Relationship of land use and elevated ionic strength in Appalachian watersheds



Coal mining activities have been implicated as sources that increase stream specific conductance in Central Appalachia. The present study characterized potential sources of elevated ionic strength for small subwatersheds within the Coal, Upper Kanawha, Gauley, and New Rivers in West Virginia. From a large monitoring data set developed by the West Virginia Department of Environmental Protection, 162 < 20–km2-watersheds were identified that had detailed land cover information in southwestern West Virginia with at least one water chemistry sample. Scatter plots of specific conductance were generated for nine land cover classifications: open water, agriculture, forest, residential, barren, total mining, valley fill, abandoned mine lands, and mining excluding valley fill and abandoned mine lands. Conductivity was negatively correlated with the percentage of forest area and positively associated with other land uses. In a multiple regression, the percentage of area in valley fill was the strongest contributor to increased ionic strength, followed by percentage of area in urban (residential/buildings) land use and other mining land use. Based on the 10th quantile regression, 300 µS/cm was exceeded at 3.3% of area in valley fill. In most catchments, HCOmath image and SOmath image concentrations were greater than Cl concentration. These findings confirm coal mining activities as the primary source of high conductivity waters. Such activities might be redressed with the goal of protecting sources of dilute freshwater in the region. Environ. Toxicol. Chem. 2013;32:296–303. © 2012 SETAC


Coal mining activities have been implicated as causes of adverse biological effects in Central Appalachia, which are potentially associated with bicarbonate and sulfate salts that increase stream-specific conductance (hereafter referred to as conductivity) 1–6. The concentration of dissolved ions has been recognized as a key physiological determinant of the distribution of aquatic organisms, but the ionic constituents are also important 7–9. In freshwater Appalachian streams, 5% of genera are extirpated when ion concentration exceeds 295 µS/cm 10. One mechanism appears to be the disruption of bicarbonate gradients that maintain pH and ionic homeostasis 8, 9. The predominant ionic mixture that raises conductivity levels above background in Central Appalachia contains much greater concentrations of HCOmath image/COmath image, as well as SOmath image, Cl, Ca2+, and Mg2+ 5, 10. In the present study, a regional analysis was undertaken to determine the contribution of different types of land cover to high ionic strength from mixtures of predominately HCOmath image/COmath image and SOmath image anions.

Mineral extraction in the region has been shown to be a source of predominately HCOmath image/COmath image, Cl, and SOmath image anionic mixture 1, 5, 6. Some of these sources include surface and underground coal mining, effluent from coal preparation plants and associated slurry impoundments, and effluent from coal fly ash impoundments 3, 11, 12. Besides coal mining, other sources that could increase ionic strength in the study region include road construction, winter road maintenance, treatment of waste water, human and animal waste, scrubbers at coal-fired electric plants, soil amendments and fertilizers, natural gas and coal-bed methane extraction, and construction run–off. The dominant anion associated with these sources, however, is Cl (Table 1) 12–14.

Table 1. Dominant ions associated with different sources
SourceDominant ionsReference
  • a

    Deep coal mine discharges can have higher chloride levels compared to those from surface coal mines 30.

Crushed rock (e.g., surface coal mining)aCa2+, Mg2+, HCOmath image, Cl, SOmath image5, 30
Wastewater treatment plantsNa+, Cl, NHmath image, NOmath image, POmath image


Road saltNa+, Cl, Ca2+, Mg+


Salt water intrusionNa+, Cl


Produced water from natural gas and coal bed methane productionNa+, Ca2+, Mg+-, Cl, HCOmath image


Agricultural runoffNa+, NHmath image, NOmath image, POmath image (irrigation water related ions may vary)


The present paper is one of a series related to a new method that uses field data to develop water quality benchmark values. The series begins by describing the method for deriving the benchmark 15 and a separate paper applies that method to determine the relationship between dissolved ions and the extirpation of benthic invertebrates 10. Another paper in this special section explains how to determine whether a field-derived exposure–response relationship is confounded and assesses potential confounders of the conductivity–extirpation relationship 16. Two papers in this special section first describe a method to determine whether an association observed in the field is causal 17 and then the method is used to show that increased concentrations of ions have caused the extirpation of benthic invertebrates in Appalachian streams 9. One type of evidence for assessing causes of extirpation of stream invertebrates is to demonstrate that there are sources that can and do increase dissolved ions in Appalachian watersheds. Developing that evidence is the focus of the present paper.

