden a slight increase in the proportion of grassland and agricultural land taxa (AL, i.e. Cerealia) occurs from 6000 cal yr BP onwards. Of the explanatory variables, in both regions summer temperature increases by about 1 °C between 9000–7500 cal yr BP and then gradually decreases until the end of the study period at 4000 cal yr BP. Winter temperature increases by almost 2 °C over the entire study period. Fire history differs between the regions. In S Sweden increasing fire activity is detected until 7250 cal yr BP followed by a decreasing trend. In C Sweden fire history is characterized by five peaks: at 10000 cal yr BP, at 8000 cal yr BP, at 6200 cal yr BP, at 5000 cal yr BP, and at 4000 cal yr BP. Human population size follows a similar pattern in both regions, remaining relatively low until the beginning of agriculture c. 6000 cal yr BP, when the human population size begins to increase culminating c. 5500 cal yr BP in S Sweden and slightly later in C Sweden. Thereafter, human population size declines until the end of the study period. Figure 5.2 Holocene changes in the explanatory variables: 1) simulated summer and winter temperatures (data from the LOVECLIM model), 2) fire (charcoal influx z-scores) and 3) human population size data at each study region and in the plant functional types (PFTs) based on the REVEALS estimates of 25 taxa. The meaning of each PFT see Table 5.1. Note that the studied periods are different between Sweden (4–10 cal kyr BP) and Finland (1–10 cal kyr BP).
CHAPTER 5 Finland A clear increase in the proportion of Picea (TBE1) occurs from 6500 cal yr BP in SE Finland and from 5500 cal yr BP in SW Finland. In both regions the proportion of Pinus (IBE) decreases slightly during the early Holocene and remains constant throughout the Holocene, with a slight increase during last 2000–1000 cal yr BP. In both regions there is a clear increase in the proportion of IBS around 9000 cal yr BP and then a strong decrease since 6000 cal yr BP towards the late-Holocene. The maximum proportion of TBS occurs in both regions during the Holocene Thermal maximum (HTM) at 8000-5000 cal yr BP. In SW Finland, the proportion of Juniperus (TSE) increases from 1500 cal yr BP onwards together with Salix (TSD), Calluna vulgaris (LSE), grassland taxa (GL) and agricultural taxa (AL). Similar trends are seen in SE Finland, but slightly later. Climate variables have similar trends in Finland as in the Swedish regions. However, in SW Finland there is slightly more variation in the early Holocene winter temperatures. In SW Finland the early Holocene is characterized by the highest fire activity. Then fire activity decreases until 4000 cal yr BP, while a slight increase is detected afterwards. In SE Finland low fire incidence is detected until circa 1500 cal yr BP, and then fire increases. The human population size remains relatively low between 10 000 and 6000 cal yr BP, rapidly increases from 6000 to 5500 cal yr BP, and then decreases to 3500 cal yr BP in both regions. In SW Finland, the population record shows increase between 3500 and 1000 cal yr BP. The main difference in human population size patterns between the study regions is the mid-Holocene population peak, which is more pronounced in SE Finland. In SE Finland, human population size at 1000 cal yr BP remains smaller than during the midHolocene, whereas in SW Finland human population size reaches its highest levels at the end of the study period. 5.3.2 Variation partitioning results The whole Holocene Climate individually explains the highest fraction of variation in PFTs in all regions between 4000 – 10 000 cal yr BP in Sweden and 1000–10 000 cal yr BP in Finland (Fig. 5.3). In S and C Sweden, the relative importance of winter temperature on PFT variation is higher than that of summer temperature. In SW and SE Finland, summer temperatures explain a higher proportion of the variation in vegetation than winter temperatures (Table 5.2). Winter and summer temperatures have significant effect on vegetation changes in all regions, with the exception of C Sweden, where the effect of summer temperature on vegetation composition is not significant (Fig. 5.3). In general, when the whole Holocene is considered, the variation explained by human population size and forest fires is relatively low. Human population size has significant impact on vegetation changes in C Sweden and SW Finland, while forest fires have significant effect in S Sweden and SE
Finland. Climate, forest fires and human population size together explain more than 75% of the variation in PFTs in all regions. Table 5.2 PFTs variation explained individually by mean summer and winter temperatures for each study region. The statistical significance (*** p < 0.001, ** p < 0.01, * p < 0.05 and NS=non-significant) of each parameter on PFT variation is shown.
Whole period Sweden S Sweden C Finland SW Finland SE
WinT 7 *** 28 *** 6 *** 8 ***
SumT 3** 1 NS 24 *** 15 ***
before onset of farming WinT 9 ** 37 *** 14 *** 17 ***
SumT 5* 3 NS 13 ** 10 ***
after onset of farming WinT <0 NS <0 NS <0 NS <0 NS
SumT 16 NS 17 NS 4 NS <0 NS
Figure 5.3 Percentage of variation in vegetation (PFTs) during the Holocene explained by climate (winter temperature: WinT, summer temperature: SumT), fires (Charc) and human population size (Pop) in S Sweden (a), C Sweden (b), SW Finland (c) and SE Finland (d). The statistical significance (*** p < 0.001, ** p < 0.01, * p < 0.05 & p < 0.1) of each parameter on variation in vegetation is shown.
Pre-agriculture Before the onset of farming (6000 cal yr BP in Sweden and 4000 cal yr BP in Finland) climate explains most of the variation in PFTs in S Sweden (13%), C Sweden (54%), SW Finland (25%) and SE Finland (32%) (Fig. 5.4). Although winter temperatures explain a higher proportion of variation in PFTs than summer temperatures, both climate parameters have significant effect on changes in vegetation dynamics in all regions, except for summer temperatures in C Sweden (Table 5.2, Fig 5.4). Fires have significant effect on vegetation composition in both regions in Finland, explaining 4% of the variation in vegetation both in SW and SE Finland. The shared effect of climate and forest fires is relatively high in S Sweden (22%), SW Finland (25%) and SE Finland (20%). Human population size explains
CHAPTER 5 8% of the variation in vegetation in SW Finland, but is not significant in other regions. In general, in all regions, climate, forest fires and human population size explain at least 70% of the variation in vegetation composition before the onset of agriculture.
Figure 5.4 Relative importance of climate (WintT, SumT), forest fires (Charc) and human population size (Pop) on the variation in vegetation during pre-agricultural period in S Sweden (a), C Sweden (b), SW Finland (c) and SE Finland (d).The statistical significance (*** p < 0.001, ** p < 0.01, * p< 0.05 and . p < 0.1) of each parameter is shown.
Figure 5.5 Relative importance of climate (WintT, SumT), forest fires (Charc) and human population size (Pop) on the variation in vegetation during agricultural period in S Sweden (a), C Sweden (b), SW Finland (c) and SE Finland (d).The statistical significance (*** p < 0.001, ** p < 0.01, * p < 0.05 and . p< 0.1) of each parameter is shown.
Agricultural time After the onset of farming, human population size explains the highest amount of the variation in PFTs in S Sweden and SW Finland (23% and 22%, respectively)(Fig 5.5). Individually climate explains relatively low fraction of the variation in PFTs in all regions. However, the joint effect of climate and other variables is relatively high. In general, summer temperatures explain a higher proportion of variation than winter temperatures (Table 5.2). However, neither summer nor winter temperatures have a significant effect on vegetation in the study regions. Forest fires do not explain individually any variation in vegetation; however, together with human population size, they explain 10% of variation in vegetation dynamics in SE Finland. In general, a relatively high proportion of variation in vegetation is left unexplained in all regions during the agricultural time. Moving window approach Results of moving window approach (Table S5.1, Fig. 5.2), when variation partitioning is applied in shorter time periods, show strong impact of climate on vegetation dynamics from early to mid-Holocene in S and C Sweden, especially between 8600–6000 cal yr BP, when climate explains 60–80% of the variation in PFTs. In both regions in Sweden the relative importance of climate decreases ca 6000–5500 cal yr BP remaining relatively low until it increases again at the end of the study period. The importance of human population size increases ca 6000–5500 cal yr BP in S Sweden explaining up to 60% of the variation in PFTs (Fig. 5.6). In C Sweden the explanatory value of human population size remains relatively low throughout the Holocene. In Finland, the results of moving window approach (Table S5.1 & Fig. 5.3) demonstrate notable fluctuations in the importance of climate in both regions. In SW and SE Finland climate explains ca 40–60% of the variation at ca 8000–7000 cal yr PB, at ca 6200–5200 cal yr BP and at ca 4500–3000 cal yr BP. In SW Finland the relative importance of climate remains low after ca 3000 cal yr BP, while in SE Finland the importance of climate is low ca 3000–1500 cal yr BP after which it explains over 30% of the variation in PFTs at the end of the study period. In general, the importance of human population size is low, except for the period of 5200–4500 cal yr BP, when human population size and climate both explain 20% of the variation in PFTs. In SW Finland the relative importance of human population size increases from ca 3500 cal yr BP exceeding the explanatory value of climate and during the period of 2500–1000 cal yr BP human population size explains 20% of the variation in PFTs (Fig. 5.6). In SE Finland the relative importance of human population size remains notably low during the late-Holocene.
Figure 5.6 Relative importance of climate and anthropogenic influence on variation in vegetation dynamics through the Holocene in northern Europe. Panel A shows the fraction of variation in vegetation explained by human population size and climate in S Sweden and SW Finland for ten 200 year subsets of data in 2000-year mowing time windows. Panel B shows the fraction of variation explained by climate and human impact for 14 subsets of data in 1200-year moving time window in Estonia (Reitalu et al. 2013). In panel A and B the results of each time window is shown from the midpoint of the time period, and the used time windows are shown in Table S5.1. Panel C shows the fraction of variation explained by climate and land use average across southern Fennoscandia and Baltic countries (Marquer et al. submitted).
5.4 Discussion 5.4.1 From natural vegetation dynamics towards increasing anthropogenic influence Role of climate and forest fires as natural drivers of vegetation dynamics In the Holocene perspective, climate is evidently the main driver behind regional vegetation dynamics in Fennoscandia. The importance of climate on variations in vegetation dynamics is clearly highest during the beginning of the HTM ca. 8000–7000 cal yr BP and again around 6000–5000 cal yr BP corresponding with the main temperature shifts during the Holocene. This can be seen especially in S Sweden and SW Finland (Fig. 5.6A). Similar trend in the importance of climate can be seen also in Marquer et al. (in review) (Fig. 5.6C). The strong role of climate in vegetation dynamics at the beginning of the HTM is in agreement with the well-established northward range shift of temperate tree
taxa such as hazel (Corylus), lime (Tilia), elm (Ulmus) and oak (Quercus) during this period (e.g. Alenius & Laakso, 2006; Miller et al. 2008; Seppä et al. 2015) and can be seen as an increase of broadleaf forest cover in Fennoscandia. This corresponds with the present ecological understanding that temperatures and moisture conditions are the main limiting factors for the northern limits of tree species (Bonan and Shugart, 1989; Woodward, 2004; Jackson et al. 2009). However, it is important to note that, when the climatic conditions are relatively constant, the role of climate decreases suggesting that other factors (such as the human impact, disturbances and site characteristics) become more important. The relatively strong impact of climate on vegetation dynamics between 6000–3000 cal yr BP in Finland, demonstrated by the results of moving window approach (Figs. 5.3 & 5.6, Table S5.1) can be connected to the large scale shift from temperate broadleaf forests to spruce-dominated conifer forests following the migration of spruce across Finland ca. 6500–3000 cal yr BP (Tallantire 1972; Giesecke and Bennett 2004; Seppä et al. 2009a). The drivers behind the migration and expansion of spruce to Fennoscandia are still unclear, however it coincides with the cooler climate conditions in the last half of the Holocene and it is plausible that spruce has benefited from cooling climate. Although, the importance of human population size on vegetation dynamics increases from the early Neolithic in Sweden and during the late Holocene in Finland, climate remains as important driver behind the changes in vegetation cover, especially in C Sweden and SE Finland (Table 5.2, Fig. 5.2). Furthermore, the results of Reitalu et al. (2013) and Marquer et al. (in review) demonstrate strong effect of climate on vegetation dynamics in Estonia and southern Scandinavia during the last 1000 years. These results suggests that the projected accelerating future warming in northern areas (Christensen et al. 2013) will most probably act as main driver of the future changes in boreal ecosystems, at least in regional scale. The change from a natural- to a more human-induced fire impact on vegetation dynamics is not clearly defined in our data. The higher importance of forest fires on vegetation dynamics during the pre-agriculture than after the onset of farming in both Sweden and Finland is most probably due to the fact that in Sweden the study period ends at 4000 cal yr BP and in Finland 1000 cal yr BP. Therefore it is plausible that the impact of increased fire activity due to human impact on the late Holocene vegetation dynamics (Molinari et al. 2013) is not detected in our analysis. However, pre-agriculture fires and climate have a notable joint effect on variation in vegetation, whereas after the onset of farming, vegetation changes are partly explained by the joint effect of fires and human population size. Since slash-and-burn practices were often used in connection to early farming (Huttunen 1980; Granström & Niklasson 2008), it is plausible that the increased joint effect between human population size and fires indicate the change from a natural- to more human-induced fires affecting the vegetation dynamics during the late Holocene.
CHAPTER 5 Increasing anthropogenic influence from early Neolithic In Figure 5.6 we compare our results with those of Reitalu et al (2013) for Estonia and Marquer et al. (in review) for Fennoscandia and Baltic, all based on a moving time window approach. The results demonstrate an increasing role of human population size on vegetation dynamics from the mid-Holocene in S Sweden (Fig. 5.6). At this time Neolithic farming was widespread in central Europe (Shennan 2009; Fyfe et al. 2015: Marquer et al. 2014) and more intensive land use practices were spreading towards northern Europe. Currently, the general view is that the shift from hunter-gatherer societies to agricultural societies was gradual (Sørensen & Karg 2012) and this can be seen in the notable differences in the impact of human population size on vegetation dynamics between the study regions. In S Sweden notable increase in the importance of human population size ca 6500 cal yr BP onwards suggests more intense land-use by growing population. This is in accordance with the view that agrarian expansion to southern Scandinavia was relatively rapid process lasting only few centuries (Sørensen & Karg 2012). It is plausible that this rapid shift was due to suitable climate conditions during the HTM and pressure form the growing population in central Europe. The Mesolithic hunter-gatherer-fisher communities had relatively low impact on surrounding vegetation and the shift to agrarian society caused changes in the landscape, when Neolithic communities started to migrate inland and broadleaf forests were first used as woodland pastures and then cleared for field cultivation (e.g. Lagerås 1996; Berglund et al. 2008; Sköld et al. 2010). During the mid-Holocene, human population size was notably lower in SW Finland than in S Sweden and climate was still the main driver of vegetation dynamics. There is no detectable indication of agricultural land use or other notable anthropogenic impact during mid-Holocene in SW Finland, and the low human impact until 3500 cal yr BP is in line with the small human population size. This is in corresponds with the view that in Finland vegetation stayed longer under natural processes than in S Scandinavia and in the Baltic regions (Marquer et al. 2014; Fyfe et al. 2015). The increasing importance of human impact in SW Finland corresponds with the onset of agriculture ca 4000 cal yr BP (Taavitsainen et al. 1987, 1998; Lahtinen et al. 2017). Results from the moving window approach suggest that human population size was the major driver of vegetation dynamics in SW Finland during the late-Holocene and especially during the period of 2500–1000 cal yr BP. The increasing human population impacted on vegetation dynamics by creating forest pasture for cattle and furthermore with forest clearances creating more open landscapes with increasing proportion of grassland (herbs) and arable land. Furthermore, cultivated fields were commonly left as fallow for few years in order to maintain the soil productivity. This may have resulted as the increase in the proportion of shade-intolerant summergreen trees such as birch and alder together with shrubs (i.e. Salix) in vegetation cover. A similar impact of early cultivation on vegetation composition is seen also in Estonia (Reitalu et al. 2013). At the same time with increasing population size, decline of
spruce forests can be seen in Finland (Fig 5.2). A similar late-Holocene decline in spruce population is also recorded in central Sweden, in the Baltic regions (Niinemets & Saarse, 2006; Seppä et al. 2009; Heikkilä & Seppä, 2010). Since spruce prefers fertile soils that could have been exploited also for cultivation, it is possible that the decline of spruce dominated conifer forests is due to forest clearances. However, Kuosmanen et al. (2014) recorded the late-Holocene spruce decline is also in NW Russia from remote small forest hollow sites, where no evidence of human impact around the sites were present during the decline in spruce population. This suggests that climate may have been the driving factor behind the late-Holocene spruce decline and that climate, together with human impact, has strong impact on late-Holocene vegetation dynamics. The markedly lower importance of human population size on regional vegetation dynamics in C Sweden and especially in SE Finland could be explained by the fact that these regions were still more forested and less populated during the length of the study period. One reason for the later expansion of more intensified cultivation to SE Finland might be the shortage of suitable areas for field establishment. Due to the land uplift substantially large areas of SE Finland were raised rapidly above the Baltic see-level after the deglaciation in contrast to SW Finland (Tikkanen & Oksanen 2002) and therefore SE Finland is lacking the thick fine-grained mineral soil layers, such as clay and silts, which provide fertile soils for cultivation in SW Finland. It is plausible that due to the lack of suitable areas for field cultivation, human population size remained relatively low in Se Finland and animal husbandry together with sporadic slash-and-burn cultivation were practice longer than in SW Finland (Taavitsainen et al. 1998). Alenius et al. (2008, 2013) showed that forest clearances and cultivation intensified only from 600 A.D. onwards in eastern Finland, which coincide with the increasing human population size in the area. It is noteworthy that our dataset only extends up to 4000 cal yr BP in Sweden and 1000 cal yr BP in Finland and it is probable that the more intensive human impact in SE Finland during last 1500 cal yr BP is not detect by variation partitioning method. It is evident that the role of human population size on variation in vegetation increases already from the early Neolithic. The decreasing importance of climate together with increasing importance of anthropogenic impact in Holocene vegetation dynamics at ca. 7000–6000 cal yr BP in Sweden and at ca. 4000–3000 cal yr BP in Finland is connected to increasing human population size. However, the relatively high importance of climate throughout the Holocene in our results together with results of Reitalu et al. (2013) and Marquer et al. (in review) suggests that after the onset of farming climate, together with human impact, remains an important factor in vegetation dynamics, and although the anthropogenic impact may have regionally overruled the effect of changing climate, climate can be considered as the main driver of long-term regional vegetation dynamics in Fennoscandia during the Holocene.
