Historical Analysis of Tunnel Approach Displacements with Satellite Remote Sensing Edward J. Hoppe, Young-Jun Kweon, Brian S. Bruckno, Scott T. Acton, Lauren Bolton, Andrew Becker, and Andrea Vaccari tion of seismic-induced displacements, landslide monitoring, and sinkhole detection (1). One major advantage of using active radar systems is that data can be acquired in any weather and light conditions because the atmospheric absorption of microwave frequencies is very low. Radar satellites typically operate in low Earth orbit (i.e., at altitudes of 600 to 800 km).
Historical displacements of tunnel boat sections at the approaches to the Monitor–Merrimac Memorial Bridge–Tunnel (MMMBT) in Virginia were investigated as a potential reason for ongoing seawater infiltration. Archived data collected from December 2001 to March 2010 by the Radarsat-1 Earth-orbiting radar satellite were analyzed. Millimetric precision was achieved in displacement measurements over an area of approximately 100 km2, including the MMMBT and adjacent regions of Suffolk and Newport News, Virginia. Data consisting of 42 radar acquisitions were processed by using the interferometric synthetic aperture radar (InSAR) differential technique. Additional statistical analyses were conducted on selected points of interest. Results of the historical analysis of satellite radar remote sensing data indicated no significant displacements of the tunnel boat sections during the study period. The annual displacement rate precision of the tunnel boat sections was estimated to be 61 mm/year at the 95th percentile confidence level. Thus, settlement of the constructed islands was unlikely to have been a reason for the ongoing water infiltration.
Background The Monitor–Merrimac Memorial Bridge–Tunnel (MMMBT) is a 7.4-km river crossing for Interstate 664 between Newport News and Suffolk, Virginia. The four-lane structure is composed of twin trestle bridges, two constructed islands, and a 1.5-km-long tunnel under a portion of the Hampton Roads harbor where the James, Nansemond, and Elizabeth Rivers meet. This large transportation project cost approximately $400 million and opened to traffic in 1992. Islands at the tunnel approaches required approximately 2.1 million m3 of hydraulic fill to construct. Large precast concrete elements were used to build open-cut tunnel approaches at each island. These structures, called boat sections, were connected in series for the length of the approach. The north and south tunnel approaches consist of 33 and 39 boat sections, respectively. Joints between adjacent boat sections were waterproofed during construction to prevent water infiltration. Unlike approaches to some other tunnels in the area, the MMMBT approaches are not supported on piles. Characterized by layers of sedimentary deposits, the MMMBT lies in the Coastal Plain physiographic province of Virginia. Around the tunnel entrance, the Norfolk Formation is composed of thick sequences of fine sand, silty sand, silt, and clay (2). Basement strata along the Atlantic continental margin consist of eastward-dipping granitic and metasedimentary rocks (3). The tunnel entrance lies on a passive tectonic margin well to the east of the Central Virginia Seismic Zone and generally is seismically quiescent. The Virginia Coastal Plain is one of the most groundwater extraction–stressed regions on the U.S. East Coast. Groundwater extraction can result in localized ground subsidence because of the reduction of pore water pressure in aquifers and consolidation of clays (4). Local ground deformations are further complicated by the proximity of the tunnel entrance to the Chesapeake Bay impact structure. First identified by the Deep Sea Drilling Project Site 612, the structure resulted from a meteorite impact approximately 35.5 million years ago (5). The impact was centered near Cape Charles, Virginia, and resulted in a large crater approximately 85 km in diameter (6). According to various available interpretations of gravity and seismic
A satellite-based measurement technique known as synthetic aperture radar (SAR) can be used for monitoring large surfaces with a reasonably high ground resolution. By taking advantage of the phase information contained in SAR images, interferometric SAR (InSAR) directly measures the topography of a particular scene, and a set of techniques known as differential InSAR (DInSAR) accurately measure changes in elevation between sequential observations. This latter approach is particularly attractive for forensic analysis because measurements typically do not require any ground-based monitoring devices and rely fully on the natural scattering of radar signals off various site features (radar scatterers). Practically any area on the Earth’s surface that is visible from the satellite can be analyzed for temporal displacement (i.e., settlement or heave). Specialized DInSAR techniques theoretically can measure millimetric surface deformations. Practical applications include the detecE. J. Hoppe and Y.-J. Kweon, Virginia Center for Transportation Innovation and Research, 530 Edgemont Road, Charlottesville, VA 22903. B. S. Bruckno, Virginia Department of Transportation, 811 Commerce Road, Staunton, VA 24401. S. T. Acton, A. Becker, and A. Vaccari, Department of Electrical and Computer Engineering, and L. Bolton, Department of Civil and Environmental Engineering, University of Virginia, Charlottesville, VA 22904. Corresponding author: E. J. Hoppe,