The extent of land use alteration associated with surface mining is greater in Appalachia than anywhere else in the United States 3. Our analysis of land use/land cover characterizes ionic strength associated with land uses and estimates the relative contribution of surface coal mining as a source of ions in streams. The present study also provides information to estimate expected conductivity levels associated with the percentage of watershed area filled with crushed rock; that is, stream valleys filled with the overburden from blasting and removing mountaintops to access underlying coal. This form of coal mining is called mountain top mining with valley fill construction.


General approach

Small watersheds were delineated from the upland to the pore point, and the proportions of land cover types were regressed against water quality chemical parameters. Watershed size was limited to <20 km2 and from headwater to pore point ranged between 1.85 to 19.53 km for the smallest to largest catchment based on stream length. Catchments represent true watersheds 18. Watershed size was kept small to minimize the variety of land use and cover types within a single watershed, thereby providing a clearer signal for each potential source of ionic strength. Because the region has a long history of mining, however, and land cover information may not include legacy mining, persistent effects of legacy mining are potentially present even when no current record of past or present mining activity exists in publically available land cover databases. Also, buildings and roads are present in areas where mining occurs. Therefore, there are potential influences from multiple sources in most of the 162 watersheds, but these are minimized by using smaller subwatersheds.

These smaller subwatersheds were delineated from the larger basins of the Coal, Upper Kanawha, Gauley, and New Rivers of Ecoregion 69D (Dissected Appalachian Plateau) in West Virginia (Fig. 1) 19. Selecting the subwatersheds was based on the availability of at least one water chemistry sample and detailed land cover information. Water quality parameters were obtained from the West Virginia Department of Environmental Protection's (WVDEP) Watershed Assessment Branch database. The number of water samples from each station varied from 1 to 18 samples (median number of samples was 12 with 161 out of 162 stations sampled at least twice). In a prior analysis 10, we examined the seasonal patterns of conductivity with at least one water quality sample taken in the first and last halves of the year. On average, the mean conductivity between January and June was less than the mean conductivity between July and December. Although we recognized that this would increase variability in our measurement of water chemistries, we chose to maximize the overall sample size and calculated the average of the sample values for each station and compared the average with land uses. Land use information came from many sources (Table 2). No new data were collected for the present study.

Figure 1.

Sampling locations used to develop evidence of sources of high conductivity inputs. The 162 stations (black dots) at the terminus of each <20–km2 catchment are shown within the larger 8–digit hydrologic unit catchments (HUC) in southwestern West Virginia. Major rivers are depicted as solid lines draining northward.

Table 2. Publicly available GIS data used to generate land cover estimates
Source TypeData description
  1. AML = abandoned mine lines; DMR = Division of Mining Reclamation; GAP = Gap Analysis Program; GIS = geographic information system; NHD = National Hydrography Data Set; NLCD = National Land Cover Database; NPDES = National Pollutant Discharge Elimination System; TMDL = total maximum daily load; USGS = U.S. Geological Survey; WVDEP = West Virginia Department of Environmental Protection

General sources of land use/land cover information

General West Virginia Universities GIS data repository location 20

was accessed at

General WVDEP's GIS data repository location 21

was accessed available at

Coal River TMDL Land use spreadsheet appendix 22

was accessed at

Upper Kanawha River TMDL Land use spreadsheet appendix 23

is was accessed

Gauley River TMDL Land use spreadsheet appendix 24

was accessed

New River TMDL Land use spreadsheet appendix 25

was accessed at
Base land use/land

GAP land use 35

was accessed from at

NLCD 2001 land use 36

was accessed from
Watershed boundary data sets

USGS 8–digit hydrologic unit code boundaries 37

was accessed from
NHD streams

National Hydrography Data Set streams 38

was accessed from
Abandoned mine lines (AML–highwalls) and polygons (AML areas)