CHAPTER 5 5.4.2 Potential biases in the method and explanatory variables Here we further tested the benefits of variation partitioning as a method to detect the drivers behind past vegetation dynamics. Kuosmanen et al. (2016a) suggested that the variation partitioning is more suitable for detecting trends, such as the effect of climate on vegetation compositions, rather than the effect of abrupt events, such as a fire. Our results, with spatially more comparable datasets for response and explanatory variables, demonstrate that overall climate, forest fires and human population size explain more of the variation in Holocene vegetation dynamics than the same variables used in Kuosmanen et al. (2016a). It is notable that over 75% of the variation in vegetation during the whole study period was explained by these three variables. However, the results from the agricultural time period show notably larger amounts of unexplained variation. This might be explained by the shorter length of the study period (2000 years in Sweden and 4000 years in Finland) together with increased small scale fluctuations (higher unpredictability) in PFTs and also in regional fires and human population size data (Fig 5.2). In addition to climate, forest fires and human population size, several other factors such as soils, productivity, local hydrology, light availability, topography and biotic interactions (Selikhovin 2005; Kuneš et al. 2011; Post 2013; Mitchell et al. 2006) can have short-term effect on vegetation dynamics and be behind the variation left unexplained in the analysis. These results support the previous assumption that when assessing the factors behind past vegetation dynamics variation partitioning is better suited for detecting the long-term drivers of vegetation, while the role played by factors characterized by abrupt or short-term effects on vegetation needs to be addressed with different methods. However, environmental factors can affect vegetation dynamics individually or in interaction with other variables and one benefit of variation partitioning is that it indicates also the variation explained jointly by individual variables. This can provide insight into the complex interactions between different environmental variables affecting vegetation dynamics. It is also important to note that variation partitioning as a method does not consider the causality, but just the covariation between variables. Of the explanatory variables used in this study the causality between climate and vegetation is relatively clear and it can be assumed that climate is leading the changes in vegetation dynamics. Furthermore, during the agricultural time it can be assumed that the human impact is the cause of decrease in forest cover and increase in arable land. However, the question of causality between the changes in vegetation and in forest fires and hunter-gatherer population size is more difficult to deduct. Hunter-gatherers shaped the vegetation by favoring suitable tree species such hazel (Holst 2010) and by using fire to create suitable conditions for game. However, hunter-gatherer populations were small and scattered, affecting primarily vegetation dynamics on the local scale (Regnell et al. 1995; Hörnberg et al. 2005) and most probably did not leave an impact that would be detectable in regional vegetation. Regional scale vegetation changes reflect broader ecosystem changes and certain ecosystems are more
favorable to hunter-gatherers than others. For example, boreal forests provided poorer environments for hunter-gatherers in terms of food abundance compared to temperate mixed forest (Tallavaara & Seppä 2012). In Finland, the decline in the hunter-gatherer population size along with the decline of temperate forests coincides with the expansion of spruce population and consequent boreal ecosystem (Seppä et al. 2009a). Therefore, it can be speculated, that the change from temperate broadleaf forest ecosystem to boreal conifer forests ecosystem had an impact on the human population size rather than that huntergatherers influenced the change in vegetation. Similarly, Ohlson et al. (2011) showed that fire frequency in Fennoscandia decreased after the expansion of spruce. However, it is still speculative, if spruce expanded because of the decrease in fire activity or did the fires decrease due to the spruce expansion. However, it is also plausible that both (i.e. changes in vegetation dynamics and human population size) were driven by climate (Tallavaara & Seppä 2012) and instead of a causal relationship between human population size and vegetation, these variables co-vary because of the changes in climate. Since disentangling these issues of causality with variation partitioning is challenging, would future studies would benefit from the use of methods that provides more insight into the direction of the interaction between response and explanatory variables, for example time-series analysis such as wavelet coherence or different modelling approaches. Simulated climate data Climate parameters, summer and winter temperature, for this study was derived from simulation performed with the LOVECLIM climate model (Zhang et al. 2016). However, it is noteworthy that the results might differ depending on the applied climate model. The used climate simulation is realistic on large sub-continental scale in terms of the forcing, but there might be uncertainties in simulating finer regional differences in climate history. However, the output of the LOVECLIM model simulations have been compared with proxy data (pollen, glacier frequency and chironomid data) and these comparisons reveal that the simulation is generally reliable in the Fennoscandian region (Zhang et al. accepted). Furthermore, derived temperatures correspond with the δ18O records from Finland (Heikkilä et al. 2010) and Sweden (Reference) and with the climate reconstructions for Fennoscandia (e.g. Heikkilä & Seppä 2003: Seppä et al 2009b). Charcoal data In Kuosmanen et al. (2016a, b), charcoal and vegetation data were derived from individual sites and the importance of forest fires on vegetation changes were surprisingly low. They argued that the reason for this could be 1) the bias caused by the time-lag between the peak in charcoal record and corresponding response in vegetation, and 2) the fact that variation partitioning records the overall trends, but not the direction of change and therefore the short-term negative association between fire and vegetation dynamics can be deleted by short-term positive association resulting in a low overall fraction of the explained variation.
CHAPTER 5 To overcome this problem, we use composite charcoal curves that provide regional trends of fire history. The protocol for the construction of the regional-grouping transformed charcoal influx z-scores is widely used within the paleofire community (Power et al. 2010, Daniau et al.2012, Molinari et al. 2013) and can be considered as a reliable method for identifying shared features and trends in fire history that may exist in a given spatial or temporal domain (Marlon et al., 2016). Averaging multiple records can provide insights into changes in fire history that only manifest at a broad spatial scale and can thus overcome the high variability existing within individual charcoal time series. In general, the results with our spatially more comparable dataset demonstrate that the role of fire history in explaining past vegetation dynamics through the Holocene (Fig. 5.3) is more pronounced than in Kuosmanen et al. (2016a). However, the overall importance of forest fires remains still relatively low and the results are notably different between study periods and regions. Human population size proxy Over the last couple of decades, the use of radiocarbon dates as a proxy for regional or continental past human population size has developed into a mainstream technique in archaeology (Rick 1987; Shennan and Edinborough, 2007; Shennan et al. 2013). The advantage of this proxy is that it is independent from the reconstructed vegetation dynamics. However, there has been criticism against the approach, mainly in the form of lists of potential factors that may bias demographic interpretations of temporal distributions of archaeological radiocarbon dates (Crombé and Robinson 2014; Mökkönen 2014; Attenbrow and Hiscock 2015; Torfing 2015; Torfing 2016). However, when applied carefully, this approach is extremely conservative. The global research community has been highly proactive in developing this demographic method by explicitly assessing the impacts of different biases (Edinborough 2005; 2009; Shennan and Edinborough 2007; Surovell et al. 2009; Shennan et al. 2013; Timpson et al. 2014; Crema et al. 2016). Researchers have also continuously evaluated radiocarbon date-based demographic proxies against other independent measures of past population size (e.g. Williams 2012; French and Collins 2015; Williams et al. 2015). For example, using EUROEVOL radiocarbon date data, Downey et al. (2014) were able to show strong correspondence between radiocarbon date-based human population size and changes in population growth rates previously inferred from cemetery data (Bocquet-Appel, 2011). In addition, extensive evaluations of the Finnish data have shown that all available population proxies, some of which are totally independent from archaeological data and methods (e.g., genetics), support the pattern revealed by radiocarbon date-based proxy and its demographic interpretation (Oinonen et al. 2010; Tallavaara et al. 2010; Tallavaara et al. 2014). Thus, we are confident that the archaeologically-derived relative population size proxies used here reflect true demographic signal from the past to a first order of approximation.
5.5 Conclusions Hunter-gatherer populations did not significantly affect vegetation dynamics in Fennoscandia and climate was the main driver of change at that time. Large scale shifts in vegetation are driven by climate. However, when the climatic conditions stay relatively constant, the role of climate may decrease and other factors, such as disturbances, site characteristics, competition, and human impact may have greater effect on vegetation dynamics. Agricultural populations, however, had greater effect on vegetation dynamics and the role played by human population size became more important during the late Holocene. There is a clear regional-dependency related to human population size, i.e. human impact on vegetation dynamics was notably higher in S Sweden and SW Finland, where land use was more intensive, than in C Sweden and southeast Finland. Variation partitioning approach provides important insights into the drivers of past vegetation dynamics and gives a stepping stone for finding more precise methods to assess the processes behind past changes in long-term vegetation dynamics. Our results underline main changes in the role of climate, forest fires and human population size on Holocene vegetation dynamics in Fennoscandia. Although climate can be considered as the main driver of long-term Holocene vegetation dynamics, our results suggest that in some regions the role of human population size in vegetation dynamics may have been greater than climate already from the early Neolithic.
Chapter 6 Synthesis and outlook
6.1 Summary of the main findings This thesis focused on the Holocene climate history by employing various methods in different chapters. The overall goal was to obtain a better understanding of ice sheet– ocean–atmosphere processes during the early Holocene transition and thus a reliable estimation of Holocene climate, as that has potential implications for ongoing climate change. Chapter 2 examines the effects of different forcings on the climate and tests two FWF scenarios with the LOVECLIM model, while Chapter 3 compares four sets of model results to evaluate the reliability of Holocene climate simulations. In Chapter 4, we use proxy data (including pollen, chironomids and δ18O in Greenland) to further evaluate the simulations and analyse the most likely Holocene climate history over five regions at high latitudes. Finally, Chapter 5 investigates the role of climate together with human disturbance and forest fires in Holocene vegetation dynamics with variation partitioning statistical analysis. The main findings are summarized below by addressing the research questions presented in the introductory Chapter 1. 6.1.1 Effects of the ice sheets on Holocene climate change in the NH extratropics The main question addressed in Chapter 2 is ‘How does the Holocene climate in the NH extratropics respond to the main forcings?’. To answer this question, the spatial pattern and temporal characteristics of temperature were analysed in climate simulations performed with the LOVECLIM model. Spatial patterns of anomalous temperature at the onset of the Holocene a) The main spatial characteristics of temperature at the onset of the Holocene under
CHAPTER 6 different forcings, especially with the decaying ice sheets At the onset of the Holocene, the climate shows latitudinal temperature variations in response to the ORB and GHG, and ice sheets forcing induces spatial heterogeneity in the simulated temperature (Fig. 6.1A). The results of OG11.5 (forced with ORBGHG) reveal that the positive ORB forcing overwhelms minor negative GHG radiative anomalies, and thus causes positive summer temperature anomalies of 2–4 °C over most of the extratropical continents relative to 0 ka, with a maximum deviation of 5 °C over the central parts of the NH continents. The simulated temperatures also show latitudinal patterns, especially in winter. A north–south gradient exists, going from a clear warm anomaly at high latitudes to areas further to the south with less warming or even cooler conditions. The OGIS11.5 simulation (forced by full forcings by additionally including the ice-sheet forcing) suggests clear spatial heterogeneity of climate responses at 11.5 ka. Ice sheets induce strong summer cooling over ice-covered areas, which contributes to these heterogeneous patterns. In particular, the most distinct feature is a thermally contrasting pattern over North America with simulated temperatures being around 2 °C higher than at 0 ka over Alaska, contrasting with more than 3 °C lower temperatures over most of Canada. Relatively small temperature anomalies can be observed in other regions, for instance slightly negative temperature anomalies in Siberia and small positive values in the Arctic Ocean. b) The possible mechanisms behind these simulated temperatures The maximum temperature reduction in the simulation performed with full forcings was found over the centre of the LIS. Although the ORB and GHG forcings resulted in overall higher temperature than the pre-industrial era due to overwhelming positive anomalies of the ORB, the ice-sheet forcings primarily induced cooling of the climate. The underlying mechanisms of this cooling include enhanced surface albedo over ice sheets, anomalous atmospheric circulation, reduced AMOC and relevant sea-ice feedbacks. The anomalous atmospheric circulation and AMOC can influence the regions beyond the boundary of the ice sheets, in comparison to simulations without the ice sheets. The cooler climate over the northern Labrador Sea and the North Atlantic was related to both reduced northward heat transport due to reduced AMOC strength and enhanced sea-ice feedbacks, which was induced by the freshwater from the melting ice sheets. A small summer temperature anomaly in Siberia is probably because the positive insolation anomaly was offset by the cooling effect of the high albedo associated with the relatively extensive tundra cover in the early Holocene. The relatively warm early-Holocene climate in Alaska was mainly due to increased southerly winds induced by the LIS. Overall, the combination of ORB, GHG and ice-sheet forcings at 11.5 ka resulted in cooling over most of regions, but over most of the Arctic Ocean the climate was warmer than preindustrial, because strong insolation anomalies overwhelm effects of a weakened AMOC induced by meltwater releases.
Figure 6.1 (A) Simulated temperature anomalies at 11.5 ka relative to 0 ka. (B) Holocene temperature evolution in simulations performed with different forcings, with colored lines indicating different regions.
Temporal trends of Holocene temperatures a) Temporal trends of Holocene temperatures over different regions The Holocene temperature evolution, especially the early-Holocene temperature trends, reveals geographical variability (Fig. 6.1B). In Alaska, the climate has constantly cooled throughout the Holocene due to the decreasing insolation and anomalous atmospheric circulation. In contrast, northern Canada experienced strong early-Holocene warming with an overall warming rate of over 1 °C kyr-1, and this warming lasted until 7 ka. Although different forcings and mechanisms played different roles in northwestern Europe, the Arctic and Siberia, the overall warming effect was similar for these regions, with a rate of around 0.5 °C kyr-1. b) Contribution of different forcings to these temperature trends The ORB and GHG forcings lead to an overall decreased temperature trend with a larger magnitude in summer than in winter, which shows latitudinal patterns. The cooling effects of ice sheets induce the spatial heterogeneity in the Holocene temperature trends. The climate-ocean system during the early Holocene is also sensitive to FWF forcing, as
CHAPTER 6 indicated by the responses of the NH sea-ice area and AMOC to different FWF scenarios (Fig. 2.14). Different FWF forcings, including the intensity, duration and the location of freshwater release, can result in varied conditions in the oceans. For instance, the enhanced freshwater influx from the GIS and the redistributed meltwater from the FIS caused an alteration in the surface ocean freshening in the Nordic Seas and associated changes in temperature. c) Which FWF scenarios provide more a plausible Holocene temperature evolution regarding the early-Holocene warming and timing of the HTM? Clear differences between two scenarios are found in NW Europe, where abundant proxy records are available. The comparison of Holocene temperatures regarding the earlyHolocene warming and HTM over northwestern Europe with proxy records suggests that the updated FWF scenario (FWF-v2, with a larger FWF release from the Greenland ice sheet and a faster FWF from FIS) represents a more realistic climate. The OGIS_FWF-v1 simulation may underestimate the cooling effect of ice-sheet melting, which indicates two peaks at around 10 and 7 ka in the temperature evolution over northwestern Europe. High temperatures at 7 ka are recorded in proxy-based reconstructions, but no warm peak at 10 ka has been observed in pollen-based reconstructions (Mauri et al. 2015) in Europe. In contrast to OGIS_FWF-v1, the OGIS_FWF-v2 simulation produced a warming trend that is consistent with the highest temperature around 7 ka. Moreover, the OGIS_FWF-v1 produced a temperature decrease between these two peaks, whereas the proxies indicated a persistent temperature increase by 10 ka, followed by a more gradual warming (Brooks et al. 2012). Therefore, from the viewpoint of temperature evolution in northwestern Europe, the OGIS_FWF-v2 simulation represents a more realistic climate than OGIS_FWF-v1, which also demonstrates that the existing uncertainties in the reconstructions of ice-sheet dynamics can be evaluated by applying different freshwater scenarios. 6.1.2 Inter-model comparisons of the Holocene climate trends 1) What are the consistencies and divergences among these Holocene simulations? The LOVECLIM simulation OGIS_FWF-v2 was compared with similar simulations performed with CCSM3, FAMOUS and HadCM3. The multiple simulations suggest generally consistent temperatures in the NH extratropics as a whole, with better agreements in summer than in winter. On a regional scale, reasonably consistent temperature trends are found over the regions where climate is strongly influenced by the ice sheets, including Greenland, N Canada, N Europe and central-West Siberia. These simulated temperatures generally follow a similar pattern of early-Holocene warming, HTM and a gradual decrease toward 0 ka (Fig. 6.2A). Within the above general patterns, the warming extent and rate during the early Holocene varies across these regions. The strongest early-Holocene warming, up to 5 °C in summer and 10 °C in winter, is found in
N Canada; whereas NE Europe and central-west Siberia show the lower warming magnitude among these regions, with 4 °C warming in winter and 1–3 °C cooling in summer. An intermediate degree of warming is found in Greenland and NW Europe, with about 2 °C in summer and 8 °C in winter. Overall, these generally consistent temperature trends illustrate that the structural and parametric uncertainties across the individual models are smaller than climate forcing effects and internal climate variability in these regions.
Figure 6.2 (A) Multi-model ensemble mean of simulated Holocene temperatures (in °C) over the regions where multiple simulations are broadly consistent. (B) Simulated temperatures over the regions where temperatures are less consistent across the simulations, with grey indicating the ensemble range and thick black lines depicting the ensemble mean.