[email protected] Transportation Research Record: Journal of the Transportation Research Board, No. 2510, Transportation Research Board, Washington, D.C., 2015, pp. 15–23. DOI: 10.3141/2510-03 15
16
Transportation Research Record 2510
data, the tunnel entrance lies in the trough between the central uplift and the outer rim (3, 7). The magnitude and extent of the ongoing ground deformation attributable to faulting at the Chesapeake Bay impact structure are currently undefined. Problem Statement The Virginia Department of Transportation (DOT) is conducting a comprehensive engineering study designed to address persistent water intrusion at both tunnel approaches to the MMMBT. Noticeable quantities of seawater have seeped through joints and cracks at the boat sections. Even though this condition poses no imminent safety concerns, it causes substantial maintenance problems. Therefore, the Virginia DOT is collaborating with expert consultants to develop remedial measures designed to reduce, capture, and divert water inflow away from the pavement section. Water inflow has been a problem at the MMMBT for many years. In 2009, in response to observations of increased seepage, the Virginia DOT employed several consulting firms to investigate the problem (8, 9). Parsons Brinckerhoff noted water intrusion at “locations in the walls and along the joint between the wall and the exterior jersey barrier where the waterproofing membrane appears to have failed, allowing water to enter the walls in cracks and at joints particularly at the base of the jersey barriers” (9). Of significance, the report also states that “without a testing program of significant scope, it is impossible to definitively determine the cause or location of the failed waterproofing system.” A May 2013 report by Whitman, Requardt, and Associates, LLC, indicates that “water infiltration has occurred in varying degrees within the approach structures since the asset was completed” (8). The report concludes that “the most likely cause of failure of the waterproofing system is the inability of the waterproofing membrane and waterstop materials to accommodate joint movement experienced between boat sections, either due to initial concrete shrinkage or thermal movement over time.”
Purpose and Scope The purpose of this study was to test the hypothesis that the settlement of MMMBT boat sections at constructed islands was one reason for the waterproofing system failure. If a significant amount of consolidation settlement took place at some point after construction, then seawater infiltration may have been exacerbated by the differential displacement of adjacent boat sections and resulting openings at the joints. The scope of the study was limited to the analysis of available historical records. Methods The Virginia DOT asked the Virginia Center for Transportation Innovation and Research (VCTIR) to conduct a study of MMMBT approaches. No postconstruction elevation surveys of the MMMBT constructed islands had been performed in the 22 years since the tunnel opened, and no other conventional survey records were available for review. This study analyzed remote sensing data previously acquired by Earth-orbiting radar satellites. Historical radar records, when available for a given area of interest (AOI), can be used to detect past surficial deformations. The study was a collaborative effort among researchers from VCTIR, the University of Virginia (UVA), and the commercial company TRE Canada. From available historical records, the research team selected data consisting of 46 radar images acquired by the Radarsat-1 satellite over the MMMBT area between December 4, 2001, and March 16, 2010. Actual acquisition dates and time intervals between acquisitions are listed in Figure 1. Significant gaps (no data available) in the archive records occur in 2003, 2007, and 2008. The study period coincided with increased awareness of the seawater infiltration, as documented by Parsons Brinckerhoff in 2009 (9). Remote sensing data were acquired from an orbit approximately 800 km above the Earth’s surface, with a side-looking radar beam
FIGURE 1 Radarsat-1 data used in historical analysis.