West Virginia abandoned mine lands coverages. Highwall mine coverage and AML area 39

was accessed
DMR mining NPDES permits and outlets

WVDEP Office of Mining and Reclamation NPDES permit and outlet coverages 40

was accessed from
Mining related fills, southern West Virginia

WVDEP valley fills coverage from 2003 41

was accessed from
Mining permit boundaries

WVDEP mining permit boundaries 21

was accessed from
Roads paved2000 TIGER/Line GIS and WV_roads shapefiles 20, 42 was accessed from Available at and from
Roads unpaved2000 TIGER/Line GIS shapefile and digitized from aerial photographs and topographic maps 20, 43 was accessed from and

Geographical Information Systems (GIS) data descriptions

Numerous GIS data sets are available for the state of West Virginia. One such repository for data, the West Virginia GIS Technical Center (WVGISTC) 20, maintains publicly available shapefiles. The West Virginia Department of Environmental Protection 21 also maintains a publicly available repository of statewide GIS data sets ( All relevant GIS metadata are available for the data housed at each repository site. All GIS coverages are in or were converted to universal Transverse Mercator 1983 Zone 17, with the units in meters. Table 2 describes some of the publicly available GIS shapefiles that were originally used to develop base files for WVDEP's program to remediate waters listed as impaired. We used the WVDEP's land cover data from this program as the starting point to select stations for the analyses (Table 3). The area in valley fill is from a 2003 coverage that the WVDEP developed with ground-truthing performed; note that some water samples were taken in 2004. Both the analytical sample and land use data are from a discrete time span (2001 through 2004) and accurately reflect the existing chemistry and land use at that time.

Table 3. Detailed land use category derivation and land use derivation
Detailed WV TMDL land use categoryData sourceBase land use from which new source area was subtracted

Land use categories used in scatter plots in Fig. 2

  • a

    Valley fill land use was not part of the base TMDL land use and was specifically incorporated into the detailed land use analysis.

    See Table 2 for the source file.

    AML = abandoned mine line; LU = land use; TMDL = total maximum daily load; WQ = water quality; WV = West Virginia; WVDEP = West Virginia Department of Environmental Protection; NA = not available.

WaterWater—base LU coverageNAWater
WetlandWetland—base LU coverageNAWater
ForestForest—consolidated all forested types from base LU coverageNAForest
GrasslandGrassland—base LU coverageNAAgriculture
CroplandCropland—consolidated all cropland types from base LU coverageNAAgriculture
Urban perviousUrban—consolidated urbanized types from base LU coverageNAUrban/residential
Urban imperviousUrban—consolidated urbanized types from base LU coverageNAUrban/residential
BarrenBarren—base LU coverageNABarren
PastureWVDEP source trackingNew area subtracted from grasslandAgriculture
Paved roadsRoads shapefilesNew area subtracted from urban imperviousUrban/residential
Unpaved roadsRoads shapefilesNew area subtracted from urban perviousUrban/residential
Revoked mining permitsAML informationNew area subtracted from BarrenAML
Abandoned mine landAML shapefileNew area subtracted from barrenAML
QuarryMining shapefileNew area subtracted from barrenMining
HighwallAML shapefileNew area subtracted from barrenMining
Oil and gasOil and gas shapefileNew area subtracted from barrenMining
Surface mine water quality permitsMining shapefileNew area subtracted from barrenMining
Surface mine technology permitsMining shapefileNew area subtracted from barrenMining
Comingled mine deep ground gravity dischargeMining shapefileNew area subtracted from barrenMining
Comingled mine deep ground pump dischargeMining shapefileNew area subtracted from barrenMining
Undeveloped surface mine WQ permitsMining shapefileNew area subtracted from forestMining
Undeveloped surface mine technology permitsMining shapefileNew area subtracted from forestMining
Undeveloped comingled mine gravity dischargeMining shapefileNew area subtracted from forestMining
Undeveloped comingled mine pump dischargeMining shapefileNew area subtracted from forestMining
Burned forestForestry Dept. informationNew area subtracted from forestBarren
Harvested forestForestry Dept. informationNew area subtracted from forestBarren
Skid roadsForestry Dept. informationNew area subtracted from forestBarren
TMDL land use considers valley filla area as part of the surface mine water quality and technology permit informationWVDEP valley fills coverage from 2003New area subtracted from mining, barren, and forest, as appropriateValley fill