By contrast, large inter-model variations exist in the regions where the cooling effects of the ice sheet were relatively weak, such as in Alaska, the Arctic and E Siberia, through a series of indirect influences (Fig. 6.2B). In particular, during the early Holocene, the signals of multi-model simulations are incompatible in winter, when both positive and negative early-Holocene anomalies are suggested by the different models. At 11.5 ka, the spread of the simulated temperatures anomalies ranges from +2 °C in LOVECLIM to -6 °C in FAMOUS in Alaska. This distinct multi-model difference in winter thus amounts up to 8 °C, which is considerably larger than in summer when the inter-model difference is only 1 °C. Over the Arctic, the discrepancies between the simulations also primarily exist in winter. At 11.5 ka, the winter temperature anomaly is slightly positive in the LOVECLIM
CHAPTER 6 simulation, while more than 8 °C cooling is produced in FAMOUS. Nevertheless, the ensemble mean temperature suggests 1 °C cooling in summer and 4 °C warming in winter throughout the Holocene. Relatively large multi-simulation differences are found over E Siberia in both summer and winter, reaching up to 3 °C at the onset of the Holocene. Over the course of the Holocene, the simulations show decreased trends of summer temperatures over E Siberia, with the largest decrease (more than 4 °C) in CCSM3, which contributes to the large variation among these simulations. In winter, over 2 °C Holocene cooling is simulated by LOVECLIM, contrasting with up to 5 °C warming in FAMOUS. 2) What climate variables in models cause these inter-model divergences? Divergent temperatures in the above regions can be attributed to inconsistent responses of model variables, such as southerly winds, surface albedo and sea ice. The strong southerly winds induced by the LIS over Alaska and part of E Siberia result in an anomalously warm climate (more than 3 °C) in LOVECLIM. During the early Holocene, higher summer temperatures (1–2 °C) over E Siberia in CCSM3 than in other models are caused by a strong difference in surface albedo (more than 20%) between 11.5 and 0 ka, which is ultimately associated with an unrealistically high albedo at 0 ka. The wide spread of simulated Arctic temperatures in winter may be attributed to inter-model differences in sea ice. Extended early-Holocene sea ice cover in the Arctic Ocean in FAMOUS influences the strength of the albedo-related feedback and could explain why the winter temperature in FAMOUS is 2–3 °C lower than the ensemble mean, which contrasts with only limited sea ice changes between the 11.5 and 0 ka in CCSM3. Relatively low winter temperatures in HadCM3 over the Arctic are probably attributable to the thick sea ice at 11.5 ka, which can block the ocean heat release into the atmosphere in comparison with LOVECLIM. 3) Where do these discrepancies originate? The multi-model comparisons reveal that divergent responses in climate variables across these simulations are partially caused by model differences (e.g. different model physics and resolution). The newly adopted formulation of a turbulent transfer coefficient in CCSM3 causes an overestimated albedo over Siberia at 0 ka (Oleson et al. 2003; Collins et al. 2006), which leads to stronger early-Holocene warmth (temperature anomalies) than in others. Moreover, the relatively simplified sea ice representation in FAMOUS probably leads to overestimated sea ice cover in the Arctic Ocean. Furthermore, the coarse vertical resolution in LOVECLIM might introduce strong responses in atmospheric circulation over Alaska. Apart from the different models, the transient feature of the early-Holocene climate driven by the retreating ice sheets also influences the inter-model comparisons. In particular, large uncertainty exists in the FWF forcing, which has a major impact on the early-Holocene climate between 11.5 ka and 7 ka.
6.1.3 Model–data comparisons of Holocene temperatures at NH high latitudes a) To what degree do proxy-based reconstructions agree with model results? The consistencies between the model simulations and proxy-based reconstructions are region-dependent. In Fennoscandia (Fig. 6.3), the ensemble mean summer temperature of the simulations is consistent with pollen-based reconstructions, although there are minor differences between individual simulations (±0.5 °C) and proxy records (about 2 °C before 8 ka). The simulations and composite pollen-based reconstructions reveal a clear earlyHolocene warming, although a stable early-Holocene temperature is indicated by chironomid data. The simulated annual temperature and δ18O-based climate reconstruction in Greenland (Fig. 6.3) are consistent well regarding the Holocene trend. The temperature at 11.5 ka is 6 °C lower than at 0 ka, followed by warming, and reaching a maximum at ~7 ka. The only exception is that δ18O data suggest a strong warming of over 8 °C by 9 ka, leading to an earlier temperature peak than in the simulations. The temperatures based on borehole measurements show a weaker cooling at the onset of the Holocene than in the δ18O data, which agrees better with the simulated temperatures. However, the temperature during the HTM in borehole data was up to 2.5 °C higher than 0 ka, which slightly diminishes its agreement with the simulations. The simulations and pollen data show similar summer temperature trends over the course of the Holocene in northern Canada (Fig. 6.3). The ensemble mean of simulations indicates a warming of 4 °C between 11.5 and 8 ka with an inter-model spread of ~3 °C, and the compiled pollen data extending back to 10 ka suggest 1~2 °C warming until ~7 ka with a spread range of 4 °C. From 7 ka onwards, simulations and pollen data are consistent, showing a cooling temperature trend of ~1 °C toward 0 ka. In contrast, the simulated summer temperature and proxy-based reconstruction are mismatched in Alaska, with opposite signs (Fig. 6.3), particularly in comparisons of the simulations with pollen-based temperature reconstructions. The ensemble mean of the simulations represents a 2 °C cooling trend during the Holocene with a range of 1–2 °C across individual models, while pollen data suggest a warming of 4 °C with a range of 2 °C. Meanwhile, the chironomid-based temperatures show an almost stable level with a 2 °C variation across individual records until 9 ka. Comparisons between the simulations and proxy-based reconstructions during the Holocene are generally noisy in high-latitude Siberia (Fig. 6.3), despite the simulations matching the chironomid results better than pollen data. The ensemble mean of simulations indicates a 2.5 °C cooling trend, with a similar range of inter-model spread from 4 °C in CCSM3 to 0.5 °C in LOVECLIM. In spite of the large spread in the chironomid records, the median of the dataset suggests a Holocene cooling trend of 1 °C, which is slightly less than in the ensemble mean of simulations. Pollen data indicate an increasing temperature until 7 ka, reaching a distinct
CHAPTER 6 maximum of +2 °C, and this warmth lasts until 4 ka, after which the temperatures decrease to the 0 ka level.
Figure 6.3 Comparisons of simulated temperatures with proxy-based reconstructions.
b) What are the potential sources of uncertainty in simulated temperatures and proxy-based reconstructions? 1) Potential uncertainties in proxy data & implications for the interpretation of proxy data Multiple factors contribute to the uncertainty of proxy-based reconstructions, and the dominant sources of uncertainty vary between regions. First, different proxies and transfer functions can explain an important part of the mismatch between proxy-based temperature reconstructions. During the early Holocene, pollen- and chironomid-based temperature reconstructions are different in some regions. Apart from the complexity of reconstructing early-Holocene climate change, one important issue is the difference in the representation of seasonality in these proxies. The occurrence and abundance of chironomids in lakes are strongly determined by summer air and water temperatures (Heiri et al. 2014), while the pollen data can to some extent also reflect winter temperatures, especially for plant species with low cold tolerances. In some of the pollen-based temperature records from Alaska and
Canada, the Modern Analog Technique (MAT) method rather than Weighted Averaging Partial Least Square regression and calibration (WAPLS) was applied to quantitatively reconstruct the climate, which might explain some of the above model–data discrepancies. MAT and WAPLS each have their own merits and weaknesses, and one method can outperform the other for particular datasets, depending on differences in the training-set size, taxonomic diversity and complexity of the species–environment relationship (ter Braak & Juggins 1993; Telford & Birks 2005; 2009; Juggins & Birks 2012). Simple δ18O conversion to temperature in Greenland may induce uncertainty, as other changes, such as atmospheric circulation (Charles et al. 1994), the seasonality of precipitation (Fisher et al. 1983) and its variation through time (Cuffey et al. 1995), affect the isotopic composition, in addition to the local environmental temperature. The borehole temperatures might include regional signals in individual records, as they are down-core measurements of the GRIP ice core (Dahl-Jensen et al. 1998). Secondly, the representativeness of climate signals in proxies might be disturbed by other factors. Disequilibrium vegetation dynamics during the transient early Holocene (e.g. in N Canada) may influence the accuracy of pollen-based temperature records. It has been suggested that the post-glacial migration of trees to deglaciated regions may have been constrained by their limited speed dispersal and population growth rates (Birks 1986a; Webb 1986). Such a time lag in their migration history would prevent the pollen records from tracking rapid climate changes. Thus, it is possible that the considerable temperature rise indicated by the models was not fully reflected in pollen-based climate reconstruction, leading to an underestimation of the early-Holocene warming in pollen data. Non-climatic factors following the last deglaciation influence climate records derived from chironomids data. Apart from temperature, the inconstant hydrobiological and limnological factors, such as nutrient availability, trophic state, and dissolved and total organic carbon in lakes, influenced chironomid distribution and abundance, and thus potentially caused bias in the assumed relationship between chironomids and temperature (Brooks & Birks 2001; Velle et al. 2010). In Fennoscandia, the absence of early-Holocene warming in the chironomid data is probably associated with the influences of non-climatic factors. No correction for palaeo-topography was taken into account in reconstructed temperatures by assuming that the effects of ice thickness and post-glacial rebound are of the same order and they are roughly balanced out, which could induce some uncertainties in model–data comparison. The simulated temperature (e.g. in north Canada and Fennoscandia) may be influenced by palaeo-topographic changes due to the reducing thickness of the LIS and the undergoing post-glacial rebound. Thirdly, statistical uncertainties exist in the composite reconstructions of some regions and also in selection processes. Each individual record carries some site-specific signals, which can be incorporated into compiled reconstructions when only a low number of proxy records are available. Given the extremely low number of records (only five) in Siberia,
CHAPTER 6 the reliability of the proxy-based reconstruction in high-latitude Siberia is statistically lower than in other regions. In addition, uncertainties related to a coarse temporal resolution might be induced in individual records when the expanded selection criteria were applied in record selection, such as in N Canada. 2) Potential uncertainties in simulations and implications for model simulations Similarly to the proxy approach, multiple uncertainty sources exist in simulations, which also and show region-dependency. First, palaeo-topographic changes related to the retreating ice sheets and associated post-glacial rebound potentially induce uncertainties in the simulations over N Canada and Fennoscandia. Moreover, it is well known that simulating the temperature over the ice sheets is a challenge. The accuracy of climate simulation over ice sheets strongly depends on the model resolution, as a high resolution allows a detailed representation of topography and precise description of thermodynamics (Ettema et al. 2009). From this viewpoint, potential uncertainties in simulations could also be attributable to a slightly underestimated mid-Holocene warmth over Greenland in simulations. Uncertainty might also be included by taking the rectangular box of the simulations as a representative of the Greenland region, as simulated temperatures are sensitive to the southeastern coastal area. Additionally, various representations of physical processes in models might be overestimated for some parameters. In high-latitude Siberia, CCSM3 tends to simulate a high albedo bias at 0 ka in the later adopted formulation of the turbulent coefficient (Collins et al. 2006; Oleson et al. 2003). This high albedo leads to anomalously low temperatures at 0 ka, and a deceptively warm early-Holocene in Siberia (more than 4 °C warmer) in CCSM3, which is far above either pollen or chironomid-based temperature reconstructions, implying overestimated early Holocene warming in CCSM3. c) What are the most probable Holocene temperature trends? Additionally available Holocene data include glacier frequency data from Fennoscandia and frequency data for peatland initiation in Alaska, which can provide additional evidence to climate history. The glacier data (Nesje et al. 2009) also suggest a clear early-Holocene warming in Fennoscandia, which matches better with pollen-based reconstructions and model simulations than with the chironomid data. Therefore, multiple lines of evidence suggest a pronounced early-Holocene warming in Fennoscandia, which is followed by a gradual decrease temperature towards the preindustrial value. The high frequency of peatland initiation during the early Holocene in Alaska (Jones & Yu 2010) could result from a high temperature, high soil moisture, or large seasonality (Kaufman et al. 2004; Zona et al. 2009; Jones & Yu 2010). Accordingly, the Holocene temperature trend in Alaska is still inconclusive. Overall, these comparisons of multi-model simulations with proxy reconstructions further confirm the Holocene climate evolution patterns in Fennoscandia, Greenland and North
Canada. This implies that the Holocene temperatures in these regions have been relatively well established, with a reasonable representation of the Holocene climate in the multiple simulations and a plausible explanation for the underlying mechanisms. However, the Holocene climate history and underlying mechanisms in the regions of Siberia and Alaska remain inconclusive. 6.1.4 Vegetation dynamics and the main drivers Contributions of various factors to vegetation dynamics Variation partitioning analyses with climate, human population size and fire as explanatory variables were first performed in (C & S) Sweden and (SW & SE) Finland. The results reveal that over the study period as a whole (10–4 ka in Sweden, 10–1 ka in Finland), climate explains the highest proportion of variation in PFTs in all regions (Fig. 6.4). Winter temperatures have a significant effect on vegetation changes in all regions, while summer temperatures are significant in all regions except for C Sweden. In S and C Sweden, the relative importance of winter temperatures in PFT variation is higher than that of summer temperature, while in SW and SE Finland, summer temperatures explain a higher proportion of the variation (Table 5.2). Meanwhile, the variation explained by human population size and forest fires is relatively low, when considering the Holocene as a whole study period. Apart from the contribution of climate change, human population size has a significant contribution to vegetation changes in C Sweden and S Finland, whereas forest fires have a significant effect in S Sweden and SE Finland. Generally, climate, forest fires and human population size together explain over 75% of the variation in all these regions. Before the onset of farming (at 6 ka in Sweden and 4 ka in Finland), climate explains most of the variation in PFTs, with 13% in S Sweden, 54% in C Sweden, 25% in SW Finland and 32% in SE Finland. Although winter temperatures explain a higher proportion of variation in PFTs than summer temperatures, both climate parameters have significant effects on changes in vegetation dynamics in all regions (Fig. 6.4). Forest fires have a significant effect on vegetation dynamics in all regions except in C Sweden, and the highest contribution is found in SW and SE Finland, with up to 4%. The joint effects of climate and forest fires explain relatively a high proportion of the variation in vegetation, with 22% in S Sweden, 25% in SW Finland and 20% in SE Finland. The contribution of human population size to variation in vegetation is insignificant in these regions, except in SW Finland where 8% of the variation in vegetation can be explained by human population size. Overall, climate, forest fires and human population size explain at least 70% of the variation in vegetation composition in the studied regions.
Figure 6.4 Relative importance of climate, forest fires and human population size to vegetation dynamics. A, B, C present the results during the Holocene, pre-agricultural, and agricultural period; and from right to left representing S Sweden, C Sweden, SW Finland and SE Finland.
After the onset of farming, human population size explains the highest amount of the variation in PFTs, with 23% in south Sweden and 22% in southwest Finland (Fig. 6.4). The contribution of climate to the vegetation dynamics in all regions is lower than during other periods, even though summer temperatures explain a higher proportion of variation than winter temperatures (Table 2 of Chapter 5). Although the influence of forest fires on the variation in vegetation is minor, 10% of the variation in vegetation composition is explained by their joint effects with human population size in southeastern Finland. Overall, a relatively high proportion of the variation in vegetation is left unexplained after the onset of farming than in other periods in all regions. Transition of the main drivers of vegetation dynamics Variation partitioning reveals that a shift from natural (climate) driven vegetation dynamics towards human-dominated vegetation occurred during the Holocene. According to our study, Mesolithic populations did not significantly affect the vegetation dynamics in Fennoscandia, and climate was the main driver of change at that time. Neolithic populations, however, had major effects on vegetation dynamics and human population size became a more important driver of vegetation change than climate. A transition from
climate- to human-driven Holocene vegetation dynamics in Fennoscandia can be estimated to have occurred at 7–6 ka in Sweden and 4–3 ka in Finland. There is a clear regional dependency of changes caused by the human population. In particular, the impact of human population size is higher in south Sweden and southwest Finland, suggesting more intense land use in these regions. The low importance of human impacts in central Sweden and southeast Finland might be related to the higher forest cover in these regions and to the smaller scale and more scattered farming compared to S Sweden and SW Finland.
6.2 Remaining issues and Outlook This thesis investigates the climate responses to the main forcings and temperature trends over the course of the Holocene. As the most important forcing—the melting ice sheets— exerted spatial heterogeneity to climate before they finally disappeared, and the simulated climate was sensitive to different scenarios of the FWF forcing. This spatially heterogeneous response of climate implies that regionally different responses should be taken into account when studying the current global change. Moreover, our comparison of multiple simulations and with compiled proxy data suggestes that the multi-model and model-data agreements were regionally dependent. On the one hand, reasonably consistent temperatures were found over the several regions, including Greenland, N Canada and Fennoscandia where the climate is strongly influenced by the ice sheets. This implies that the Holocene temperatures in these regions have been relatively well established, with a reasonable representation of the Holocene climate in the multiple simulations and that our study offers a plausible explanation for the underlying mechanisms. On the other hand, large inter-model variation existed in the regions over which the ice sheet effects on the climate were relatively weak or the influences were indirect, such as in Alaska and E Siberia. This divergence implies that temperature history is still elusive in these regions. More important, that climate modelling results need to be validated by independent other models and proxy-based data, as these cross-validations can reduce the overestimation or underestimation in a specific model. For the context of the current global change, this suggests that apart from considering different scenarios of the GHG forcing, inter-model comparison would be a good option to reduce model-dependency in estimations of the future climate change. Additionally, statistical analyses revealed that vegetation dynamics in Fennoscandia was mainly driven by the climate change. However, the dynamics of human population became a more important driver of variations in vegetation composition than climate once farming practices systematically formed, indicated by a relatively stable and high population size. Although this thesis provides insights into the Holocene climate change as a whole, several issues remain unsolved and/or require additional research. Based on current understanding, including the results of this thesis, below are listed those issues considered as deserving more attention and further investigation. However, this list is not intended to be complete.