Hoppe, Kweon, Bruckno, Acton, Bolton, Becker, and Vaccari
oriented at 42° off the vertical. Managed by the Canadian Space Agency, the Radarsat-1 satellite was equipped with SAR operating in the C-band frequency (5.3 GHz) and moved in a quasi-polar orbit of approximately 100-min duration, providing nearly complete global landmass coverage every 24 days (10). Historical data were processed by TRE Canada with the SqueeSAR algorithm (11). Project deliverables included spatial data [e.g., georeferenced location, mean and annual deformation rates (velocities), total deformation, and time series elevation for every identified radar target at the MMMBT approaches] in shapefile (*.shp) format. TRE Canada provided a comprehensive report on data analysis and a summary of findings (12). In addition, TRE Canada produced an Internet-based portal designed for viewing analysis results using a web browser: TREmaps. Results were processed further by the UVA image processing and analysis group and VCTIR researchers. Additional statistical analyses were conducted on selected points of interest. Results and Discussion Area of Interest TRE Canada processed historical radar data over the approximately 100-km2 AOI, which includes the entire MMMBT complex and adjacent regions of Suffolk and Newport News (Figure 2). Radar scatterer locations identified by SqueeSAR analysis are superimposed on an orthophoto basemap in Figure 2. A total of 20,708 distinct points (scatterers) were located in the AOI, and the average density was 195 scatterers/km2. Temporal displacements were computed for each scatterer. The SqueeSAR algorithm identifies two scatterer
FIGURE 2 Area of interest [adapted from Bohane et al. (12)].
17
types: permanent, which represents displacement–time series at a specific location, and distributed, which represents a time series of an irregular area with a statistically uniform temporal deformation behavior. The horizontal positional accuracy of individual scatterers was determined to ±2 m (A. Bohane, personal communication, February 2014). Reference Point The SqueeSAR algorithm is a differential technique that compares the displacements of all identified scatterers with a single reference datum. Reference point selection involves a complex optimization procedure. A reference point is characterized by strong stability properties of the radar signal in amplitude and phase across all the SAR acquisitions. The scatterer with the lowest standard deviation and highest average amplitude values usually is chosen. The selected point effectively becomes a global reference for all reported displacements in the AOI. The reference point for the AOI is northwest of the tunnel entrance in the area of the Chesapeake and Ohio Railroad yard in Newport News at the coordinates 3521327.80 N, 12089691.15 E (NAD 1983 State Plane Virginia South). The point is approximately 2,530 m from the north MMMBT approach and 4,480 m from the south MMMBT approach. Displacements Results of the TRE Canada analysis indicated that the MMMBT approaches had low or negligible displacement rates during the study period; average velocity was +0.25 mm/year and +0.15 mm/year
18
Transportation Research Record 2510
(b)
(a) FIGURE 3 Boat section scatterers at north approach.
at the north and south approaches, respectively. UVA and VCTIR researchers subsequently carried out additional statistical analyses that focused specifically on the scatterers associated with boat sections. Scatterers were selected on the basis of their geolocation and data quality, as expressed by a metric called coherence. Coherence is a measure of the temporal stability of the radar characteristics of the
return signal expressed in a range of 0 to 1. The coherence data were supplied for each scatterer identified by the SqueeSAR analysis. Only the boat section scatterers with a coherence value of 0.8 (arbitrary threshold) or more were used. These criteria were satisfied by four locations at the north boat section (Figure 3) and 19 locations at the south boat section (Figure 4).
(b) (a) FIGURE 4 Boat section scatterers at south approach.
Hoppe, Kweon, Bruckno, Acton, Bolton, Becker, and Vaccari
Regression analysis results indicate that the mean rate of displacement was +0.32 to +0.80 mm/year for the north boat section and +0.27 to +0.42 mm/year for the south boat section, both at a 95% confidence level. Results also indicated a slight movement toward the satellite during the monitoring period. To resolve the ambiguity associated with the nonzero trend in displacement data and compare adjacent displacements more accurately, researchers decided to compare the behavior of tunnel boat sections with that of apparently stable areas nearby. Proximal scatterers were selected for comparison on the basis of their proximity, stability, and coherence and represent areas unassociated with constructed islands. The local reference selected for the north approach was at the loading terminal, approximately 635 m to the west (Scatterer A0VUI), and the one selected for the south approach was at the adjoining trestle bridge, approximately 500 m away (Scatterer A0VI9). Figures 5 and 6 show the locations and associated displacements of local reference points near the north and south approaches, respectively. Representative boat section scatterers are shown for comparison (A0VT0 and A0VOX). The results indicated that the boat sections were stable during the entire monitoring period. There was no evidence of a significant difference in the slopes of regression lines depicting displacements of the boat section and the adjacent local reference points. Additional statistical analyses using a linear
FIGURE 5 North approach and nearby displacements.