Catchments with available data that met the needs of the analysis involved a multistep selection process that resulted in 162 small watershed catchments. The steps were performed in sequence. From the WVDEP Watershed Assessment Branch database, we selected all small catchments (≤20 km2) located within Ecoregion 69D that were sampled between 2001 and 2004. We then removed sites with an average pH <6, thereby focusing the present study on sources of increased ionic concentrations in the neutral to alkaline range. We selected chemistry stations that coincided with an appropriate scale of land use that were already determined by previous WVDEP efforts. Catchments were located in the Coal, Upper Kanawha, Gauley, and New River watersheds. The total number of water chemistry stations and catchments within Ecoregion 69D is 162 (Fig. 1).

Land use analysis

Our land use analysis began by first obtaining the WVDEP's existing electronic land use information in spreadsheet format for the following WVDEP studies: Coal 22, Upper Kanawha 23, Gauley 24, and New River 25. These land uses were used as the starting point for further analysis. The WVDEP's land uses were created originally by consolidating the available base land use GIS raster files (GAP Analysis Program 2000 or National Land Cover Data 2001) into more general categories and then adding more detailed source land use categories (e.g., mining, oil and gas, and roads) from detailed source information such as permits or field verification. Table 3 contains the land use categories, the data source from which the extent of the area and its location were determined, and the base land use from which any newly created land use categories were subtracted. In brief, nine land use categories were generated: (1) total percentage area in mining (% total mining), which is the sum of % abandoned mine, % valley fill, and % mining; (2) percentage in mountaintop mining valley fill (% valley fill); (3) percentage of abandoned mine lands (% abandoned mine); (4) percentage of mining (% mining), which is % total mining minus % valley fill and % abandoned mine; (5) percentage barren land use (% barren); (6) percentage of residences, buildings, and roads (% urban); (7) percentage in agriculture, pasture and grassland (% agriculture); (8) percentage in forest (% forest); and (9) percentage in open water (% water).

Because the WVDEP land use characterization process has been revised and enhanced over the years, the land use data sets for the Upper Kanawha, Coal, Gauley, and New Rivers were normalized to have equivalent land use classifications. This yielded seven basic land use categories for the 162 sampling stations. The valley fill GIS coverage was then incorporated into the land use characterization by subtracting the valley fill acreage 26 from the mining land use category. If more area was present in the valley fill coverage than was present in the original mining area for each subwatershed, the remainder was subtracted from forest. The nine land use categories calculated for each of the 162 Watershed Assessment Branch database sampling stations used seven categories consolidated from the land use (Table 3) and then included the addition of the valley fill area. The % total mining category is the sum of the % mining, % valley fill, and % abandoned mine land categories. The % mining land use represents all known types of mining activities minus % abandoned mine and % valley fill areas.

Water quality data correspond to land use percentages at the time of sampling from 2001 through 2004. Mining areas present at the time of land use characterization are based on active mining permits obtained from the WVDEP. Subsequent permits that became active after the land use characterizations were developed are not used in our analysis because using them would over estimate mining land use and would not accurately reflect the condition of the water quality at the time of sampling.

Statistical analysis

Summary statistics were computed for physical–chemical variables. Environmental variables were logarithm-transformed as appropriate to obtain normal distributions.

The percentage of catchment area for nine land cover classes were determined including open water (% water), agriculture (% agriculture), forest (% forest), urban/residential/buildings (% urban), barren (% barren), total mining (% total mining), valley fill (% valley fill), abandoned mine lands (% abandoned mine), and mining (% mining) (excluding valley fill and abandoned mine lands). All but one land use class (% forest) were transformed (log10 + 1) to normalize the data distribution.

Scatter plots between conductivity and the nine land cover classes were generated and their correlations were estimated using Spearman correlation. The trends of relationships between conductivity and each land use class were visualized using locally weighted scatterplot smoothing (span = 0.75). We also used ordinary least square regression and quantile regression to model the relationship between % valley fill and conductivity levels. Quantile regression and ordinary least square models were used to predict the % valley fill associated with conductivity. A multiple regression model was also evaluated, because multiple predictors, that is, land uses, could contribute to increased ionic strength in streams.