CHAPTER 6 (1) On the ice sheets and their influence on the climate (focus on the early Holocene) Dynamic ice sheets serve both causes and effects of climate evolution, with multiple glacial–interglacial cycles being experienced in geological history. On the one hand, the existence, building-up or decay of ice sheets are results of a changing climate. On the other hand, dynamic ice sheets exert multiple impacts on the climate system. The thesis investigated the complex response of the atmosphere–ocean system to melting ice sheets during the early Holocene, including their extent, topography and associated melting release with prescribed ice sheets based on ice-sheet reconstructions. In particular, responses of the climate–ocean system to different FWF scenarios were examined. However, reconstructions of decaying ice sheets are still uncertain to some extent, especially regarding the spatial-temporal distributions of FWF. Therefore, more evidence on decaying ice sheets during the early Holocene is still highly demanded to improve our understanding of dynamic ice sheets and their responses to the environment. Meanwhile, with increasing knowledge of the interaction between the atmosphere and ice sheets, the next step would be apply the fully coupled (to a dynamical ice-sheet component) iLOVECLIM climate model (Roche et al. 2014) for simulating glacial-interglacial cycles, which can be constrained by increasingly available ice-sheet reconstructions. This coupling may provide better understanding of the interactions between dynamic ice sheets and the environment. The final goal is to disentangle the buildup and decay of the ice sheets throughout the Earth’s history. (2) On inter-model comparisons This thesis revealed that multiple factors at different levels can lead to divergences between simulations performed with different models. Different parameterizations and representations of physics in models are important sources of such divergences, which can be accessed by comparing the simulated modern climate with observations. Meanwhile, differences in experimental setups and forcings hampered our inter-model comparisons. In particular, non-uniform FWF across simulations has a major impact on the divergent earlyHolocene climate (11.5–7 ka) among these simulations. Accordingly, we suggest that constructing a standardized FWF for the early Holocene would be advantageous for future inter-model comparisons. (3) On the model–data comparisons The model–data comparisons highlight that more work is required to reduce uncertainties from both simulation and proxy record aspects, especially in the case of the divergent temperature trends. Combinations of different proxies are expected to increase the accuracy of proxy-based reconstructions given the strengths and weaknesses of each type of proxy. However,
mismatched temperature reconstructions can be yielded by different types of proxies, which challenges the integration of multiple types of proxies into one reconstruction. Systematic comparisons of different types of proxies are valuable to efficiently combine different proxy datasets. Spatial consistency between the simulations and proxy data is another critical issue given the spatial heterogeneity of the climate. Proxy data are regularly site-based and thus represent relatively small-scale climate conditions, although various spatial scales can be represented by different types of proxies and even within the same proxy. For instance, pollen data can represent vegetation composition at various spatial scales from local (stand scale) to regional, depending on the openness of the landscape and the size of the sedimentary basin (Parshall & Calcote 2001; Sugita 2007). By contrast, the simulated temperatures in a model are the average over an area covered by grid cells, representing the large-scale climate. Therefore, maintaining a consistent spatial coverage has a crucial role in model–data comparison, despite the challenges in determining the exact representative scale of a given proxy dataset. (4) On further investigation of the transitional early-Holocene climate The highly dynamic characteristics of the early-Holocene climate and its general warming pattern provide a potential analogue for current global change. Apart from the methods employed in this thesis (model simulations in Chapter 2, inter-model comparisons in Chapter 3 and model–data comparisons in Chapter 4), the novel approach of dataassimilation would be another method that can be applied to investigate the Holocene climate. The data-assimilation approach makes use of proxy data in a statistical and dynamical way. Thus, compared with purely statistical approaches, data assimilation offers the potential advantage of ensuring dynamical consistency in the assessment of past climate change, despite its own limitations, such as the potential incompatibility between proxy data and model simulation due to the simplified physics of the model (Goosse et al. 2010b). This method has been applied to investigate model–data consistencies during the MH and also to Southern Hemisphere cooling from 10–8 ka (Mairesse et al. 2013; Mathiot et al. 2013). However, no study has investigated the NH extratropics climate for the whole Holocene yet. Increasingly available proxy data cover the whole Holocene provide the basis for applying data assimilation, which could ultimately improve knowledge of the early-Holocene climate. (5) Problematic Holocene temperatures in Alaska and Siberia According to this thesis, the Holocene temperature trends are still inconclusive in Alaska and high-latitude Siberia. In Alaska, multiple simulations are incompatible in winter: LOVECLIM suggests a decreasing Holocene temperature, while FAMOUS indicates an increasing trend. In summer, although multiple simulations consistently suggest a decreasing temperature trend, the pollen- and chironomid-based reconstructions are
CHAPTER 6 divergent. Additional peatland initialization data are available, but have multiple interpretations. Therefore, further investigation of the contribution of climate change to peatland initialization could provide evidence to narrow down the uncertainty of the Alaskan Holocene temperature. For high-latitude Siberia, a relatively wide spread was found in simulated temperatures in both summer and winter. Meanwhile, important issues exist in proxy-based reconstructions, such as the limited number of available records and the discrepancies between pollen-based and chironomid-based temperature reconstructions. Therefore, more proxy records are required to further examine their temperature evolution during the Holocene and differences between alternative types of proxy data. In addition, the early-Holocene climate in Beringia was influenced by various relevant changes, such as the flooding of the shelf and the connection between the Pacific and Arctic Oceans (Bartlein et al. 2015). Including the potential influence of these processes can improve the agreement between different models. (6) On the application of our simulations to investigate the dynamics of other parameters With confirmation of inter-model and model–data comparisons, our simulations can reasonably represent climate variations throughout the Holocene in most the NH extratropics. Simulated Holocene temperatures have the potential to be used in analyses of the contribution of climate change to other components/parameters, such as vegetation dynamics. However, other data (e.g. on forest fires and pollen data) are typically sitespecific records and represent the small-scale environment. Thus, the establishment of a downscaling procedure on our simulations would enhance its applicability in these relevant fields.
Appendix Appendix A: Abbreviations AMOC AOGCMs
Atlantic Meridional Overturing Circulation Atmosphere-ocean Global Climate Models
Community Climate System Model version 3
Climate Sensitivity to Radiative (Freshwater) forcing
Earth system Models of Intermediate Complexity
FAst Met Office/UK Universities Simulator
Fennoscandian Ice Sheet
Freshwater Flux (Meltwater flux) Greenhouse gases
Hadley Center climate model
Holocene Temperature Maximum
Last Glacial Maximum Last Interglacial
Laurentide Ice Sheet
Modern Analog Technique Mid-Holocene
North Atlantic Deep Water production
Northern Hemisphere Orbital
Plant Functional Types
Regional Estimates of VEgetation Abundance from Large Sites
Weighted Averaging Partial Least Square regression and
Appendix B: List of figures and tables (only include brief captions) Fig. 1.1 A schematic illustration of the components and interactions in the climate system (P. 20) Fig. 1.2 The Holocene period and early Holocene transition in the climate system (P. 24) Fig. 1.3 The climate components of the LOVECLIM model (P. 30) Fig. 1.4 Schematic illustration of approaches employed in the thesis (P. 33) Fig. 2.1 Evolution of greenhouse gas (GHG) concentrations and June insolation at 65°N derived from orbital configuration (P. 44) Fig. 2.2 The prescribed ice-sheet forcing during the early Holocene (P. 45) Fig. 2.3 Simulated temperatures for 11.5 kyr (P. 47) Fig. 2.4 Simulated temperature evolution since the early Holocene at high latitudes (P. 48) Fig. 2.5 Same as Fig. 2.4 but for northwestern Europe (P. 50) Fig. 2.6 Same as Fig. 2.4 but for northern Canada (P. 50) Fig. 2.7 Same as Fig. 2.4 but for Alaska (P. 51) Fig. 2.8 Same as Fig. 2.4 but for Siberia (P. 51) Fig. 2.9 Model–data comparison over the latitudinal band of 30–90°N (P. 54) Fig. 2.10 Geopotential height anomalies from global mean at 800 hPa in NH extratropics (P. 55) Fig. 2.11 Summer surface albedo in the NH extratropics (P. 55) Fig. 2.12 Meridional overturning stream function in the Atlantic Ocean basin (P. 56) Fig. 2.13 Minimum sea-ice thickness in September (P. 56) Fig. 3.1 Ice sheet related forcings during the early Holocene (P. 68) Fig. 3.2 Comparison of stacked proxy reconstruction with simulated summer (a) and annual mean temperature (b) in NH extratropics (P. 70) Fig. 3.3 Spatial distribution of simulated temperature anomalies in NH extratropics for the time windows of 11.5 ka (a) and 6 ka (b) (P. 71) Fig. 3.4 Temperature trends over the regions where multiple simulations have good agreements, and corresponding multi-model ensemble mean (P. 73) Fig. 3.5 Simulated temperatures over the regions where temperatures are less consistent across the simulations (P. 74) Fig. 3.6 Atmospheric circulation anomaly induced by the ice sheets at 11.5 ka and associated winds anomalies are indicated by the vectors (P. 76) Fig. 3.7 Distribution of maximum sea ice (P. 77) Fig. 3.8 Surface albedo anomalies at 11.5 ka compared to 0 ka (P. 81) Fig. 4.1 Map showing the locations of 61 proxy records and the domains of the five regions (P. 91) Fig. 4.2 Climate forcings in the simulations. (A) GHG forcings and summer (JJA) insolation at 65°N during the Holocene. (B) Change of ice sheet areas and FWF release into the oceans during the early Holocene (P. 95) Fig. 4.3 Comparisons of simulated summer (JJA) temperatures with pollen- and chironomid-based temperature reconstructions in Fennoscandia (P. 97)
Fig. 4.4 (A) Glacier frequency in Fennoscandia; (B) Composited geochemical proxy in Iceland; (C) Frequency of peatland initiation in Alaska; (D) simulated soil moisture in LOVECLIM (P. 98) 18 Fig. 4.5 Comparisons of Greenland temperatures between simulations and δ O-based reconstruction and borehole measurements at GRIP (P. 99) Fig. 4.6 Comparisons of simulated temperatures with pollen-based reconstruction in N Canada (P. 101) Fig. 4.7 Comparisons of simulated temperatures with pollen- and chironomid-based reconstructions in Alaska (P. 103) Fig. 4.8 Comparisons of simulated temperatures with pollen- and chironomid-based reconstructions in high-latitude Siberia (P. 105) Fig. 5.1 Location of the pollen and charcoal records and the study area (P. 115) Fig. 5.2 Holocene changes in the explanatory variables: 1) simulated summer and winter temperatures; 2) fire (charcoal influx z-scores) and 3) human population size data and in the plant functional types (PFTs) (P. 121) Fig. 5.3 Percentage of variation in vegetation during the Holocene explained by climate, fires and human population size in four regions (P. 123) Fig. 5.4 Same as the 5.3, but for the pre-agricultural period (P. 124) Fig. 5.5 Same as the 5.3, but for the agricultural period (P. 124) Fig. 5.6 Relative importance of climate and anthropogenic influence on variation in vegetation dynamics through the Holocene in northern Europe (P. 136) Fig. 6.1 (A) Simulated temperature at 11.5 ka, (B) Holocene temperature evolution in simulations performed with different forcings (P. 137) Fig. 6.2 (A) Multi-model ensemble mean of simulated temperature trends over the regions where multiple simulations are broadly consistent. (B) Simulated temperatures over the regions where temperatures are inconsistent across the simulations (P. 139) Fig. 6.3 Comparisons of simulated temperature with proxy-based reconstructions (P. 142) Fig. 6.4 Relative importance of climate, forest fires and human population size to the variation in vegetation (P. 146) Table 1.1 Inter-model and model–data comparison studies included in PMIP (P. 27) Table 2.1 Boundary conditions for 11.5 kyr and the preindustrial (PI) era (P. 46) Table 2.2 Experiments and corresponding setup (P. 46) Table 3.1 Summary of participated climate model (P. 66) Table 3.2 Main features of the setup of involved simulations (P. 66) Table 4.1 Summary of involved climate models and simulations (P. 94) Table 5.1 Correspondence between the pollen taxa and the plant functional types (P. 110) Table 5.2 PFTs variation explained individually by summer and winter temperatures (P. 123) Fig. S2.1 Monthly insolation anomaly at 11.5 ky compared to 0 ky Fig. S2.2 Selected region denoted by the box over where the temperature evolution is shown
APPENDIX Fig. S2.3 Simulated temperatures for 10 ky Fig. S3.1 Orbital-scale insolation and GHG related radiative forcing during the Holocene Fig. S3.2 The eight selected regions are marked as boxes Fig. S3.3 Atmospheric circulations changes induced by the topography of the LIS. Fig. S3.4 Total area of sea ice in NH Fig. S3.5 Meridional overturning streamfunction of the Atlantic Basin in the transient simulations Fig. S3.6 Changes of maximum AMOC over the course of the Holocene Fig. S3.7 Albedo changes over the course of the Holocene Fig. S4.1 Boxplot of reconstructed temperature in Greenland based on 6 δ18O-based temperature records. Fig. S5.1 Summed probability distributions (SPD) of archaeological radiocarbon dates and the results of Monte Carlo tests Fig. S5.2 Results of variation partitioning for 2000 year time windows moving in 200 year time slices for S Sweden and C Sweden Fig. S5.3 Same as Fig. S5.2, but for SW Finland and SE Finland Table S4.1 Site information on proxy records used in the present study Table S5.1a Moving window results for S Sweden and SE Finland for ten 200 year subsets of data in 2000-year mowing time windows. Table S5.1b Moving window results for Estonia 1200-yr moving time window in Estonia Table S5.1c Fraction of variation explained by climate and land use average across S Fennoscandia and Baltic countries
Appendix C: Supplementary information
Figure S2.1 Monthly insolation anomaly (W m-2) at 11.5 ky compared to 0ky derived from Berger (1978).
Figure S2.2 Selected region denoted by the box over where the temperature evolution is shown. Shown in the background is the simulated annual mean temperature at the onset of the Holocene.
Figure S2.3 Simulated temperatures for 10 ky (shown as the deviation of 100-yr average from PI). Left, middle and right columns show the simulations ORBGHG, OGIS_FWF-v1 and OGIS_FWF-v2, respectively. Upper, middle and lower panel indicate summer, winter and annual temperatures, respectively.
Figure S3.1 Orbital-scale insolation and GHG related radiative forcing (in W m-2) during the Holocene.
Figure S3.2 The eight selected regions are marked as boxes. The background color indicates the simulated annual mean temperature in LOVECLIM at 11.5 ka.
Figure S3.3 Atmospheric circulations changes induced by the topography of the LIS, shown as the differences between the LIS and N Pacific of geopotential height. Given to different vertical layers among models, the results are standardized by calculating the anomalies and percentage regarding 0 ka condition. The results are shown as 100-yr average in LOVECLIM and 1 kyr interval in FAMOUS and HadCM3.
Figure S3.4 Total area of sea ice in NH (in 1012 m2)
Figures S3.5 Meridional overturning streamfunction (in Sv) of the Atlantic Basin in the transient simulations
Figure S3.6 Changes of maximum AMOC (in the box of 500-2000 m, 34S-50N, according to the definition of Hofer et al 2011; Drijfhout et al 2012) over the course of the Holocene. Results are shown as 100-yr averages.
Figure S3.7 Albedo changes over the course of the Holocene.
Figure S4.1 Boxplot of reconstructed temperature in Greenland based on 6 δ18O-based temperature records. Segments inside the rectangles show the median and whiskers above and below the box indicate the minimum and maximum of all records. Open dots give the suspected outliers that are either 1.5×IQR (the interquartile range) above the up quartile, or 1.5×IQR below the low quartile.
Table S4.1 Site information on proxy records used in the present study No Name
Oldest yr. BP
Fréchette&de Vernal 2009
Seppä et al. 2009
Ilyashuk et al. 2005
Seppä et al. 2009
Velle et al. 2005
δ18O paleoVinther et al. 2009 thermometer
Bjune et al. 2004
Miller et al 2005
Wolfe et al. 2000
RESP&MAT Kerwin et al. 2004
δ18O paleoVinther et al. 2006 thermometer
15 Dyer Lower
RESP&MAT Kerwin et al. 2004
Antonsson et al. 2006
paleoAlley 2000 thermometer
δ18O paleoVinther et al. 2006 thermometer
Eide et al. 2006; Seppä et al. 2009
Kerwin et al. 2004
Kerwin et al. 2004
Fallu et al. 2005
Jones et al. 2011; Salonen et al. 2011
Seppä et al. 2008
APPENDIX 30 Lily
Gajewski et al 1993
Seppä et al. 2009
Andreev et al. 2005
Seppä et al. 2009
36 Nain Pond
RESP&MAT Kerwin et al. 2004
δ18O paleo- Vinther et al. 2006; thermometer NGRIP members 2004
Bigler et al. 2003
Bjune et al. 2005; Velle et al. 2005
Wooller et al. 2012
Clegg et al. 2011
Velle et al. 2005
paleoVinther et al. 2009 thermometer
Clegg et al. 2011
Hammarlund et al. 2004; Velle et al. 2005
Nazarova et al 2013
Hammarlund et al. 2002
Seppä&Birks 2002; Seppä et al. 2002
Irvine et al. 2012
Seppä&Birks 2001, Seppä et al. 2009
Bigler et al. 2002
Seppä et al. 2008
Method abbreviations: MAT: modern analog technique; WAPLS: weight average partial least squares; RESP: response surface. The δ18O sensitivity value to temperature in paleo-therometer is based on Cuffey et al 1995.
Figure S5.1 Summed probability distributions (SPD) of archaeological radiocarbon dates and the results of Monte Carlo tests. A: Southern Sweden. B: Central Sweden. C: Southwestern Finland. D: Southeastern Finland. The thick lines show the 200-years rolling mean, whilst the grey polygon represents the 95% confidence interval for the exponential null model. Red and blue vertical bands represent intervals with significant positive and negative deviations from the null model. Ntot is the total number radiocarbon dates in a sample. Nbin is the number of dates after binning to 200 radiocarbon year bins. SPDs are based on these binned dates. The figure also shows the p-value of the global significance test, which indicates whether the observed SPD differs significantly from the null model.
Figure S5.2 Results for variation partitioning for 2000 year time windows moving in 200 year time slices. Fraction of the variation in vegetation dynamics (PFTs) explained by climate and human population size in a.) S Sweden and b.) C Sweden.
Figure S5.3 Results for variation partitioning for 2000 year time windows moving in 200 year time slices. Fraction of the variation in vegetation dynamics (PFTs) explained by climate and human population size in a.) SW Finland and b.) SE Finland.