19
trend model were performed to compare all the boat section scatterers with the adjacent local reference points. The difference in linear regression slopes was not significant at the 90th percentile confidence level, implying statistically identical overall trends. The analyses of overall data trends assumed that the displacement rate over the 10-year monitoring period was a straight line. However, the trend may have been nonlinear. A local regression (loess) model was used to examine this possibility. The loess model is a nonparametric regression model that uses locally weighted linear regressions to smooth the data (13). Because it does not impose any functional form, the model is useful for detecting possible nonlinear trends. After trend shifts are identified, a piecewise linear regression model is used. The piecewise model fits a separate line for each shifted trend. Loess model analysis results for the south approach indicate a change in slope at approximately 2,000 days of monitoring, which corresponds to mid-2007 (Figure 7). A total of 42 measures made from 2001 through 2010 were used for each of the 19 sites; the solid line and gray band represent mean predictions and their 95th percentile intervals, respectively. Because characteristics were almost identical for nearby reference points, no significant displacement could be inferred for the tunnel approaches. Therefore, small nonzero displacement trends probably are caused by the artifacts associated with data processing.
20
Transportation Research Record 2510
FIGURE 6 South approach and nearby displacements.
Displacements at the North Helipad Many field-verified deformations are detected in the AOI, including settlement at the north approach helicopter landing pad, where the line of sight displacement values approach −20 mm near the end of the monitoring period (Figure 8). The asphalt pavement in this area experienced considerable distress in the past; on inspection, Virginia DOT personnel reported an undulating surface with numerous cracks (J. Deusebio, personal communication, June 2014).
Even though it was substantial, the helipad settlement cannot be considered indicative of boat section behavior. The settlement likely stems from the sequence of tunnel construction. The open-cut boat sections were constructed on preloaded constructed islands, but the fill adjoining the retaining walls and supporting nearby paved areas was placed at a much later date. Thus, in some places, compaction may have been inadequate, resulting in localized settlement. Figure 9 shows the tunnel approach during construction, with boat sections already established and the surrounding fill yet to be placed.
Settlement (mm)
10 5 0 –5 –10
0
1,000
2,000
Days Since 12/04/2001 FIGURE 7 Loess model predictions for south approach boat section.
3,000
Hoppe, Kweon, Bruckno, Acton, Bolton, Becker, and Vaccari
(a)
21
(b)
FIGURE 8 North helipad displacements.
Measurement Error TRE Canada reported that, on the basis of experience, the precision of individual InSAR measurements was within 5 mm (i.e., ±5 mm for any single measurement) (A. Bohane, personal communication, February 2014). To quantify this error empirically for the MMMBT study, measurements collected in uniform 24-day intervals (cor responding to the satellite revisit time) were extracted and analyzed,
FIGURE 9 Tunnel approach during construction.
resulting in 26 differenced measures (i.e., differences between the consecutive displacement measurements). Because a significant deformation is unlikely to have taken place in such a short time, the difference between two successive 24-day measurements can be attributed to a random systemic error. The distributions of differenced measures at the four sites on the north approach are shown in Figures 10 and 11. All sites have a zero median value at the 95th percentile confidence level, confirming the expectation of no significant change during 24 days. The t-test results indicated that mean values were statistically zero. Density plots imply that, in general, the data were distributed symmetrically around zero, approximating a normal distribution. To estimate the measurement error empirically, a standard deviation of the differenced measures was calculated for each site. Standard deviation was computed for the difference in two measures. If each measure is assumed to be normally distributed, then a standard deviation of the measure itself should be half of the standard deviation of the difference. Then, an empirical error can be calculated by multiplying the t-statistics corresponding to a desired confidence level and an appropriate degree of freedom by the standard deviation of the variable. The results of these calculations are presented in Tables 1 and 2 for the north and south approaches, respectively. For the north approach, the average empirical measurement errors were ±3.790 mm at the 95th percentile confidence level and ±5.128 mm at the 99th percentile confidence level (Table 1). The measurement error of an individual site varied from ±3.232 mm (A0VT0) to ±4.483 mm (A0VT6) at the 95th percentile confidence level and from ±4.373 mm (A0VT0) to ±6.065 mm (A0VT6) at the 99th percentile confidence level. A total of 26 differenced measures based on 27 measures made in a 24-day interval were used for each site. For the south approach, the average empirical measurement errors were ±3.133 mm at the 95th percentile confidence level and ±4.239 mm at the 99th percentile level (Table 2). At the 99th percentile confidence level, the measurement error of an individual site varied from ±2.528 (A0VPE) to ±6.605 (A0VN9); four sites had errors larger than ±5 mm (A0VM9, A0VMY, A0VN6, and A0VN9). At the
22
Transportation Research Record 2510
Site ID
A0VT6 A0VT0 A0VSE A0VSB –10
–5
0
5
10
Difference in Settlement (mm) FIGURE 10 Box plots of differenced measures at north approach. 0.15
Site ID
0.10 Probability
A0VSB A0VSE A0VT0 A0VT6
0.05
0.00 –10
–5
0
5
10
Difference in Settlement (mm) FIGURE 11 Density plots of differenced measures at north approach.