Characterization of catchments and ionic matrix

The 162 small watersheds used in the present analysis are located near the borders of the 8–digit hydrologic unit codes where elevations are greater and headwaters of these small perennial streams are located (Fig. 1). The ionic composition of these waters is not uniform, but bicarbonate and sulfate are usually greater than chloride (Table 4) 5, 10. Because we were interested in all ions as well as the mixture, we did not exclude high Cl sites.

Table 4. Summary statistics of water quality parameters in the 162 catchments
ParameterUnitsMin25th percentileMedian75th percentileMaxMeanValid n
  1. TSS = total suspended solids; Mean = geometric mean except for temperature, pH, and dissolved oxygen (DO).

Al, totalmg/L0.,170
Al, dissolvedmg/L0.,169
Ca, totalmg/L1.9314.123.458.310323.425
Cu, totalmg/L0.0010.0030.0030.0040.010.00316
Cu, dissolvedmg/L0.0010.0030.0030.0041.910.00519
Fe, totalmg/L0.,170
Fe, dissolvedmg/L0.,165
Mg, totalmg/L1.285.618.428.988.912.225
Mn, totalmg/L0.0030.0210.090.29927.30.0961,169
Se, totalmg/L0.0010.0050.0050.0050.0450.005395
Zn, totalmg/L0.0050.0050.
Zn, dissolvedmg/L0.0050.0080.010.0120.7260.01319
pHstandard units4.,479

Correlations with in–stream water quality parameters

Pairs of land use and anionic water quality parameters from >25 stations and at least one Spearman's correlation coefficient with an r > |0.50| are listed in Table 5.

Table 5. Spearman correlation coefficients between pairs of land use and anionic water quality parameters in the land use data seta
Water quality parameter% Valley fill% Total mining% Mining% Forestn
  • a

    Parameters yielding r < |0.50| and sample sizes of less than 30 are not shown.


Mining-related land uses are strongly correlated with each other and negatively correlated with % forest but not with other land uses. The % total mining is strongly correlated with % mining (r = 0.89) and the % valley fill (r = 0.69), and negatively correlated with % forest (r = −0.88).

At relatively low residential land use, the range of conductivity is highly variable. In contrast, there is a clear pattern of increasing conductivity as percentage of area in valley fill increases and of decreasing conductivity with increasing forest cover (Fig. 2). The scatter plots illustrate that there are clear sources of increased conductivity, but that the percentage of area in valley fill has the strongest correlation with conductivity (r = 0.66), and percentage of mining without a valley fill has a moderate correlation (r = 0.42). Multiple regression analysis using the three land use categories that could potentially contribute to rising conductivity in streams (% valley fill, % mining, and % urban), showed that all three variables are strong predictors (p < 0.01). Notably, the slope of % valley fill is the steepest (0.5), followed by % urban (0.21) and % mining (0.1), indicating that % valley fill has the strongest contribution to ionic strength in these streams (Table 6).

Figure 2.

Geometric mean conductivity associated with different land uses in 162 watersheds in Ecoregion 69D and Spearman's correlation coefficient. Conductivity increases with increasing % valley fill and % total mining, and decreases with increasing % forest, but other land uses are either less clear or show no pattern. From left to right, they are (A) % total mining (percentage of deep, surface, quarry mining, valley fill, and abandoned mine land); (B) % valley fill (from mountaintop mining [MTM] overburden); (C) % abandoned mine; (D) % mining (inclusive of all types of mining except valley fill and abandoned mine), (E) % barren, (F) % urban, (G) % agricultural, (H) % forest, and (I) % water. All but one land use class (% forest) were transformed (log10 + 1) to normalize the data distribution, and were then were transformed back to the original scale to inspect trends conveniently. The fitted lines are the locally weighted scatter plot smoothing lines with span set at 0.75.