Table S5.1a Moving window results for S Sweden and S Finland for ten 200 year subsets
of data in 2000-year mowing time windows. Panel A. Time perid
Southern Sweden Midpoint used in fig. 6 Human pop. Size (%)
APPENDIX Table S5.1b Moving window results for Estonia 1200-year moving time window in Estonia (Reitalu et al. 2013). Panel B. Time period 1250-50 1550-350 1850-650 2150-950 2450-1250 2750-1550 3050-1850 3350-2150 3650-2450 3950-2750 4250-3050 4550-3350 4850-3650 5150-3950
Estonia Midpoint used in fig. 6 Human impact (%) 650 10.46 950 7.11 1250 5.68 1550 6.67 2150 7.58 2450 7.60 2750 11.80 3050 11.62 3350 13.65 3650 14.37 3950 8.58 4250 7.36 4550 3.44 4850 0.15
Bibliography Abe-Ouchi A, Saito F, Kawamura K, Raymo ME, Okuno J, Takahashi K, Blatter H, 2013. Insolation-driven 100,000-year glaial cycles and hysteresis of ice-sheet volume. Nature 500, 190–193. Alenius T, Laakso V, 2006. Palaeoecology and archaeology of the village of Uukunniemi, Eastern Finland. ActaBorealia: A Nordic Journal of Circumpolar Societies 23, 145–16. Alenius T, Mikkola E, Ojala AEK, 2008. History of agriculture in Mikkeli Orijärvi, eastern Finland as reflected by palynological and archaeological data. Veg Hist Archaeobot 17, 171–183. Alenius T, Mönkkönen T, Lahelma A, 2013. Early farming in the northern boreal zone: Reassessing the history of land use in southeastern Finland through high-resolution pollen analysis. Geoarchaeology 28, 1–24. Attenbrow V, Hiscock P, 2015. Dates and demography: are radiometric dates a robust proxy for long-term prehistoric demographic change? Archaeology in Oceania 50, 30–36. Bakker P, Masson-Delmotte V, Martrat B et al., 2014. Temperature trends during the present and last interglacial periods—a multi-model–data comparison. Quat Sci Rev 99, 224–243. Barlow LK, Rogers JC, Serreze MC, Barry RG, 1997. Aspects of climate variability in the North Atlantic sector: discussion and relation to the Greenland Ice Sheet Project 2 high-resolution isotopic signal. J Geophys Res 102, 26,333–26. Bartlein PJ, Edwards ME, Hostetler WS, Shafer SL, Anderson PM, Brubaker LB, Lozhkin AV, 2015. EarlyHolocene warming in Beringia and its mediation by sea-level and vegetation changes. Clim Past 11, 1197–1222. Behre KE, 1988. The role of man in European vegetation history. Vegetation History (ed. 850 by B. Huntley and T. Webb), pp. 633–672. Dordrecht, Netherlands. Berger A, 1978. Long-term variations of caloric insolation resulting from the Earths orbital elements. Quat Res 9, 139–167. Berger A, 1988. Milankovitch Theory and Climate. Rev Geophys 26, 624–657. Berger A, Loutre MF, 1991. Insolation values for the climate of the last 10 million years. Quat Sci Rev 10, 297–317. Berglund BE, Persson T, Björkman L, 2008. Late Quaternary landscape and vegetation diversity in a North European perspective. Quat Int 184, 187–194. Berner KS, Koç N, Divine D, Godtliebsen F, Moros M, 2008. A decadal-scale Holocene sea surface temperature record from the subpolar North Atlantic constructed using diatoms and statistics and its relation to other climate parameters. Paleoceanography 23, PA2210. Bigelow NH, Brubaker LB, Edwards ME et al., 2003. Climate change and Arctic ecosystems: Vegetation changes north of 55°N between the last glacial maximum, mid-Holocene, and present, J Geophys Res 108(D19), 8170. Bintanja R, van de Wal RSW, Oerlemans J, 2002. Global ice volume variations through the last glacial cycle simulated by a 3-D ice-dynamics model. Quat Int 95–96, 11–23. Birks HH, 2015. South to north: Contrasting late-glacial and early-Holocene climate changes and vegetation responses between south and north Norway. Holocene 25, 37–52. Birks HJB, 1986a. Numerical zonation, comparison and correlation of Quaternary pollen-stratigraphical data. In: Berglund, B.E., Ralska-Jasiewiczowa, M. (Eds.), Handbook of the Holocene Palaeoecology and Palaeohydrology. Wiley & Sons, Chichester, pp. 743–774.
BIBLIOGRAPHY Birks HJB, 1986b. Late-Quaternary biotic changes in terrestrial and lacustrine environments, with particular reference to north-west Europe. In Berglund, BE. (ed) Handbook of Holocene palaeoecology and palaeohydrology. Wiley, Chichester, pp 3–65. Birks HJB, Heiri O, Seppä H, Bjune, AE, 2010. Strengths and weaknesses of quantitative climate reconstructions based on late-Quaternary biological proxies. The Open Ecology Journal 3, 68–110. Biskaborn BK, Subetto DA, Savelieva LA, Vakhrameeva PS, Hansche A, Herzschuh U, Klemm J, Heinecke L, Pestryakova LA, Meyer H, Kuhn G, Diekmann B, 2016. Late Quaternary vegetation and lake system dynamics in north-eastern Siberia: Implications for seasonal climate variability. Quat Sci Rev 147, 406– 421. Blarquez O, Vanniére B, Marlon JR, Daniau AL, Power MJ, Brewer S, Bartlein PJ, 2014. Paleofire: an R package to analyse sedimentary charcoal records from the Global Charcoal Database to reconstruct past biomass burning. Comput Geosci 72, 255–261. Blaschek M, Renssen H, 2013. The Holocene thermal maximum in the Nordic Seas: the impact of Greenland Ice Sheet melt and other forcings in a coupled atmosphere–sea-ice–ocean model. Clim Past 9, 1629– 1643. Boer GJ, Yu B, 2003. Climate sensitivity and climate state. Clim Dyn 21, 167–176. Bond G, Broecker W, Johnsen S, McManus J, Labeyrie L, Jouzel J, Bonani G, 1993. Correlations between climate records from North Atlantic sediments and Greenland ice, Nature, 365, 143–147. Bonfils C, de Noblet-Ducoudr N, Guiot J, Bartlein P, 2004. Some mechanisms of mid-Holocene climate change in Europe, inferred from comparing PMIP models to data. Clim Dyn 23, 79–98. Borcard D, Legendre P, Drapeau P, 1992. Partialling out the spatial component of ecological variation. Ecology 73, 1045–1055. Bothe O, Jungclaus JH, Zanchettin D, 2013. Consistency of the multi-model CMIP5/PMIP3-past1000 ensemble. Clim Past 9, 2471–2487. Boyle EA, Keigwin LD, 1985. Comparison of Atlantic and Pacific paleochemical records for the last 215,000 years: Changes in deep ocean circulation and chemical inventories. Earth Planet Sci Lett 76, 135–150. Braconnot P, Harrison SP, Kageyama M, Bartlein PJ, Masson-Delmotte V, Abe-Ouchi A, Otto-Bliesner B, Zhao Y, 2012. Evaluation of climate models using palaeoclimatic data. Nat Clim Change 2, 417–424. Braconnot P, Joussaume S, de Noblet N, Ramstein G, 2000. Mid-Holocene and Last Glacial Maximum African monsoon changes as simulated within the Paleoclimate Modelling Inter-comparison Project. Planet Change 26, 51–66. Braconnot P, Otto-Bleisner B, Harrison SP, Joussaume S, Peterschmitt JY, Abe-Ouchi A, Crucifix M, Driesschaert E, Fichefet Th,Hewitt CD, Kagayama M, Kitoh A, Loutre MF, Marti O,Merkel U, Ramstein G, Valdes P, Weber L, Yu Y, Zhao Y, 2007. Results of PMIP2 coupled simulations of the mid-Holocene and Last Glacial maximum, part 1: experiments and large-scale features. Clim Past 3, 261–277. Brewer S, Guiot J, Torre F, 2007. Mid-Holocene climate change in Europe: a data–model comparison. Clim Past 3, 499–512. Briner JP, McKay NP, Axford Y, Bennike O, Bradley RS, de Vernal A, Fisher D, Francus P, Fréchette B, Gajewski K, Jennings A, Kaufman DS, Miller G, Rouston C, Wagner B, 2016. Holocene climate change in Arctic Canada and Greenland. Quat Sci Rev 147, 340–364. Bromwich DH, Toracinta ER, Oglesby RJ, Fastook J, Hughes TJ, 2005. LGM summer climate on the southern margin of the Laurentide ice sheet: wet or dry? J Climate 18, 3317–3338. Bromwich DH, Toracinta ER, Wei H, Oglesby RJ, Fastook JL, Hughes TJ, 2004. Polar MM5 simulations of the winter climate of the Laurentide Ice Sheet at the LGM. J Climate 17, 3415–3433. Brooks SJ, Axford Y, Heiri O, Langdon PG, Larocque-Tobler I, 2012. Chironomids can be reliable proxies for
Holocene temperatures. A comment on Velle et al. (2010). Holocene 22, 1495–1500. Brooks SJ, Birks HJB, 2000. Chironomid-inferred Late-glacial and early Holocene mean July air temperatures for Krakenes Lake, western Norway. J Paleolimnol 23, 77–89. Brooks SJ, Birks HJB, 2001. Chironomid-inferred air temperatures from Lateglacial and Holocene sites in north-west Europe: progress and problems. Quat Sci Rev 20, 1723–1741. Brooks SJ, Matthews IP, Birks HH, Birks HJB, 2012. High resolution Lateglacial and early-Holocene summer air temperature records from Scotland inferred from chironomid assemblages. Quat Sci Re. 41, 67–82. Brovkin V, Ganopolski A, Svirezhev Y, 1997. A continous climate-vegetation classification for use in climatebiosphere studies. Ecol Model 101, 251–261. Brown WA, 2015. Through a filter, darkly: population size estimation, systematic error, and random error in radiocarbon-supported demographic temporal frequency analysis. J Archaeol Sci 53, 133–147. Bryan F, Danabasoglu G, Nakashiki N, Yoshida Y, Kim D, Tsutsui J, Doney S, 2006. Response of the North Atlantic Thermohaline circulation and ventilation to increasing carbon dioxide in CCSM3. J Climate 19, 2382–2397. Buizert C, Gkinis V, Severinghaus JP, He F, Lecavalier BS, Kindler P, Leuenberger M, Carlson AE, Vinther B, Masson-Delmotte V, White JWC, Liu Z, Otto-Bliesner B, Brook EJ, 2014. Greenland temperature response to climate forcing during the last deglaciation. Science, 345, 1177–1180. Came RE, Oppo DW, McManus JF, 2007. Amplitude and timing of temperature and salinity variability in the subpolar North Atlantic over the past 10 ky. Geology 35, 315–318. CAPE Members, 2006. Last Interglacial Arctic warmth confirms polar amplification of climate change. Quat Sci Rev 25, 1383–1400. CAPE Members, 2001. Holocene paleoclimate data from the arctic: Testing models of global climate change, Quat Sci Rev 20, 1275–1287 Carlson A.E., Clark P.U., Haley B.A., Klinkhammer G.P., Simmons K., Brook E.J., Meissner K.J., Geochemical proxies of North American freshwater routing during the Younger Dryas cold event. Proc Natl Acad Sci USA 104, 2007, 6556–6561. doi:10.1073/pnas.0611313104 Carlson AE, Winsor K, Ullman DJ, Brook E, Rood DH, Axford Y, Le Grande AN, Anslow F, Sinclair G, 2014. Earliest Holocene south Greenland ice-sheet retreat within its late-Holocene extent. Geophys Res Lett 41, 5514–521. Cattle H, Crossley JF, 1995. Modelling Arctic climate change. Philos Trans R Soc London Ser A 352,201–213. Charles CD, Rind D, Jouzel J, Koster RD, Fairbanks RG, 1994. Glacial–interglacial changes in moisture sources for Greenland: Influences on the ice core record of climate. Science 263, 508–511. Claquin T, Roelandt C, Kohfeld KE, Harrison SP, Tegen I, Prentice IC, Balkanski Y, Bergametti G, Hansson M, Mahowald N, Rodhe H, Schulz M, 2003. Radiative forcing of climate by ice-age atmospheric dust. Clim Dyn 20, 193–202. Claussen M, Brovkin V, Ganopolski A, 2001. Biogeophysical versus biogeochemical feedbacks of large-scale land cover change. Geophys Res Lett 28, 1011–1014. Clear JL, Molinari C, Bradshaw RHW, 2014. Holocene fire in Fennoscandia and Denmark. Int J Wildland Fire 23, 781–789. Collins M, Tett SFB, Cooper C, 2001. The internal climate variability of HadCM3, a version of the Hadley Centre coupled model without flux adjustments. Clim Dyn 17, 61–81. Collins WD, Bitz CM, Blackmon ML et al., 2006. The Community Climate System Model Version 3 (CCSM3). J Climate 19, 2122–2243. Coumou D, Rahmstorf S, 2012. A decade of weather extremes. Nat Climate Change 2, 491–496. Cox P, Betts R, Bunton C, Essery R, Rowntree P, Smith J, 1999. The impact of new land surface physics on the
BIBLIOGRAPHY GCM simulation of climate and climate sensitivity. Clim Dyn 15,183–203. Crema ER, Habu J, Kobayashi K, Madella M, 2016. Summed Probability Distribution of 14C Dates Suggests Regional Divergences in the Population Dynamics of the Jomon Period in Eastern Japan. PLOS ONE 11. doi:10.1371/journal.pone.0154809 Crombé, P., Robinson, E., 2014. 14C dates as demographic proxies in Neolithisation models of northwestern Europe: a critical assessment using Belgium and northeast France as a case-study. J Archaeol Sci 52, 558– 566. Cuffey KM, Clow GD, Alley RB, Stuiver M, Waddington ED, Saltus RW, 1995. Large Arctic Temperature Change at the Wisconsin-Holocene Glacial Transition. Science 270, 455–458. Curry JA, Schramm JL, Ebert EE, 1995. Sea ice-albedo climate feedback mechanism. J. Climate 8, 240–247. Cuzzone JK, Clark PU, Carlson AE, Ullman DJ, Rinterknecht VR, Milne GA, Lunkka JP, Wohlfarth B, Marcott SA, Caffee M, 2016. Final deglaciation of the Scandinavian Ice Sheet and implications for the Holocene global sea-level budget. Earth Planet Sci Lett 448, 34–41. Dahl E, 1998. The Phytogeography of Northern Europe. Cambridge University Press, Cambridge. Dahl-Jensen D, Mosegaard K, Gundestrup N, Clow GD, Johnsen S J, Hansen AW, Balling N, 1998. Past Temperatures Directly from the Greenland Ice Sheet. Science 282, 268–271. Daniau AL et al. 2012. Predictability of biomass burning in response to climate changes. Global Biogeochem Cy 26, GB4007. Dansgaard W, Johnsen SJ, Clausen HB, Dahl-Jensen D, Gundestrup NS, Hammer CU, Hvidberg CS, Steffensen JP, Sveinbjornsdottir AE, Jouzel J, Bond G, 1993. Evidence for general instability of past climate from a 250 kyr ice-core record. Nature 364, 218–220. Davis BAS, Collins PM, Kaplan JO, 2015. The age and post-glacial development of the modern European vegetation: a plant functional approach based on pollen data. Veg Hist Archaeobot 24, 303–317. de Vernal A, Hillaire-Marcel C, Rochon A, Fréchette B, Henry M, Solignac S, Bonnet S, 2013. Dinocyst-based reconstructions of sea ice cover concentration during the Holocene in the Arctic Ocean, the northern North Atlantic Ocean and its adjacent seas. Quat Sci Rev 79,111–121. Deblonde G, Peltier WR, Hyde WT, 1992. Simulations of continental ice sheet growth over the last glacial– interglacial cycle: experiments with a one level seasonal energy balance model including seasonal ice albedo feedback. Global Planet Change 98, 37–55. Denton GH, Anderson RF, Toggweiler JR, Edwards RL, Schaefer JM, Putnam AE, 2010. The last glacial termination. Science 328, 1652–1656. Downey SS, Bocaege E, Kerig T, Edinborough K, Shennan S, 2014. The Neolithic Demographic Transition in Europe: Correlation with Juvenility Index Supports Interpretation of the Summed Calibrated Radiocarbon Date Probability Distribution (SCDPD) as a Valid Demographic Proxy. PLoS ONE 9. doi:10.1371/journal.pone.0105730 Drijfhout S, van Oldenborgh GJ, Cimatoribus A, 2012. Is a decline of AMOC causing the warming hole above the North Atlantic in observed and modeled warming patterns? J Climate 25(24), 8373–8379. Dyke AS, Moore A, Robertson L, 2003. Deglaciation of North America, Open-file report-geological survey of Canada, Geological Survey of Canada, Ottaw. Eby M, Weaver AJ, Alexander K et al., 2013. Historical and idealized climate model experiments: an intercomparison of Earth system models of intermediate complexity. Clim Past 9, 1111–1140. Ehlers J, 1990. Reconstructing the dynamics of the north-west European Pleistocene ice sheets. Quat Sci Rev 9, 71–83. Eldevik T, Risebrobakken B, Bjune AE, Andersson C, Birks HJB, Dokken TM, Drange H, Glessmer MS, Li C, Nilsen JEØ, Otterå OH, Richter K, Skagseth Ø, 2014. A brief history of climate the northern seas from the Last Glacial Maximum to global warming. Quat Sci Rev 106, 225–246.
