Environmental Management

, Volume 34, Issue 1, pp 62–74

A Watershed-Scale Model for Predicting Nonpoint Pollution Risk in North Carolina


    • Department of ForestryNorth Carolina State University
  • Frederick W. Cubbage
    • Department of ForestryNorth Carolina State University
  • Gary B. Blank
    • Department of ForestryNorth Carolina State University
  • Rex H. Schaberg
    • Nicholas School of the Environment and Earth Sciences, Duke University

DOI: 10.1007/s00267-004-0117-7

Cite this article as:
Potter, K., Cubbage, F., Blank, G. et al. Environmental Management (2004) 34: 62. doi:10.1007/s00267-004-0117-7


The Southeastern United States is a global center of freshwater biotic diversity, but much of the region’s aquatic biodiversity is at risk from stream degradation. Nonpoint pollution sources are responsible for 70% of that degradation, and controlling nonpoint pollution from agriculture, urbanization, and silviculture is considered critical to maintaining water quality and aquatic biodiversity in the Southeast. We used an ecological risk assessment framework to develop vulnerability models that can help policymakers and natural resource managers understand the impact of land cover changes on water quality in North Carolina. Additionally, we determined which landscape characteristics are most closely associated with macroinvertebrate community tolerance of stream degradation, and therefore with lower-quality water. The results will allow managers and policymakers to weigh the risks of management and policy decisions to a given watershed or set of watersheds, including whether streamside buffer protection zones are ecologically effective in achieving water quality standards. Regression analyses revealed that landscape variables explained up to 56.3% of the variability in benthic macroinvertebrate index scores. The resulting vulnerability models indicate that North Carolina watersheds with less forest cover are at most risk for degraded water quality and steam habitat conditions. The importance of forest cover, at both the watershed and riparian zone scale, in predicting macrobenthic invertebrate community assemblage varies by geographic region of the state.

Nonpoint source pollutionEcological risk assessmentAquatic ecosystemsLand use planningWater qualityForest cover

The Southeastern United States is a region with extensive forestlands and high-quality aquatic resources, despite the extensive loss of forest and the degradation of water quality since European settlement (West 2002). Because of the region’s diverse physical geography, a favorable climate, and a dynamic natural history (Chaplin and others 2000), rivers and streams of the Southeast harbor an extraordinary variety of life, including 864 rare aquatic species distributed among seven taxonomic groups — fish, mussels, snails, insects, crustaceans, reptiles, and amphibians (Herrig and Shulte 2002).

In 1998, 45% of the total river miles assessed by Southeastern states — totaling 210,010 km (103,441 miles) — was classified as impaired by some form of pollution or habitat degradation (West 2002). Between 1988 and 1998, 70% of stream degradation in the Southeast was caused by nonpoint source pollutants. Controlling nonpoint pollution from its most significant sources — urban development and agriculture — is considered critical to maintaining future water quality in the Southeast (West 2002). Nonpoint pollution can have negative consequences for the integrity of aquatic biota by altering physical habitat, modifying seasonal water flow, altering the systemic food base, contaminating water with toxic chemicals, and modifying interactions among organisms (Chen and others 1994, Karr 1999). Thus, resident biota of aquatic ecosystems serve as continuous monitors of cumulative effects on those systems, and are often used as endpoints in ecological risk assessments (US Environmental Protection Agency 1990, Zandbergen 1998, Bryce and others 1999, Diamond and Serveiss 2001).

Studies have found strong positive relationships between diverse assemblages of stream benthic macroinvertebrates that are intolerant of water quality degradation and watershed-wide forested land cover (Lenat and Crawford 1994, Stewart and others 2001, Weigel and others 2003) or forested land cover within riparian zones (Basnyat and others 1999, Sponseller and others 2001, Stewart and others 2001, Weigel and others 2003). Meanwhile, research has shown less diverse and more intolerant macrobenthic communities to be correlated with agricultural land cover (Lenat and Crawford 1994, Richards and others 1996, Weigel and others 2000, Genito and others 2002) and urban land use (Lenat and Crawford 1994, Morley and Karr 2002, Morse and others 2003, Roy and others 2003, Volstad and others 2003, Wang and Kanehl 2003). Current land cover, however, may not be as important as past land use in predicting current stream invertebrate diversity, because sustained anthropogenic disturbance such as agriculture or urban development could profoundly alter biotic communities with lasting effects (Harding and others 1998).

