Development of a Ground-Water-Vulnerability Data Base for the Southeastern United States Using a Geographic Information System
By John S. Clarke and Jerry W. Sorensen
GIS-derived values of hydraulic conductivity, depth to water, precipitation, evaporation, and net precipitation were compared to corresponding values assigned by EPA Field Investigation Teams (FIT) to test the accuracy of the coverages at 28 hazardous-was te sites. Site data for the hydraulic-conductivity and depth-to-water factors were divided into the Coastal Plain, Valley and Ridge/Interior Low Plateau, and Piedmont/Blue Ridge physiographic provinces to determine if differences in hydrologic setting aff ected the comparison.
GIS-derived values for evaporation, precipitation, and net precipitation showed good correlation to FIT data values, having median differences of 0, 1.96, and 0 percent, respectively. The GIS-derived values for depth to water generally were higher than FI T values, having median differences ranging from about 33 to about 40 percent higher. GIS-derived values for hydraulic conductivity showed median differences that ranged from 0 to about 2 orders of magnitude.
This paper describes development of a computerized data base consisting of seven factors that influence ground-water vulnerability: (1) lithology of soil and rock; (2) hydraulic conductivity of soil and rock; (3) sorptive capacity of soil and rock; (4) de pth to water; (5) annual precipitation; (6) annual lake evaporation; and (7) net precipitation. The HRS uses factors 2, 3, 4, and 7 to rank relative vulnerability. The pilot study area, which includes Alabama, Florida, Georgia, Kentucky, Mississippi, Nort h Carolina, South Carolina, and Tennessee, is located within EPA Region IV ( fig. 1).
To create GIS map coverages, published maps were digitized as polygons and assigned either specified values or ranges of values. Within each map coverage, adjacent polygons having the same value for a given attribute were joined, and common attribute boun daries were removed, thus creating a series of larger composite polygons. Values for each of the seven ground-water-vulnerability factors were assigned to 28 hazardous-waste sites to evaluate the accuracy of the GIS-derived map coverages. The values were assigned to each site using GIS to overlay each location on the composite map coverage.
Hydraulic conductivity represents the potential rate at which geologic materials can transmit ground water; whereas sorptive capacity reflects the potential for geologic materials to chemically adsorb hazardous substances and, therefore, retard hazardous substance migration. Hydraulic-conductivity and sorptive-capacity values were estimated based on the lithology of soil and rock (U.S. Environmental Protection Agency, 1988, p. 85, 88).
Values derived from the geologic map and input to a GIS map coverage are shown in figures 4 and 5, and values derived from the surficial-deposits map are shown in figures 6 and 7. Because bedrock is exposed or near land surface in some areas, composite coverages were generated to show the average hydraulic conductivity and average sorptive capacity for both bedrock and overlying materials ( figs. 8, 9). To generate the composite coverages, polygons for the geologic and surficial-deposits coverages were joined, and ranges of values in the geologic coverages were averaged with ranges of values in the surficial-deposits coverages. Note that the resolution of this composite coverage is only as good as the resolution of the smaller scale surficial- deposits coverage (1:14,000,000).
Depth to aquifer represents the distance that hazardous substances must travel to reach an aquifer. Because the evaluation was limited to the uppermost aquifer, depth to aquifer was considered to be equal to the water-table depth. Water-level data from 20 ,309 wells stored in NWIS were compiled to create a point coverage of water-table depth (Clarke and others, 1989) ( fig. 10).Wells selected from NWIS were limited to those less than 100 feet deep because these wells were assumed to represent water-table co nditions. Where multiple measurements for an individual well were encountered, the mean of the data was used for plotting purposes. Because measurements used for the point coverage were collected over a wide range of years, time periods were not designate d, and seasonal fluctuations were not accounted for. Well-data density was highest in Kentucky (about 0.04 well per square mile (mi)) and lowest in South Carolina and Tennessee (about 0.0003 well per square mi).
By using GIS proximity analysis, water-table depths for hazardous-waste sites were calculated by averaging water-table depths for wells located within a designated radius. To determine the relative accuracy of averaging water depths at increasing distance s from sites, wells were selected with 10-, 15-, and 20-mi radii from each site. Intervals of less than 10 mi were not evaluated separately because well data were limited. Well data were not available for waste sites 1, 4, 5, 16, 22, 25, and 27 within 20 mi of the site; and thus, water-table values were not estimated.
Net precipitation, as used in the HRS, is the amount of water potentially available to recharge the ground-water system on an annual basis. Net precipitation is derived from the difference between annual rainfall and annual lake evaporation and, thus, doe s not account for overland runoff or for transpiration from plants. National Oceanic and Atmospheric Administration (NOAA) maps (U.S. Department of Commerce, 1968) showing mean annual precipitation for the period 1931-60 (scale, 1:10,000,000) and mean ann ual lake evaporation for the period 1946-55 (scale, 1:20,000,000) were digitized into GIS coverages ( figs. 11 , 12). Net precipitation was calculated using GIS by subtracting the evaporation coverage from the precipitation coverage ( fig. 13). Polygons for the two coverages were joined, and ranges of values in the evaporation coverage were subtracted from ranges of values in the precipitation coverage. Some apparent discrepancies on the net-precipitation coverage exist because the plots of the precipitation and evaporation coverages shown in figures 11 and 12 use ranges of values instead of contours, which were too numerous to show on the figures.