EPICA community members, 2004. Eight glacial cycles from an Antarctic ice core. Nature 429, 623-628. Ettema J, van den Broeke MR, van Meijgaard E, van de Berg WJ, Bamber JL, Box JE, Bales RC, 2009. Higher surface mass balance of the Greenland ice sheet revealed by high-resolution climate modeling. Geophys Res Lett 36, L12501. Fairbanks RG, 1989. A 17 000-year glacio-eustatic sea level record: influence of glacial melting dates on the Younger Dryas event and deep ocean circulation. Nature 342, 637–642. Fallu MA, Pienitz R, Walker IR, Lavoie M, 2005. Paleolimnology of a shrub-tundra lake and response of aquatic and terrestrial indicators to climatic change in arctic Quebec, Canada. Palaeogeog Palaeoclim Palaeoecol 215, 183–203. Fang K, Morris JL, Salonen JS, Miller PA, Renssen H, Sykes MT, Seppä H, 2013. How robust are Holocene treeline simulations? A model–data comparison in the European Arctic treeline region. J Quat Sci 28, 595–604. Felzer B, Oglesby RJ, Webb T, Hyman DE, 1996. Sensitivity of a general circulation model to changes in northern hemisphere ice sheets. J Geophys Res-Atmos 101, 19077–19092. Fichefet T, Maqueda MAM, 1997. Sensitivity of a global sea ice model to the treatment of ice thermodynamics and dynamics. J Geophys Res-Oceans 102, 12609–12646. Fisher DA, Koerner RM, Bourgeois JC, Zielinski G, Wake C, Hammer CU, Clausen HB, Gundestrup N, Johnsen S, Goto-Azuma K, Hondoh T, Blake E, Gerasimoff M, 1998. Penny Ice Cap cores, Baffin Island, Canada, and the Wisconsinan Foxe Dome connection: two states of Hudson Bay ice cover. Science 279, 692–695. Fisher D, Koerner R, Paterson W, Dansgaard W, Gundestrup , Reeh N, 1983, Effect of wind scouring on climate records from ice-core oxygen-isotope profiles, Nature 301, 205–209. Fisher DA, Koerner, RM, Reeh N, 1995. Holocene climatic records from Agassiz Ice Cap, Ellesmere Island, NWT, Canada. Holocene 5, 19–24. Flato G et al., 2013. Evaluation of Climate Models, in: Climate Change 2013: The Physical Science Basis. Contribution of Working Group I to the Fifth Assessment Report of the Intergovernmental Panel on Climate Change, Stocker TF, Qin D, Plattner GK, Tignor M, Allen SK, Boschung J, Nauels A, Xia Y, Bex V, Midgley PM (eds), Cambridge University Press, Cambridge, UK and New York, NY, USA. French JC, Collins C, 2015. Upper Palaeolithic population histories of Southwestern France: a comparison of the demographic signatures of 14C date distributions and archaeological site counts. J Archaeol Sci 55, 122–134. Friedrich T, Timmermann A, Menviel L, Elison Timm O, Mouchet A, Roche D, 2010. The mechanism behind internally generated centennial-to-millennial scale climate variability in an earth system model of intermediate complexity. Geosci Model Dev 3(2), 377–389. Fyfe RM, Twiddle C, Sugita S et al., 2013. The Holocene vegetation cover of Britain and Ireland: overcoming problems of scale and discerning patterns of openness. Quat Sci Revi 73, 132–148. Fyfe RM, Woodbridge J, Roberts N, 2015. From forest to farmland: pollen-inferred land cover change across Europe using the pseudobiomization approach. Glob Change Biol 21, 1197–1212. Gaillard MJ, SugitaS, Mazier F, Trondman AK, Brostrom A et al., 2010. Holocene land-cover reconstructions for studies on land cover-climate feedbacks. Clim Past 6, 483–499. Gamble C, Davies W, Pettitt P, Hazelwood L, Richards M, 2005. TheArchaeological and Genetic Foundations of the European Population during the Late Glacial: implications for ‘Agricultural Thinking’. Camb Archaeol J 15, 193–223. Ganachaud A, Wunsch C, 2000. Improved estimates of global ocean circulation, heat transport and mixing from hydrographic data. Nature 408, 453–457. Ganopolski A, Calov R, Claussen M, 2010. Simulation of the last glacial cycle with a coupled climate ice-sheet model of intermediate complexity. Clim Past 6, 229–244.
BIBLIOGRAPHY Geirsdóttir A, Miller GH, Larsen DJ, Ólafsdóttir S, 2013. Abrupt Holocene climate transitions in the northern North Atlantic region recorded by synchronized lacustrine records in Iceland. Quat Sci Rev 70, 48–62. Gent PR, McWilliams JC, 1990 Isopycnal mixing in ocean circulation models. J Phys Oceanogr 20, 150–155. Genthon C, Jozel J, Déqué M, 1994. Accumulation at the surface of polar ice sheets: observation and modelling for global climate change. NATO ASI Series 126, 53–75. Giesecke T, Basil Davis B, Simon Brewer S, Finsinger W, Wolters S, Blaauw M, de Beaulieu JL, Binney H, Fyfe RM, Gaillard MJ, Gil-Romera G, van der Knaap WO, Kuneš P, Kühl N, van Leeuwen JFN, Leydet M, Lotter AF, Ortu E, Semmler M, Bradshaw RHW, 2014. Towards mapping the late Quaternary vegetation change of Europe. Veg Hist Archaeobot 23, 75–86. Gladstone RM, Ross I, Valdes PJ, Abe-Ouchi A, Braconnot P, Brewer S, Kageyama M, Kitoh A, Legrande A, Marti O, Ohgaito R, Otto-Bliesner B, Peltier WR, Vettoretti G, 2005. Mid-Holocene NAO: A PMIP2 model intercomparison. Geophys Res Lett 32, L16707. Golledge NR, Fogwill CJ, Mackintosh AN, Buckley KM, 2012. Dynamics of the last glacial maximum Antarctic ice-sheet and its response to ocean forcing. Proc Natl Acad Sci USA 109, 16052–16056. Goosse H, Brovkin V, Fichefet T, Haarsma R et al., 2010a. Description of the Earth system model of intermediate complexity LOVECLIM version 1.2. Geosci Model Dev 3, 603–633. Goosse H, Crespin E, de Montety A, Mann ME, Renssen H, Timmermann A, 2010b. Reconstructing surface temperature changes over the past 600 years using climate model simulations with data assimilation. J Geophys Res 115, D09108. Goosse H, Fichefet T, 1999. Importance of ice-ocean interactions for the global ocean circulation: A model study. J Geophys Res-Oceans 104, 23337–23355. Gordon C, Cooper CA, Senior CA, Banks H, Gregory JM, Johns TC, Mitchell JFB , Wood RA, 2000. The simulation of SST, sea ice extents and ocean heat transports in a version of the Hadley Centre coupled model without flux adjustments. Clim Dyn 16, 147–168. Granström A, 2001. Fire management for biodiversity in the European boreal forest, Scand. J Forest Res 16 (3), 62–69. Gregory JM, Dixon KW, Stouffer RJ, Weaver AJ, Driesschaert E, Eby M, Fichefet T, Hasumi H, Hu A, Jungclaus JH, Kamenkovich IV, Levermann A, Montoya M, Murakami S, Nawrath S, Oka A, Sokolov AP, Thorpe RB, 2005. A model intercomparison of changes in the Atlantic thermohaline circulation in response to increasing atmospheric CO2 concentration. Geophys Res Lett 32, L12703. Grootes PM, Stuiver M, White JWC, Johnsen S, Jouzel J, 1993. Comparison of oxygen isotope records from the GISP2 and GRIP Greenland ice cores, Nature, 366, 552–554. Gupta AK, 2004. Origin of agriculture and domestication of plants and animals linked to early Holocene climate amelioration. Current Science 87, 54–59. Hald M, Andersson C, Ebbesen H, Jansen E, Klitgaard-Kristensen D, Risebrobakken B, Salomonsen GR, Sarnthein M, Sejrup HP, Telford RJ, 2007. Variations in temperature and extent of Atlantic Water in the northern North Atlantic during the Holocene. Quat Sci Rev 26, 3423–3440. Hallegatte S, 2009. Strategies to adapt to an uncertain climate change. Global Environ Change 19, 240–247. Halme P, Allen KA, Auniᶇš A et al., 2013. Challenges of ecological restoration: lessons from forests in northern Europe. Biol Conserv 167, 248–256. Hargreaves JC, Annan JD, Ohgaito R, Paul A, Abe-Ouchi A, 2013. Skill and reliability of climate model ensembles at the Last Glacial Maximum and mid-Holocene. Clim Past 9, 811–823. Harrison SP et al., 1998. Intercomparison of simulated global vegetation distributions in response to 6 kyr BP orbital forcing. J Climate 11, 2721–2742. Hawkins E, Sutton R, 2009. The potential to narrow uncertainty in regional climate predictions. Bull Am
Meteorol Soc 90, 1095. He F, 2011. Simulating transient climate evolution of the last deglaciation with CCSM3. Dissertation, The university of Wisconsin-Madison. Heiri O, Brooks SJ, Renssen H, Bedford A et al., 2014. Validation of climate model-inferred regional temperature change for late-glacial Europe. Nat Commun 5, 4914. Hewitt CD, Mitchell FB, 1997. Radiative forcing and response of a GCM to ice age boundary conditions: cloud feedback and climate sensitivity. Clim Dyn 13, 821–824. Hofer D, Raible CC, Merz N, Dehnert A, Kuhlemann J, 2012. Simulated winter circulation types in the North Atlantic and European region for preindustrial and glacial conditions. Geophysical Research Letters 39. Hofer D, Raible CC, Stocker TF, 2011. Variations of the Atlantic meridional overturning circulation in control and transient simulations of the last millennium. Clim Past 7, 133–150. Holland MM, Bitz CM, Hunke EC, Lipscomb WH, Schramm JL, 2006. Influence of the sea ice thickness distribution on polar climate in CCSM3. J Climate 19, 2398–2414. Holst D, 2010. Hazelnut economy of early Holocene hunter-gatherers: a case study from Mesolithic Duvensee, northern Germany. J Archaeol Sci 37, 2871–2880 Hörnberg G, Bohlin E, Hellberg E, Bergman I, Zackrisson O, Olofsson A, Wallin JE, Påsse T, 2005. Effects of Mesolithic hunter-gatherers on local vegetation in a non-uniform glacio-isostatic land uplift area, northern Sweden. Veg Hist Archaeobot 15, 13–26. Hu A, Meehl GA, Han W, 2007. Role of the Bering Strait in the thermohaline circulation and abrupt climate change. Geophys Res Lett 34, 05704. Hughen KA, Overpeck J, Peterson L, Trumbore S, 1996. Rapid climate changes in the tropical Atlantic region during the last deglaciation. Nature 380, 51–54. Hughes ALC, Gyllencreutz R, Lohne ØS, Mangerud J, Svendsen JI, 2016. The last Eurasian ice sheets e a chronological database and time-slice reconstruction, DATED-1. Boreas 45, 1–45. Huttunen P, 1980. Early land use, especially the slash-and-burn cultivation in the commune of Lammi, southern Finland, interpreted mainly using pollen and charcoal analyses. Acta Botanica Fennica 113, 1– 45. Huybrechts P, 2002. Sea-level changes at LGM from ice-dynamic reconstructions of the Greenland and Antarctic ice sheets during glacial cycles. Quat Sci Rev 21, 203–231. Inamdar AK, Ramanathan V, 1998. Tropical and global scale interactions among water vapor, atmospheric greenhouse effect, and surface temperature. J Geophys Res-Atmos 103, 32177–32194. IPCC, 2014. Climate Change 2014: Impacts, Adaptation, and Vulnerability. Part A: Global and Sectoral Aspects. Contribution of Working Group II to the Fifth Assessment Report of the Intergovernmental Panel on Climate Change [Field, C.B., V.R. Barros, D.J. Dokken, K.J. Mach, M.D. Mastrandrea, T.E. Bilir, M. Chatterjee, K.L. Ebi, Y.O. Estrada, R.C. Genova, B. Girma, E.S. Kissel, A.N. Levy, S. MacCracken, P.R. Mastrandrea, and L.L. White (eds.)]. Cambridge University Press, Cambridge, United Kingdom and New York, NY, USA, 1132 pp. Ivanovic RF, Gregoire LJ, Kageyama M, Roche DM, Valdes PJ, Burke A, Drummond R, Peltier WR, Tarasov L, 2016. Transient climate simulations of the deglaciation 21–9 thousand years before present (version 1) – PMIP4 Core experiment design and boundary conditions. Geosci Model Dev 9, 2563–2587. Jackson ST, Betancourt JL, Booth RK, Gray ST, 2009. Ecology and the ratchet of events: climate variability, niche dimensions, and species distributions. Proc Natl Acad Sci USA 106, 19685–19692. Jennings A, Andrews J, Pearce C, Wilson L, Ólfasdótttir S, 2015. Detrital carbonate peaks on the Labrador shelf, a 13–7ka template for freshwater forcing from the Hudson Strait outlet of the Laurentide Ice Sheet into the subpolar gyre. Quat Sci Rev 107, 62–80.
BIBLIOGRAPHY Jessen CA, Rundgren M, Björck S, Muscheler R, 2007. Climate forced atmospheric CO2 variability in the early Holocene: A stomatal frequency reconstruction. Global and Planetary Change 57, 247–260. Jiang D, Lang X, Tian Z, Ju L, 2013. Mid-Holocene East Asian summer monsoon strengthening: insights from Paleoclimate Modeling Intercomparison Project (PMIP) simulations. Palaeogeog Palaeoclim Palaeoecol 369, 422–429. Jiang D, Lang X, Tian Z, Wang T, 2012. Considerable Model–Data Mismatch in Temperature over China during the Mid-Holocene: Results of PMIP Simulations. J Climate 25, 4135–4153. Johnsen SJ, Dahl-Jensen D, Gundestrup N, Steffensen JP, Clausen HB, Miller H, Masson-Delmotte V, Sveinbjornsdottir AE, White J, 2001. Oxygen isotope and palaeotemperature records from six Greenland ice-core stations: camp Century, Dye-3, GRIP, GISP2, Renland and NorthGRIP. J Quat Sci 16, 299–307. Jones CD, Gregory JM, Thorpe RB, Cox PM, Murphy JM, Sexton DMH, Valdes P, 2005. Systematic optimisation and climate simulation of FAMOUS, a fast version of HadCM3. Clim Dyn 25, 189–204. Jones MC, Yu Z, 2010. Rapid deglacial and early Holocene expansion of peatlands in Alaska. Proc Natl Acad Sci USA 107, 7347–7352. Joos F, Spahni R, 2008. Rates of change in natural and anthropogenic radiative forcing over the past 20,000 years. P Natl Acad Sci USA 105, 1425–1430. Jouzel J, Vimeux F, Caillon N, Delaygue G, Hoffmann G, Masson-Delmotte V, Parrenin F, 2003. Magnitude of isotope/temperature scaling for interpretation of central Antarctic ice cores. J Geophys Res 108 (D12), 4361. Juggins S, Birks HJ, 2012. Quantitative environomental reconstructions from Biological Data, in: Birks HJB, Lotter SJ, Juggins S, Smol J, (eds) Tracking environmental change using lake sediments. Data handling and numerical techniques 5, 431–494. Justino F, Peltier WR, 2005. The glacial North Atlantic Oscillation. Geophys Res Lett 32, L21803 Kageyama M, Braconnot P, Bopp L, Caubel A, Foujols MA, Guilyardi E, Khodri M, Lloyd J, Lombard F, Mariotti V, Marti O, Roy T, Woillez MN, 2013. Mid-Holocene and Last Glacial Maximum climate simulations with the IPSL model. Part II: model–data comparisons. Clim Dyn 40, 2469–2495. Kageyama M, Mignot J, Swingedouw D, Marzin C, Alkama R, Marti O, 2009. Glacial climate sensitivity to different states of the Atlantic meridional overturning circulation: results from the IPSL model. Clim Past 5, 551–570. Kandiano ES, Bauch HA, Müller A, 2004. Sea surface temperature variability in the North Atlantic during the last two glacial–interglacial cycles: comparison of faunal, oxygen isotopic, and Mg/Ca-derived records. Palaeogeog. Palaeoclim. Palaeoecol. 204, 145–164. Kaplan JO, Krumhardt KM, Zimmermann N, 2009. The prehistoric and preindustrial deforestation of Europe. Quat Sci Rev 28, 3016–3034. Kaufman D, Ager TA, Anderson NJ, Anderson PM, Andrews JT, Bartlein PT, Brubaker LB, Coats LL, Cwynar LC, Duvall ML, Dyke AS, Edwards ME, Eisner WR, Gajewski K, Geirsdottir A, Hu FS, Jennings AE, Kaplan MR, Kerwin MW, Lozhkin AV, MacDonald GM, Miller GH, Mock CJ, Oswald W W, OttoBliesner BL, Porinchuw DF, Ruhland K, Smol JP, Steig EJ, Wolfe BB, 2004. Holocene thermal maximum in the western Arctic (0– 180°W). Quat Sci Rev 23, 529–560. Kaufman DS, Axford YL, Henderson ACG, McKay NP, Oswald WW, Saenger C, Anderson RS, Bailey HL, Clegg B, Gajewski K, Hu FS, Jones MC, Massa C, Routson CC, Werner A, Wooller MJ, Yu Z, 2016. Holocene climate changes in eastern Beringia (NW North America) –A systematic review of multi-proxy evidence. Quat Sci Rev 147, 312–339. Kerwin MW, Overpeck JT, Webb RS, Anderson KH, 2004. Pollen-based summer temperature reconstructions for the eastern Canadian boreal forest, subarctic, and Arctic. Quat Sci Rev 23, 1901–1924. Kitover DC, van Balen R, Roche DM, Vandenberghe J, Renssen H, 2015. Advancement toward coupling of the
VAMPER permafrost model within the Earth system model iLOVECLIM (version 1.0): description and validation, Geosci Model Dev 8, 144–1460. Klemm J, Herzschuh U, Pisaric MFJ, Telford RJ, Heim B, Pestryakova LA, 2013. A pollen-climate transfer function from the tundra and taiga vegetation in Arctic Siberia and its applicability to a Holocene record, Palaeogeog Palaeoclim Palaeoecol 386, 702–713. Knutti R, Hegerl GC, 2008. The equilibrium sensitivity of the earth’s temperature to radiation changes. Nat Geosci 1, 735–743. Koerner RM, Fisher DA, 1990. A record of Holocene summer climate from a Canadian High Arctic ice core. Nature, 343, 630–631. Korhola A, Ruppel M, Seppä H, Väliranta M, Virtanen T, Weckström J, 2010. The importance of northern peatland expansion to the late-Holocene rise of atmospheric methane. Quat Sci Rev 29, 611–617. Kröel-Dulay G, Ransijn J, Schmidt IK, Beier C, De Angelis P, Dato de G, Dukes JS, Emmett B, Estiarte M, Garadnai J, Kongstad J, Kovács-Láng E, Steenberg Larsen K, Liberati D, Ogaya R, Riis-Nielsen T, Smith AR, Sowerby A, Tietema A, Penuelas J, 2015. Increased sensitivity to climate change in disturbed ecosystems. Nat Commun 6, 6682. Kuosmanen N, Fang K, Bradshaw RHW, Clear JL, Seppä H, 2014. Role of forest fires in Holocene stand-scale dynamics in the unmanaged taiga forest of northwestern Russia. Holocene 24(11), 1503-1514. Kuosmanen N, Seppä H, Alenius T, Bradshaw RHW, Clear JL, Filimonova L, Heikkilä M, Renssen H, Tallavaara M, Reitalu T, 2016a. Importance of climate, forest fires and human population size in the Holocene boreal forest composition change in northern Europe. Boreas 45, 688–702. Kuosmanen N, Seppä H, Reitalu T, Alenius T, Bradshaw RHW, Clear JL, Filimonova L, Kuznetsov O, Zaretskaya N, 2016b. Long-term forest composition and its drivers in taiga forest in NW Russia. Veg hist Archaeobot 25, 221–236. Kutzbach JE, He F, Vavrus SJ, Ruddiman WF. 2013. The dependence of equilibrium climate sensitivity on climate state: Applications to studies of climates colder than present. Geophys Res Lett 40:3721–3726. Kutzbach JE, Liu Z, 1997. Response of the African monsoon to orbital forcing and ocean feedbacks in the Middle Holocene. Science 278, 440–443. Lagerås P, 1996. Farming and forest dynamics in agriculturally marginal area of southern Sweden, 5000 BC to present: a palynological study of Lake Avegöl in the Småland Uplands. Holocene 6, 301–314. Lahtinen M, Oinonen M, Tallavaara M, Walker JW, Rowley-Conwy P, 2017. The advance of cultivation at its northern European limit: Process or event? Holocene 27, 427–438. Lambeck K, Rouby H, Purcell A, Sun Y, Sambridge M, 2014. Sea level and global ice volumes from the Last Glacial Maximum to the Holocene, Proc Natl Acad Sci USA 111, 15296–15303. Langen P L, Vinther BM, 2009. Response in atmospheric circulation and sources of Greenland precipitation to glacial boundary conditions. Clim Dyn 32, 1035–1054. Lavento M, 2001. Textile ceramics in Finland and on the Karelian Isthmus: nine variations and fugue on a theme of C. F. Meinander, Suomen Muinaismuistoyhdistyksen aikakauskirja. The Finnish Antiquarian Society, Helsinki. Le Treut H, Somerville R, Cubasch U, Ding Y, Mauritzen C, Mokssit A, Peterson T, Prather M, 2007. Historical overview of climate change. In: Solomon S, Qin D, Manning M, Chen Z, Marquis M, Averyt KB, Tignor M, Miller HL (Eds), Climate Change 2007: The Physical Science Basis. Contribution of Working Group I to the Fourth Assessment Report of The Intergovernmental Panel on Climate Change. Cambridge University Press, Cambridge, United Kingdom, and New York, NY, USA. Lechterbeck J, Edinborough K, Kerig T, Fyfe R, Roberts N, Shennan S, 2014. Is Neolithic land use correlated with demography? An evaluation of pollen-derived land cover and radiocarbon-inferred demographic change from Central Europe. Holocene 24, 1297–1307.