95th percentile confidence level, individual empirical measurement errors for all 23 sites in the north and south approaches were ±5 mm.
TABLE 1 Empirical Measurement Errors at North Approach
Site All sites A0VSB A0VSE A0VT0 A0VT6
Standard Deviation (mm)
Empirical Measurement Error (mm)
Differenced Measure
Measuresa
95th Percentile Confidenceb
99th Percentile Confidencec
3.680 3.827 3.492 3.139 4.353
1.840 1.914 1.746 1.569 2.176
±3.790 ±3.943 ±3.597 ±3.232 ±4.483
±5.128 ±5.334 ±4.866 ±4.373 ±6.065
Note: tdf = t-value corresponding to degree of freedom df; α = significance level. a Standard deviation of differenced measures ÷ 2. b Standard deviation of measures × tdf = 25, α = 0.05, where tdf = 25, α = 0.05 = 2.060. c Standard deviation of measures × tdf = 25, α = 0.01, where tdf = 25, α = 0.01 = 2.787.
Conclusions • The results of an analysis of historical satellite radar remote sensing data indicate no significant displacements of the tunnel boat sections during the study period. • The precision of annual displacement rate measurements for the tunnel boat sections is estimated to be ±1 mm/year at the 95th percentile confidence level. • The settlement of constructed islands is unlikely to have been a cause of ongoing water infiltration at the MMMBT approaches. • Satellite interferometric radar data analysis can be performed with individual measurement precision ±5 mm at the 95th percentile confidence level. • Comparisons with nearby reference points allow more accurate interpretation of InSAR analysis results. • Satellite InSAR technology can be used effectively to monitor long-term performance of transportation infrastructure.
Hoppe, Kweon, Bruckno, Acton, Bolton, Becker, and Vaccari
TABLE 2 Empirical Measurement Errors at South Approach
Site All A0VM9 A0VMT A0VMV A0VMY A0VMZ A0VN2 A0VN4 A0VN6 A0VN7 A0VN9 A0VOX A0VP0 A0VP4 A0VP5 A0VP8 A0VPA A0VPB A0VPE A0VPI
Standard Deviation (mm)
Empirical Measurement Error (mm)
Differenced Measure
Measuresa
95th Percentile Confidenceb
99th Percentile Confidencec
3.043 4.141 2.205 3.323 4.127 1.935 2.501 3.100 3.959 2.263 4.740 2.820 2.349 3.228 3.508 2.617 2.398 2.482 1.815 3.203
1.521 2.071 1.103 1.661 2.064 0.968 1.250 1.550 1.979 1.132 2.370 1.410 1.175 1.614 1.754 1.308 1.199 1.241 0.907 1.601
±3.133 ±4.266 ±2.272 ±3.422 ±4.252 ±1.994 ±2.575 ±3.193 ±4.077 ±2.332 ±4.882 ±2.905 ±2.421 ±3.325 ±3.613 ±2.694 ±2.470 ±2.556 ±1.868 ±3.298
±4.239 ±5.772 ±3.074 ±4.629 ±5.752 ±2.698 ±3.484 ±4.320 ±5.515 ±3.155 ±6.605 ±3.930 ±3.275 ±4.498 ±4.888 ±3.645 ±3.342 ±3.459 ±2.528 ±4.462
Standard deviation of differenced measures ÷ 2. Standard deviation of measures × tdf = 25, α = 0.05, where tdf = 25, α = 0.05 = 2.060. Standard deviation of measures × tdf = 25, α = 0.01, where tdf = 25, α = 0.01 = 2.787.
a b c
Acknowledgments The authors thank the personnel of the Virginia Department of Transportation, Hampton Roads District, for extensive support and guidance, particularly John Deusebio, James Peavy, Andrew Scott, Thomas Tate, and Danny Williams. The authors acknowledge support provided by the U.S. Department of Transportation for the ongoing parallel study designed to facilitate the implementation of commercial satellite InSAR technology on the U.S. transportation network. The authors also acknowledge extensive technical support provided by Adrian Bohane, Giacomo Falorni, and Vicky Hsiao of TRE Canada, Inc. Linda Evans of VCTIR assisted with the editorial process.