Differences between values derived by FIT and GIS were computed and designated as percentages for precipitation, evaporation, net-precipitation, and water-level data. Positive differences indicate that GIS-derived values were greater than FIT values, wher eas negative values indicate GIS-derived values were lower than FIT values. The distributions of these differences were computed for the 10th, 25th, 50th, 75th, and 90th percentiles and plotted as boxplots ( figs. 14, 15). The interquartile range (IQR) of the computed differences represents the range of values between the 25th and 75th percentiles (lower and upper quartiles, respectively) and, thus, gives an indication of the spread of observations.
GIS-derived data for evaporation had a median difference from FIT data of 0 percent and an IQR of -2.35 to 2.17 percent ( fig. 14). The median difference for precipi- tation was 1.96 percent and had an IQR of -3.78 to 4.45 percent. Net precipitation had a median difference of 0 percent, but the IQR was comparatively high, ranging from -24.04 to 28.2 percent. The larger percentage difference for the net-precipitation data reflects the combined error of the precipitation and evaporation coverages.
GIS-derived values for depth to water were compared to FIT data on the basis of the search radius used to compute average water depth ( fig.15). GIS-derived water levels for the 20-, 15-, and 10-mi search radii were generally higher than FIT values ( fig. 15). For all sites, the median difference was similar for each search radius, ranging from about 33 percent for the 15-mi radius to about 40 percent for the 20-mi radius. The IQR, however, decreased with progressively smaller search radii, possibly indicat ing less variation in water levels at closer distances to a given site. For each search radius, values for sites located in the Valley and Ridge/Interior Low Plateau province were the most accurate. The median differences for sites located in this area we re about 40 percent for the 20-mi radius, about 3 percent for the 15-mi radius, and about 16 percent for the 10-mi radius. The range of differences, as indicated by the IQR, was greatest for the Piedmont/Blue Ridge province, possibly the result of the mor e complex nature of ground-water flow in fractured crystalline-rock settings.
To facilitate comparisons between data sets for the hydraulic-conductivity factor, values were first converted to logarithms; and then the differences in order of magnitude (OM) between the FIT and GIS-derived data were computed ( fig. 16). The logarithmic conversion and comparison are considered valid because the HRS scoring ranges for hydraulic conductivity are generally within 2 OM. For all sites, the median difference was 1.7 OM for the geologic map coverage and 0 OM for both the surficial-deposits and composite coverages. Although the median difference was the same, the IQR for the surficial-deposits coverage (-1 to 1 OM) showed less spread than the composite coverage (-1.52 to 2.23 OM), indicating a better match between the data sets. The correlation between GIS and FIT data sets was better for values derived from the surficial-deposits coverage than for values derived from the geologic coverage, possibly because FIT hydraulic-conductivity values were derived largely from the soils at a site without considering the underlying bedrock.
When hydraulic-conductivity data were grouped on the basis of physiographic province, the smallest differences between FIT and GIS-derived data were (1) the surficial-deposits coverage in the Piedmont/Blue Ridge province, having a median difference of 0 O M and an IQR of -0.5 to 0.5 OM; (2) the geologic coverage in the Coastal Plain province, having a median difference of 0 OM and an IQR of -0.77 to 1.68 OM; and (3) the surficial-deposits coverage in the Valley and Ridge/Interior Low Plateau province, havi ng a median difference of -1 OM and an IQR of -1 to 1 OM. Thus, for two of the three physio-graphic groupings, the best matches between data sets were the surficial-deposits coverages.
GIS-derived data sets were evaluated at 28 hazardous-waste sites. For each site, GIS-derived values for precipitation, evaporation, net precipitation, depth to water, and hydraulic conductivity were compared to values assigned by EPA Field Investigation T eams (FIT).
The GIS coverages for evaporation and precipitation were generated from published maps an show good correlation to FIT data, having median differences of 0 and 1.96 percent, respectively. A GIS coverage for net precipitation was derived by computing the d ifference between the precipitation and evaporation maps. For net precipitation, the median difference was 0 percent; but a large range of differences resulted, which reflects the combined errors of the precipitation and evaporation coverages.
GIS coverage for depth to water was computed using well data from the USGS NWIS, whereby an average water level was computed for wells within a designated radius of a hazardous-waste site. GIS-derived water levels for 20-, 15-, and 10-mi search radii were generally higher than FIT values, ranging from about 33 percent for the 15-mi radius to about 40 percent for the 20-mi radius. The range of differences, however, decreased with progressively smaller search radii, possibly indicating less variation in wat er levels at closer distances to a given site. For each search radius, values for sites located in the Valley and Ridge/Interior Low Plateau physiographic province were the most accurate.
Three GIS coverages for hydraulic conductivity were generated from (1) a published geologic map, (2) a published surficial-deposits map, and (3) a composite of the two maps. The median differences between the FIT and GIS-derived data sets were 1.7 orders of magnitude (OM) for the geologic map coverage and 0 OM for both the surficial-deposits coverage and the composite coverage. Although the median difference was the same, the range of differences for the surficial-deposits coverage showed less spread than the composite coverage, indicating a better match between the data sets.
When hydraulic conductivity data were grouped on the basis of physiographic province, the smallest differences between FIT and GIS-derived data were (1) the surficial-deposits coverage in the Piedmont/Blue Ridge province, having a median difference of 0 O M; (2) the geologic coverage in the Coastal Plain province, having a median difference of 0 OM; and (3) the surficial-deposits coverage in the Valley and Ridge Interior/Low Plateau province, having a median difference of -1 OM. Thus, for two of the three physiographic groupings, the best matches between data sets were the surficial-deposits coverages.
Return to Bulletin 2016 contents