BIBLIOGRAPHY Levis S, Foley JA, Pollard D, 1999. CO2, climate, and vegetation feedbacks at the Last Glacial Maximum, J Geophys Res 104, 31191–31198. Li J, Xu Q, Zheng Z, Lu H, Luo Y, Li Y, Li C, Seppä H, 2015. Assessing the importance of climate variables for the spatial distribution of modern pollen data in China. Quat Res 83, 287–297. Licciardi JM, Teller JT, Clark PU, 1999. Freshwater routing by the Laurentide Ice Sheet during the last deglaciation, mechanism of global climate change at millennial time scales. Geophysical Monograph 112, 177–201. Lindner M, Maroschek M, Netherer S, Kremer A, Barbati A, Garcia-Gonzalo J, Seidl R, Delzon S, Corona P, Kolström M, Lexer MJ, Marchetti M, 2010. Climate change impacts, adaptive capacity, and vulnerability of European forest ecosystems. Forest Ecol Manag 259, 698–709. Lisiecki LE, Raymo ME, 2005. A Pliocene-Pleistocene stack of 57 globally distributed benthic δ18O records. Paleoceanography 20, PA1003. Liu Z, Notaro M, Kutzbach JE, Liu N, 2006. Assessing global vegetation-climate feedbacks from observations, J Climate 19, 787–814. Liu Z, Zhu J, Rosenthal Y, Zhang X, Otto-Bliesner BL, Timmermann A, Smith RS, Lohmann G, Zheng W, Elison Timm O, 2014. The Holocene temperature conundrum. Proc Natl Acad Sci USA 111, E3501–3505. Lohmann G, Pfeiffer M, Laepple T, Leduc G, Kim JH, 2013. A model–data comparison of the Holocene global sea surface temperature evolution. Clim Past 9, 1807–1839. Loulergue L, Schilt A, Spahni R, Masson-Delmotte V, Blunier T, Lemieux B, Barnola JM, Raynaud D, Stocker TF, Chappellaz J, 2008. Orbital and millennial-scale features of atmospheric CH4 over the past 800 000 years. Nature 453, 383–386. Loutre MF, Fichefet T, Goosse H, Huybrechts P, Goelzer H, Capron E, 2014. Factors controlling the last interglacial climate as simulated by LOVECLIM1.3. Clim Past Discuss 10, 235–290. Lunt DJ, Abe-Ouchi A, Bakker P, Berger A, Braconnot P, Charbit S, Fischer N, Herold N, Jungclaus JH, Khon VC, Krebs-Kanzow U, Langebroek PM, Lohmann G, Nisancioglu KH, Otto-Bliesner BL, Park W, Pfeiffer M, Phipps SJ, Prange M, Rachmayani R, Renssen H, Rosenbloom N, Schneider B, Stone EJ, Takahashi K, Wei W, Yin Q, Zhang ZS, 2013. A multi-model assessment of last interglacial temperatures. Clim Past 9, 699– 717. Luthi D, Le Floch M, Bereiter B, Blunier T, Barnola JM, Siegenthaler U, Raynaud D, Jouzel J, Fischer H, Kawamura K, Stocker TF, 2008. High-resolution carbon dioxide concentration record 650,000-800,000 years before present. Nature 453, 379–382. MacDonald GM, Velichko AA, Kremenetski CV, Borisova OK, Goleva AA, Andreev AA, Cwynar LC, Riding RT, Forman SL, Edwards TWD, Aravena R, Hammarlund D, Szeicz JM, Gattaulin VN, 2000. Holocene Treeline History and Climate Change Across Northern Eurasia. Quat Res 53, 302–311. Mairesse A, Goosse H, Mathiot P, Wanner H, Dubinkina S, 2013. Investigating the consistency between proxy-based reconstructions and climate models using data assimilation: a mid-Holocene case study. Clim Past 9, 2741–2757. Manning K, Colledge S, Crema E, Shennan S, Timpson A, 2016. The Cultural Evolution of Neolithic Europe. EUROEVOL Dataset 1: Sites, Phases and Radiocarbon Data. J Open Archaeology Data. 5. doi: http://doi.org/10.5334/joad.40 Marcott SA, Shakun JD, Peter UC, Mix AC, 2013. A Reconstruction of Regional and Global Temperature for the Past 11 300 Years. Science 339, 1198–1201. Marlon JR, Kelly R, Daniau A-L, Vannière B, Power MJ, Bartlein P, Higuera P, Blarquez O, Brewer S, Brücher T, Feurdean A, Gil Romera G, Iglesias V, Maezumi SY, Magi B, Courtney Mustaphi CJ, Zhihai T, 2016. Reconstructions of biomass burning from sediment-charcoal records to improve data–model comparisons. Biogeosciences 13, 3225–3244.
Marquer L, Gaillard MJ, Sugita S, Poska A, Trondman AK, Mazier F, Nielsen AB, Fyfe RM, Jönsson AM, Smith B, Kaplan JO, Alenius T., Birks HJB, Bjune AE, Christiansen J, Dodson J, Edwards KJ, Giesecke T, Herzschuh U, Kangur M, Koff T, Latałowa M, Lechterbeck J, Olofsson J, Seppä H. (in review) Quantifying the effects of land use and climate on Holocene vegetation in Europe. Quat Sci rev. Marquer L, Gaillard MJ, Sugita S, Trondman AK, Mazier F, Nielsen AB, Fyfe RM, Vad Odgaard B, Alenius T, Birks HJB, Bjune AE, Christiansen J, Dodson J, Edwards KJ, Giesecke T, Herzschuh U, Kangur M, Lorenz S, Poska A, Schult M, Seppä H, 2014. Holocene changes in vegetation composition in northern Europe: why quantitative pollen-based vegetation reconstructions matter. Quat Sci Rev 90, 199–216. Masson V, Cheddadi R, Braconnot P, PMIP participants, 1999. Mid-Holocene climate in Europe: what can we infer from PMIP model–data comparisons? Clim Dyn 15, 163–182. Mathiot P, Goosse H, Crosta X, Stenni B, Braida M, Renssen H, Van Meerbeeck CJ, Masson-Delmotte V, Mairesse A, Dubinkina S, 2013. Using data assimilation to investigate the causes of Southern Hemisphere high latitude cooling from 10 to 8 ka BP. Clim Past 9, 887–901. Mauri A, Davis BAS, Collins P M, Kaplan JO, 2015. The climate of Europe during the Holocene: a gridded pollen-based reconstruction and its multi-proxy evaluation. Quat Sci Rev 112, 109–127. McGuffie K, Henderson-Sellers A, 2005. A Climate Modelling Primer, Third Edition. McManus JF, Francois R, Gherardi JM, Keigwin LD, Brown-Leger S, 2004. Collapse and rapid resumption of Atlantic meridional circulation linked to deglacial climate changes. Nature 428, 834–837. Mitchell CE, Agrawal AA, Bever JD, Gilbert GS, Hufbauer RA, Klironomos JN, Maron JL, Morris WF, Parker IM, Power AG, Seabloom EW, Torchin ME, Vázquez DP, 2006. Biotic interactions and plant invasions. Ecol Lett 9, 726–740. Mökkönen T, 2014. Archaeological Radiocarbon Dates as a Population Proxy: a Skeptical View. Fennoscandia Archaeologica 31, 125–134. Molinari C, Lehsten V, Bradshaw RHW, Power MJ, Harmand P, Arneth A, Kaplan JO, Vanniere B, Sykes MT (2013). Exploring potential drivers of European biomass burning over the Holocene: a data–model analysis. Global Ecology and Biogeography 22, 1248–1260. Monnin E, Indermuhle A, Dallenbach A, Fluckiger J, Stauffer B, Stocker TF, Raynaud D, Barnola JM, 2001. Atmospheric CO2 concentrations over the last glacial termination. Science 291(5501), 112–114. Mouchet A, and Francois LM, 1996. Sensitivity of a global oceanic carbon cycle model to the circulation and the fate of organic matter: preliminary results. Phys Chem Earth 21, 511–516. Nesje A, 2005. Briksdalsbreen in western Norway: AD 1900–2004 frontal fluctuations as a combined effect of variations in winter precipitation and summer temperature. Holocene 15, 1245–1252. Nesje A, 2009. Latest Pleistocene and Holocene alpine glacier fluctuations in Scandinavia. Quat Sci Rev 28, 2119–2136. Nielsen AB, Giesecke T, Theuerkauf M, Feeser I, Behre KE, Beug HJ, Chen SH, Christiansen J, Dörfler W, Endtmann E, Jahns S, de Klerk P, Kühl N, Latalowa M, Vad Odgaard B, Rasmussen P, Stockholm JR, Voigt R, Wiethold J, Wolters S, 2012. Quantitative reconstructions of changes in regional openness in northcentral Europe reveal new insights into old questions. Quat Sci Rev 47, 131–149. Niinemets E, Saarse L, 2006. Holocene forest dynamics and human impact in southeastern Estonia. Veg Hist Archaeobot 16, 1–13. Occhietti S, Parent M, Lajeunesse P, Robert F, Govare É, 2011. Late Pleistocene–Early Holocene decay of the Laurentide Ice Sheet in Québec–Labrador. Dev Quat Sci 15, 601–630. Oinonen M, Pesonen P, Tallavaara M, 2010. Archaeological radiocarbon dates for studying the population history of Eastern Fennoscandia. Radiocarbon 522(3), 292–407. Oleson KW, Bonan GB, Schaaf C, Gao F, Jin Y, Strahler A, 2003. Assessment of global climate model land surface albedo using MODIS data. Geophys Res Lett 30, 1143.
BIBLIOGRAPHY Opsteegh JD, Haarsma RJ, Selten FM, Kattenberg A, 1998. ECBILT: a dynamic alternative to mixed boundary conditions in ocean models. Tellus 50A, 348–367. Otto-Bliesner BL, Tomas R, Brady EC, Ammann C, Kothavala Z, Clauzet G, 2006. Climate sensitivity of moderate- and low-resolution versions of CCSM3 to preindustrial forcings. J Climate 19, 2567–2583. Pacanowski RC, Dixon K, Rosati A, 1990. Readme file for GFDL-MOM 1.0. Geophys Fluid Dyn Lab, Princeton, NJ. PALAEOSENSE Project Members, 2012. Making sense of palaeoclimate sensitivity. Nature 491, 683–691. Parshall T, Calcote R, 2001. Effect of pollen from regional vegetation on stand-scale forest reconstruction. Holocene 11, 81–87. Pausata FSR, Li C, Wettstein JJ, Kageyama M, Nisancioglu KH, 2011. The key role of topography in altering North Atlantic atmospheric circulation during the last glacial period. Clim Past 7, 1089–1101. Peltier WR, 2004. Global glacial isostasy and the surface of the ice-age earth: The ice-5G (VM2) model and grace. Annu Rev Earth Pl Sc 32, 111–149. Peros M, Gajewski K, Paull T, Ravindra R, Podritske B, 2010. Multi-proxy record of postglacial environmental change, south-central Melville Island, Northwest Territories, Canada. Quat Res 73, 247–258. Pope VD, Gallani ML, Rowntree PR, Stratton RA, 2000. The impact of new physical parametrizations in the Hadley Centre climate model. Clim Dyn 16, 123–146. Post E, 2013. Ecology of Climate Change: The Importance of Biotic Interactions. 408 pp, Princeton University Press, Princeton and Oxford. Power MJ, Marlon JR, Bartlein PJ, Harrison SP, 2010. Fire history and the Global Charcoal Database: A new tool for hypothesis testing and data exploration, Palaeogeog Palaeoclim Palaeoecol 291, 52–59. Putkinen N, Lunkka JP, 2008. Ice stream behaviour and deglaciation of the Scandinavian Ice Sheet in the Kuittijärvi area, Russian Karelia. Bull Geol Soc Finl 80, 19–37. R Development Core Team, 2016. R: A language and environment for statistical computing. Vienna: R Foundation for Statistical Computing. ISBN 3-900051-07-0, http://www.r-project.org Rahmstorf S, 2002. Ocean circulation and climate during the past 120,000 years. Nature 419, 207–214. Rahmstorf S, Crucifix M, Ganopolski A, Goosse H, Kamenkovich I, Knutti R, Lohmann G, Marsh R, Mysak LA, Wang Z, Weaver AJ, 2005. Thermohaline circulation hysteresis: A model intercomparison. Geophys Res Lett 32, L23605. Ramaswamy V et al., 2001. Radiative forcing of climate change, in: Climate Change 2001: The Scientific Basis. Contribution of Working Group I to the Third Assessment Report of the Intergovernmental Panel on Climate Change, Cambridge Univ. Press, Cambridge, 249–416. Randall DA et al., 2007. Cilmate Models and Their Evaluation. In: Solomon, S, Qin D, Manning M, Chen Z, Marquis M, Averyt KB, Tignor M, Miller HL (eds.) Climate Change 2007: The Physical Science Basis. Contribution of Working Group I to the Fourth Assessment Report of the Intergovernmental Panel on Climate Change. Cambridge University Press, Cambridge, United Kingdom and New York, NY, pp 589–662. Rasmussen SO, Andersen KK, Svensson AM, Steffensen JP, Vinther BM, Clausen HB, Siggaard-Andersen ML, Johnsen SJ, Larsen LB, Dahl-Jensen D, Bigler M, Röthlisberger R, Fischer H, Goto-Azuma K, Hansson ME, Ruth U, 2006. A new Greenland ice core chronology for the last glacial termination. J Geophys Res 111, D06102. Regnell M, Gaillard MJ, Bartholin TS Karsten P, 1995. Reconstruction of environment and history of plant use during the late Mesolithic (Ertebølle culture) at the inland settlement of Bökeberg III, southern Sweden. Veg Hist Archaeobot 4(2), 67–91. Reitalu T, Seppä H, Sugita S et al., 2013. Long-term drivers of forest composition in a boreonemoral region: The relative importance of climate and human impact. J Biogeography 40, 1524–1534.