Table 6. Regression coefficients between % land uses (log10 + 1 transformed) and conductivity (log10 transformed) in the land use data set. All % land uses included in the equations are significant at p = 0.01 level.
Water quality parameterEquationr2
  1. OLS = ordinary least square; VF = valley fill

All data (OLS, n = 162)Cond = 2.42 + 0.58VF0.45
Samples removed with 0% valley fill (OLS, n = 78)Cond = 2.41 + 0.59VF0.56
25th quantile regression lineCond = 2.25 + 0.67VF 
10th quantile regression lineCond = 2+ 0.75VF
Multiple regressionCond = 2.28 + 0.50VF + 0.21Urban + 0.1Mining0.49

Assuming that the lower conductivity values represent the best achieved with current practice, we modeled the lower 10th and 25th quantile of the percentage of area in valley fill against conductivity (Fig. 3, Table 6). For comparison, the intercept for the 10th and 25th quantile regressions at 300 µS/cm are 3.3% and 1.2% valley fill, respectively. The mean model based on samples minus those with 0% valley fill shows that the relationship is unaffected by removing the sites without valley fills. Because these estimates assesses the percentage of area and do not consider the size of the mined area, volume of the fill, construction practices, distance from the fill, or dilution from tributaries, the estimate of conductivity associated with percentage of valley fill is useful as a general characterization but will vary for specific cases.

Figure 3.

Ordinary least square regression and quantile regression of percentage of area in valley fill and conductivity in 162 small watersheds in Ecoregion 69D. The percentage of area in valley fill was first log10 + 1 transformed to perform regression analysis and then was transformed back to show the pattern in the original scale. Assuming the lowest conductivity points represent some of the best fill construction practices, the 10th and 25th quantile regression lines are shown. The intercepts for 300 µS/cm are 3.3 and 1.2% valley fill and for 500 µS/cm are approximately 7.6 and 3.7% valley fill for the 10th and 25th quantiles, respectively. The mean model based on samples minus those with 0% valley fill shows that the relationship is unaffected by the removal of sites without valley fills. See Table 6 for formulae for regression lines.


Conductivity typically increases with increasing land use, 27, but in Ecoregion 69D, the densities of agricultural and urban land cover are relatively low and associations of these land uses and conductivity are not strong. In contrast, there are clear patterns of increasing conductivity as the percentage of valley fill area increases and of decreasing conductivity with increasing percentage of forest cover area (Fig. 2). Of the land uses in the small watersheds analyzed in the Upper Kanawha, Coal, Gauley, and New Rivers, mining associated with valley fills is strongly associated with the dissolved ions that are measured as conductivity. Although the presence of buildings (% urban) was also a strong predictor in a multiple linear regression, it is not necessarily the source of increasing ion concentrations in streams. Where there is active mining, there are more buildings. However, the type of ions associated with urban land uses differs (i.e., Cl dominated), from that of coal mining land use (i.e., HCOmath image, Cl, and SOmath image dominated). The present study provides evidence of at least one strong source of high conductivity in the region and is consistent with similar reports 1–6, 9, 11, 28.

The recommended benchmark for ionic strength measured as conductivity is 300 µS/cm 10. At 300 µS/cm, 5% of genera are extirpated. For comparison, at 500 µS/cm, 17% of genera are extirpated 10. The 300 µS/cm benchmark has been met in some catchments with valley fills, albeit with a small percentage area in fill (Fig. 3; 0.3% on average and 3.3% at the 10th quantile). Additional analyses is needed to determine if the low conductivity levels at these specific locations are associated with construction design, with the size of the mined area, with dilution from tributaries, or high flow conditions, or other factors.


Table S1. (22 KB DOC)


We thank the West Virginia Department of Environmental Protection, Division of Water and Waste Management, Watershed Branch, Total Maximum Daily Load and Watershed Assessment Section for providing their data. We would also like to recognize and thank the numerous field personnel for collecting samples throughout the state. Without their effort, this research would not be possible. We also thank M. McManus, D. Petersen, H. Lattimer, and three anonymous reviewers. The work was supported by the U.S. Environmental Protection Agency. The views expressed in the present paper are those of the authors and do not necessarily represent the views or policies of the U.S. Environmental Protection Agency or any other persons or groups.