23
References 1. Vaccari, A., M. Stuecheli, E. Hoppe, B. Bruckno, and S. Acton. Detection of Geophysical Features in InSAR Point Cloud Data Sets Using Spatiotemporal Models. International Journal of Remote Sensing, Vol. 34, No. 22, 2013, pp. 8215–8234. 2. Calver, J. L. Geology of the Newport News South and Bowers Hill Quadrangles, Virginia. Virginia Division of Mineral Resources Report of Investigations 28. Richmond, 1971. 3. Plescia, J. B., D. L. Daniels, and A. K. Shah. Gravity Investigations of the Chesapeake Bay Impact Structure. Special Paper of the Geological Society of America, No. 458, 2009, pp. 181–193. 4. Galloway, D. L., and T. J. Burbey. Regional Land Subsidence Accompanying Groundwater Extraction. Hydrogeology Journal, Vol. 19, No. 8, 2011, pp. 1459–1486. 5. Poag, C. W., D. S. Powars, L. J. Poppe, R. B. Mixon, L. E. Edwards, D. W. Folger, and S. Bruce. Deep Sea Drilling Project Site 612 Bolide Event: New Evidence of a Late Eocene Impact-Wave Deposit and a Possible Impact Site, U.S. East Coast. Geology, Vol. 20, 1992, pp. 771–774. 6. Hayden, T., M. Kominz, D. S. Powars, L. E. Edwards, G. G. Miller, J. V. Browning, and A. A. Kulpecz. Impact Effects and Regional Tectonic Insights: Backstripping the Chesapeake Bay Impact Structure. Geology, Vol. 39, 2008, pp. 327–330. 7. Poag, C. W., D. R. Hutchinson, and S. M. Colman. Seismic Expression of the Chesapeake Bay Impact Crater: Structural and Morphologic Refinements Based on New Seismic Data. Geological Society of America Special Papers, Vol. 339, 1999, pp. 149–164. 8. Waterproofing Repair Recommendations for Route 0664 Over Hampton Roads. Whitman, Requardt, and Associates, LLP, Richmond, Va., May 29, 2013. 9. Russell, H. A., Jr. Monitor Merrimac Memorial Bridge Tunnel— Groundwater Intrusion in Boat Sections. Memorandum to Philip C. Rinehart. Parsons Brinckerhoff, Boston, Mass., Nov. 18, 2009. http:// media2.wavy.com/html/PDFs/BoatWallLeakAssessment.pdf. Accessed Nov. 5, 2014. 10. Satellite Characteristics. Canadian Space Agency, Ottawa, Ontario, Canada. Jan. 21, 2011. http://www.asc-csa.gc.ca/eng/satellites/radarsat /radarsat-tableau.asp. Accessed July 27, 2014. 11. Ferretti, A., A. Fumagalli, F. Novali, C. Prati, F. Rocca, and A. Rucci. A New Algorithm for Processing Interferometric Data-Stacks: SqueeSAR. IEEE Transactions on Geoscience and Remote Sensing, Vol. 49, No. 9, 2011, pp. 3460–3470. 12. Bohane, A., G. Falorni, and V. Hsiao. Historical InSAR Analysis: Analysis of Ground Deformation over the Merrimac Tunnel Approaches Between 2001 and 2010. Report No. JO13-3034-REP1.0. TRE Canada, Inc., Vancouver, British Columbia, Canada, 2014. 13. Cleveland, W. S. Robust Locally Weighted Regression and Smoothing Scatterplots. Journal of the American Statistical Association, Vol. 74, No. 368, 1979, pp. 829–836. The views, opinions, findings, and conclusions reflected in this report are the responsibility of the authors. The contents do not represent the official policies or positions of the Virginia Department of Transportation, the Commonwealth Transportation Board, or the U.S. Department of Transportation. The Standing Committee on Engineering Geology peer-reviewed this paper.