Renssen H, Driesschaert E, Loutre MF, Fichefet T, 2006. On the importance of initial conditions for simulations of the Mid-Holocene climate. Clim Past 2, 91–97. Renssen H, Goosse H, Crosta X, Roche DM, 2010. Early Holocene Laurentide Ice Sheet deglaciation causes cooling in the high-latitude Southern Hemisphere through oceanic teleconnection. Paleoceanography, 25, PA3204. Renssen H, Goosse H, Fichefet T, 2002. Modeling the effect of freshwater pulses on the early Holocene climate: the influence of high frequency climate variability. Paleoceanography, 17, 1020. Renssen H, Goosse H, Fichefet T, Brovkin V, Driesschaert E, Wolk F, 2005. Simulating the Holocene climate evolution at northern high latitudes using a coupled atmosphere-sea ice-ocean-vegetation model. Clim Dyn 24, 23–43. Renssen H, Seppä H, Crosta X, Goosse H, Roche DM, 2012. Global characterization of the Holocene Thermal Maximum. Quat Sci Rev 48, 7–19. Renssen H, Seppä H, Heiri O, Roche DM, Goosse H, Fichefet T, 2009. The spatial and temporal complexity of the Holocene thermal maximum. Nat Geosci 2, 411–414. Ritchie JC, 1986. Climate change and vegetation response. Vegetation 67, 65–74. Roche DM, Dumas C, Bügelmayer M, Charbit S, Ritz C, 2014. Adding a dynamical cryosphere to iLOVECLIM (version 1.0): coupling with the GRISLI ice-sheet model. Geosci Model Dev 7, 1377–1394. Roche DM, Renssen H, Weber SL, Goosse H, 2007. Could meltwater pulses have been sneaked unnoticed into the deep ocean during the last glacial. Geophys Res Lett 34, L24708. Roche DM, Wiersma AP, Renssen H, 2010. A systematic study of the impact of freshwater pulses with respect to different geographical locations, Clim Dyn 34, 997–1013. Romanova V, Lohmann G, Grosfeld K, 2006. Effect of land albedo, CO2, orography, and oceanic heat transport on extreme climates. Clim Past 2, 31–42. Rosenzweig C, Parry ML, 1994. Potential impact of climate change on world food supply. Nature 367, 133– 138. Ruddiman WF, 2007. The early anthropogenic hypothesis: challenges and responses. Rev Geophys 45, RG4001. Salonen JS, Seppä H, Väliranta M, Jones VJ, Self A, Heikkilä M, Kultti S, Yang H, 2011. Holocene thermal maximum and the late-Holocene cooling in the tundra of NE European Russia. Quat Res 75, 501–511. Schilt A, Baumgartner M, Schwander J, Buiron D, Capron E, Chappellaz J, Loulergue L, Schüpbach S, Spahni R, Fischer H, Stocker TF, 2010. Atmospheric nitrous oxide during the last 140000years. Earth Planet Sci Lett 300, 33–43. Sejrup HP, Seppä H, McKay NP, Kaufman DS, Geirsdóttir Á, de Vernal A, Renssen H, Husum K, Jennings A, Andrews JT, 2016. North Atlantic-Fennoscandian Holocene climate trends and mechanisms. Quat Sci Rev 147, 365–378. Seppä H, Alenius T, Bradshaw RHW, Giesecke T, Heikkilä M, Muukkonen P. 2009a. Invasion of Norway spruce (Picea abies) and the rise of the boreal ecosystem in Fennoscandia. J Ecology 97,629–640. Seppä H, Birks HJB, Odland A, Poska A, Veski S, 2004. A modern pollen–climate calibration set from northern Europe: developing and testing a tool for palaeoclimatological reconstructions. J Biogeography 31, 251– 267. Seppä H, Bjune AE, Telford RJ, Birks HJB, Veski S, 2009b. Last nine-thousand years of temperature variability in Northern Europe. Clim Past Discuss 5, 1521–1552. Seppä H, Schurgers G, Miller PA , Bjune AE, Giesecke T, Kühl N, Renssen H, Salonen JS, 2015. Trees tracking a warmer climate: The Holocene range shift of hazel (Corylus avellana) in northern Europe. Holocene 25(1), 53–63.
BIBLIOGRAPHY Seppä, H, Bennett KD, 2003. Quaternary pollen analysis: recent progress in palaeoecology and palaeoclimatology. Prog Phys Geog 27, 548–579. Shakun JD, Clark PU, He F, Marcott SA, Mix AC, Liu Z, Otto-Bliesner B, Schmittner A, Bard E, 2012. Global warming preceded by increasing carbon dioxide concentrations during the last deglaciation. Nature 484, 49–54. Shennan S, Edinborough K, 2007. Prehistoric population history: from the Late Glacial to the Late Neolithic in Central and Northern Europe. J Archaeol Sci 34, 1339–1345. Singarayer JS, Valdes PJ, 2010. High-latitude climate sensitivity to ice-sheet forcing over the last 120 kyr. Quat Sci Rev 29, 43–55. Sköld E, Lagerås P, Berglund BE, 2010. Temporal cultural landscape dynamics in a marginal upland area: agricultural expansions and contractions inferred from palynological evidence at Yttra Berg, southern Sweden. Veg Hist Archaeobot 19, 121–136. Smith RS, Gregory JM, 2009. A study of the sensitivity of ocean overturning circulation and climate to freshwater input in different regions of the North Atlantic. Geophys Res Lett 36, L15701. Smith RS, Gregory JM, Osprey A, 2008. A description of the FAMOUS (version XDBUA) climate model and control run. Geosci Mod Dev 1, 147–185. Sørensen L, Karg S, 2012. The expansion of agrarian societies towards the north e new evidence for agriculture during the Mesolithic/Neolithic transition in Southern Scandinavia, J Archaeol Sci 51, 98–114. Steffensen JP, Andersen KK, Bigler M, Clausen HB, Dahl-Jensen D, Fischer H, Goto-Azuma K, Hansson M, Johnsen SJ, Jouzel J, Masson-Delmotte V, Popp T, Rasmussen SO, Rothlisberger R, Ruth U, Stauffer B, Siggaard-Andersen ML, Sveinbjornsdottir AE, Svensson A, White JWC, 2008. High-Resolution Greenland Ice Core Data Show Abrupt Climate Change Happens in Few Years. Science, 321, 680–683. Stokes CR, Tarasov L, 2010. Ice streaming in the Laurentide Ice Sheet: A first comparison between datacalibrated numerical model output and geological evidence. Geophys Res Lett 37 (1), L01501. Stokes RC, Tarasov L, Blomdin R, Cronin MT, Fisher GT, Gyllencreutz R, Hättestrand C, Hindmarsh RCA, Hughes LCA, Jakobsson M, Kirchner N, Livingstone JS, Margold M, Murton BJ, Noormets R, Peltier RW, Peteet MD, Piper JWD, Preusser F, Renssen H, Roberts HD, Roche MD, Saint-Ange F, Stroeven PA, Teller TJ, 2015. On the reconstruction of palaeo-ice sheets: Recent advances and future challenges, Quat Sci Rev 125, 15–49. Stouffer RJ, Yin J, Gregory JM, Dixon KW, Spelman MJ, Hurlin W, Weaver AJ, Eby M, Flato GM, Hasumi H, Hu A, Jungclaus JH, Kamenkovich IV, Levermann A, Montoya M, Murakami S, Nawrath S, Oka A, Peltier WR, Robitaille DY, Sokolov A, Vettoretti G, Weber SL, 2006. Investigating the causes of the response of the thermohaline circulation to past and future climate changes. J Climate 19, 1365–1387. Strack M, Waddington JM, Lucchese MC, Cagampan JP, 2009. Moisture controls on CO2 exchange in aSphagnum-dominated peatland: results from an extreme drought field experiment. Ecohydrology 2, 454–461. Sugita S, 2007. Theory of quantitative reconstruction of vegetation I: pollen from large sites REVEALS regional vegetation. Holocene 17, 229–241. Sundqvist HS, Kaufman DS, McKay NP, Balascio NL, Briner JP, Cwynar LC, Sejrup HP, Seppä H, Subetto DA, Andrews JT, Axford Y, Bakke J, Birks HJB, Brooks SJ, de Vernal A, Jennings AE, Ljungqvist FC, Rühland KM, Saenger C, Smol JP, Viau AE, 2014. Arctic Holocene proxy climate database—new approaches to assessing geochronological accuracy and encoding climate variables. Clim Past 10, 1605–1631. Svendsen J, Alexanderson H, Astakhov V, Demidov I, Dowdeswell J, Funder S, Gataullin V, Henriksen M, Hjort C, Houmark-Nielsen M, Ingolfsson HHO, Jakobsson M, Kjaer K, Larsen E, Lokrantz H, Lunkka J, Lysa A, Mangerud J, Matiouchkov A, Murray A, Moller P, Niessen F, Nikolskaya O, Polyak L, Saarnisto M, Siegert C, Siegert M, Spielhagen R, Stein R, 2004. Late Quaternary ice sheet history of northern Eurasia, Quat Sci Rev 23, 1229–1271.
Taavitsainen JP, Simola H, Grönlund E, 1998. Cultivation History Beyond the Periphery: Early Agriculture in the North European Boreal Forest. J World Prehist 12(2), 199–253. Tallavaara M, Pesonen P, Oinonen M, 2010. Prehistoric population history in eastern Fennoscandia. J Archaeol Sci 37, 251–260. Tallavaara M, Pesonen P, Oinonen M, Seppä H, 2014. The Mere Possibility of Biases Does Not Invalidate Archaeological Population Proxies – Response to Teemu Mökkönen. Fennoscandia Archaeologica 31, 135–140. Tallavaara M, Seppä H, 2011. Did the mid-Holocene environmental changes cause the boom and bust of hunter-gatherer population size in eastern Fennoscandia? Holocene 222, 215–225. Tarasov PE, Müller S, Zech M, Andreeva D, Diekmann B, Leipe C, 2013. Last glacial vegetation reconstructions in the extreme-continental eastern Asia: Potentials of pollen and n-alkane biomarker analyses, Quat Int 290–291, 253–263. Telford RJ, Birks HJB, 2005. The secret assumption of transfer functions: problems with spatial autocorrelation in evaluating model performance. Quat Sci Rev 24, 2173–2179. Telford RJ, Birks HJB, 2009. Evaluation of transfer functions in spatially structured environments. Quat Sci Rev 28, 1309–1316. Teller JT, Leverington DW, Mann JD, 2002. Freshwater outbursts to the oceans from glacial Lake Agassiz and their role in climate change during the last deglaciation. Quat Sci Rev 21, 879–887. ter Braak CJF, Juggins S, 1993. Weighted averaging partial least-squares regression (WA-PLS) —an improved method for reconstructing environmental variables from species assemblages. Hydrobiologia 269, 485– 502. Thornalley DJ, Elderfield H, McCave IN, 2009. Holocene oscillations in temperature and salinity of the surface subpolar North Atlantic. Nature 457, 711–714. Thornalley DJR, Barker S, Becker J, Hall IR, Knorr G, 2013. Abrupt changes in deep Atlantic circulation during the transition to full glacial conditions. Paleoceanography 28, 253–262. Thornalley DJR, Elderfield H, McCave IN, 2011. Reconstructing North Atlantic deglacial surface hydrography and its link to the Atlantic overturning circulation, Global Planet Change 79, 163–175. Tikkanen M, Oksanen J, 2002. Late Weichselian and Holocene shore displacement history of the Baltic Sea in Finland. Fennia 180(1–2), 9–20. Timpson A, Colledge S, Crema E, Edinborough K, Kerig T, Manning K, Thomas MG, Shennan S, 2014. Reconstructing regional population fluctuations in the European Neolithic using radiocarbon dates: a new case-study using an improved method. J Archaeol Sci 52, 549–557. Timpson A, Manning K, Shennan S, 2015. Inferential mistakes in population proxies: A response to Torfing’s “Neolithic population and summed probability distribution of 14C-dates.” J Archaeol Sci 63, 199–202. Torfing T, 2015. Neolithic population and summed probability distribution of 14C-dates. J Archaeol Sci 63, 193–198. Torfing T, 2016. Demographic Proxies for the Neolithic: With Emphasis on the Langeland Region, Denmark. Acta Archaeologica 87, 33–48. Trenberth KE, Stepaniak DP, 2004. The flow of energy through the earth's climate system. J Roy Meteorological Society 130, 2677–2701. Trondman AK, Gaillard MJ, Mazier F et al., 2015. First pollen-based quantitative reconstructions of Holocene regional vegetation cover (plant functional types and landcover types) in Europe suitable for climate modelling. Glob Change Biol 21, 676–697. Trondman AK, Gaillard MJ, Sugita S, Björkman L, Greisman A, Hultberg T, Lagerås P, Lindbladh M, Mazier F, 2016. Are pollen records from small sites appropriate for REVEALS model-based quantitative
BIBLIOGRAPHY reconstructions of past regional vegetation? An empirical test in southern Sweden. Veg Hist Archaeobot 25, 131–151. Tudoran MM, Marquer L., Jönsson AM, 2016. Historical experience (1850–1950 and 1961–2014) of insect species responsible for forest damage in Sweden: Influence of climate and land management changes. Forest Ecol Manag 381, 347–359. van der Wal W, Wu P, Wang H, Sideris MG, 2010. Sea levels and uplift rate from composite rheology in glacial isostatic adjustment modeling. J Geodynamics 50, 38–48. Vaughan DG et al., 2013. Observations: Cryosphere, in Climate Change 2013: The Physical Science Basis. Contribution of Working Group I to the Fifth Assessment Report of the Intergovernmental Panel on Climate Change, edited by TF Stocker et al., chap. 4, pp. 317–382, Cambridge Univ. Press, Cambridge, UK. Velle G, Brodersen K, Birks H, Willassen E, 2012. Inconsistent results should not be overlooked: A reply to Brooks et al. (2012). Holocene 22, 1501–1508. Velle G, Brodersen KP, Birks HJB, Willassen E, 2010. Midges as quantitative temperature indicator species: Lessons for palaeoecology. Holocene 20, 989–1002. Vinther BM, Buchardt SL, Clausen HB, Dahl-Jensen D, Johnsen SJ, Fisher DA, Koerner RM, Raynaud D, Lipenkov V, Andersen KK, Blunier T, Rasmussen SO, Steffensen JP, Svensson AM, 2009. Holocene thinning of the Greenland ice sheet. Nature 461, 385–388. Vinther BM, Clausen HB, Fisher DA, Koerner RM, Johnsen SJ, Andersen KK, Dahl-Jensen D, Rasmussen SO, Steffensen JP, Svensson AM, 2008. Synchronizing ice cores from the Renland and Agassiz ice caps to the Greenland Ice Core Chronology. J Geophys Res 113, D08115. Vinther BM, Clausen HB, Johnsen SJ, Rasmussen SO, Andersen KK, Buchardt SL, Dahl-Jensen D, Seierstad IK, Siggaard-Andersen ML, Steffensen JP, Svensson A, Olsen J, Heinemeier J, 2006. A synchronized dating of three Greenland ice cores throughout the Holocene, J Geophys Res 111, D13102. Vorren TO, Mangerud J, Blikra L, Nesje A, Sveian H, 2008. The emergence of modern Norway, in Ramberg, I.B., Bryhni, I., Nottvedt, A., Rangnes, K. (Eds) The Making of a land geology of Norway. Norwegian Geological Association, pp. 543–559. Vuorela I, Lempiäinen T, 1988. Archaeobotany of the sire of the oldest cereal grain find in Finland. Annales Botanici Fennici 25, 33–45. Wagner F, Bohncke SJP, Dilcher DL, Kürschner W, van Geel B, Visscher H, 1999. Century-scale shifts in Early Holoceneatmospheric CO2 concentration. Science 284, 1971–1973. Wang Y, Cheng H, Edwards RL, He Y, Kong X, An Z, Wu J, Kelly MJ, Dykoski C, Li X, 2005. The Holocene Asian monsoon: links to solar changes and North Atlantic climate. Science, 308, 854–857. Weaver AJ, Eby M, Augustus FF, Wiebe EC, 1998. Simulated Influence of Carbon Dioxide, Orbital Forcing and Ice Sheets on the Climate of the Last Glacial Maximum. Nature 394, 847–853. Webb T, 1986. Is vegetation in equilibrium with climate? How to interpret Late-Quaternary pollen data. Vegetation 67, 75–91. Welinder S, 2011. Early farming households 3900–800 BC. In Myrdal J, Morell M, (ed): The Agrarian History of Sweden: From 4000 BC to AD 2000. Nordic Academic Press, Lund. Williams AN, 2012. The use of summed radiocarbon probability distributions in archaeology: a review of methods. J Archaeol Sci 39, 578–589. Williams AN, Veth P, Steffen W, Ulm S, Turney CSM, Reeves JM, Phipps SJ, Smith M, 2015. A continental narrative: Human settlement patterns and Australian climate change over the last 35,000 years. Quat Sci Rev 123, 91–112. Wohlfahrt J, Harrison, SP, Braconnot P et al., 2008. Evaluation of coupled ocean-atmosphere simulations of the mid-Holocene using palaeovegetation data from the northern hemisphere extratropics. Clim Dyn 31(7–8), 871–890.
Woodbridge J, Fyfe RM, Roberts N, Downey S, Edinborough K, Shennan S, 2014. The impact of the Neolithic agricultural transition in Britain: a comparison of pollen-based land-cover and archaeological 14C dateinferred population change. J Archaeol Sci 51, 216–224. Wu P, Wang H, 2008. Postglacial isostatic adjustment in a self-gravitating spherical earth with power-law rheology. J Geodynamics 46, 118–130. Yeager SG, Shields CA, Large WG, Hack JJ, 2006. The low-resolution CCSM3. J Climate 19, 2545–2566. Yu JY, Hartmann DL, 1995. Orographic influences on the distribution and generation of atmospheric variability in a GCM. J Atmos Sci 52, 2428–2443. Yuan D, Cheng H, Edwards RL, Dykoski CA, Kelly MJ, Zhang M, Qing J, Lin Y, Wang Y, Wu J, Dorale JA, An Z, Cai Y, 2004. Timing, duration, and transitions of the last interglacial Asian monsoon. Science, 304(5670), 575–578. Zhang Q, Sundqvist HS, Moberg A, Körnich H, Nilsson J, Holmgren K, 2010. Climate change between the mid and late Holocene in northern high latitudes Part 2: model–data comparisons. Clim Past 6, 609–626. Zhang Y, Renssen H, Seppä H, 2016. Effects of melting ice sheets and orbital forcing on the early Holocene warming in the extratropical Northern Hemisphere. Clim. Past 12, 1119–1135. Zhang Y, Renssen H, Seppä H, Valdes PJ, 2017. Holocene temperature evolution in the Northern Hemisphere high latitudes —data–model comparisons. Quat Sci Rev 173, 101–113. Zhang Y, Renssen H, Seppä H, Valdes PJ, accepted. Holocene temperature trends in the extratropical Northern Hemisphere based on inter-model comparisons. J Quat Sci. Zona D, Oechel WC, Kochendorfer J, Paw UKT, Salyuk AN, Olivas PC, Oberbauer SF, Lipson DA, 2009. Methane fluxes during the initiation of a large-scale water table manipulation experiment in the Alaskan Arctic tundra. Global Biogeochem Cy 23, GB2013.