These findings have important implications in the Southeast, where the amount of urban area is expected to increase from about 20 million acres in 1992 to 55 million acres in 2020 and 81 million acres in 2040 (Wear 2002). Much of this newly developed land will likely be converted from forest cover, as has happened in the recent past: 406,000 acres of forest was converted to urban uses in 1990, the latest year for which Forest Inventory and Analysis land conversion data are available (Conner and Hartsell 2002).

In this study, we used geographic information systems (GIS) databases to examine which landscape characteristics at multiple scales — the entire watershed, and riparian zones 30.5 m (100 feet) along both sides of streams — are the best predictors of stream macroinvertebrate community intolerance. The objectives of this research were (1) to investigate the importance of land cover, especially the amount of forest at the watershed and riparian zone scales, to benthic macroinvertebrate community composition; and (2) to develop vulnerability models to help policymakers and natural resource managers understand the impact of land cover changes on water quality in North Carolina. The results will allow managers and policymakers to weigh the risks of management and policy decisions on a given watershed or set of watersheds.


Ecological risk assessment holds promise as a tool for comparing the relative risks of human activities to water quality (Chen and others 1993, Zandbergen 1998), including land use (Diamond and Serveiss 2001, Hunsaker and others 1992). It is a process used to systematically evaluate and organize data, information, assumptions, and uncertainties to better understand and predict relationships between stressors and ecological effects in a fashion useful for environmental decision-making (US Environmental Protection Agency 1998a). Risk assessment requires using formal quantitative techniques to estimate probabilities of effects on well-defined endpoints, and to estimate uncertainties associated with the analysis. It is also clearly separated from the process of choosing among alternatives and determining the acceptability of risks, which is the responsibility of policymakers (Suter 1993).

The central objective of ecological risk assessment is to estimate the possibility of adverse impacts on ecosystems from exposure to such sources of stress as pollutants (Boroush 1998). It should provide a quantitative basis for balancing and comparing risks associated with environmental problems, and a systematic means of better estimating and understanding those risks (Graham and others 1991). An integrated and retrospective approach to ecological risk assessment is particularly well suited to evaluating multiple and cumulative effects of chemical and nonchemical stresses at regional scales (Gentile and Slimak 1992), where cumulative effects are often more apparent (Graham and others 1991).

To facilitate extrapolation among watersheds within a region and across regions, ecological risk assessments should be based on models describing the underlying factors that influence watershed response to sources of stress, particularly vulnerability to change as a result of those stresses (Detenback and others 2000). After assembling such a model for this study, we measured 10 land cover and land form characteristics for 73 North Carolina watersheds, several at both the entire-watershed scale and 30.5-m (100-foot) riparian scale (Tables 1 and 2). These variables were percent forested, percent agricultural, percent developed, precipitation, watershed area, watershed shape, relief ratio, topographic complexity, mean elevation, and soil clay content. Of these, the three land cover variables were also measured for 30.5 m on either side of streams, because the preservation of streamside vegetation is the only water quality best management practice for which we had spatial data at a watershed scale. We used ArcView 3.2 (ESRI 1999) to delineate the watersheds and assemble riparian zone data, and ArcGIS 8.1 (ESRI 2001) to project all the files to a common coordinate system: North Carolina State Plane, 1983 North American Datum. For each catchment, the watershed files and the riparian zone files were converted to grids using ArcView 3.2 (ESRI 1999), and then used to mask the statewide land cover data. To derive the three land cover variables — percent forested, percent agricultural, and percent developed — we reclassified the 15 categories in the 1992 30-m Landsat thematic mapper raster data for the state of North Carolina (Multi-Resolution Land Characterization Consortium 2000).
Table 1.

Variables used in the analyses of stream invertebrate tolerance to stream degradation (table describes each variable included in the analyses, its expected effect on water quality, and the source of the data)



Expected relationship

Data source

NC Division of Water quality, Biological Assessment Unit Data (1999, 2000b, c, d, e, f, 2001a, b, c, 2002a, b)


Macroinvertebrate index scores (response)

1) N.C. Biotic Index (NCBI) 2) EPT Biotic Index (Ephemeroptera, Plecoptera, and Trichoptera)


N.C. Division of Water Quality, Biological Assessment Unit data (1999, 2000b, 2000c, 2000d, 2000e, 2000f, 2000g, 2001b, 2001c, 2002a, 2002b)

Land cover (predictor)

Percent of watershed or riparian zone forested agricultural, developed

More forest correlated with better macrobenthos scores

Multi-Resolution Land Characterization Consortium (2000)

Precipitation (predictor)

Amount of rainfall during the 3 months before stream biota sample

More precipitation correlated with better macrobenthos scores

NOAA National Weather Service data (US Department of Commerce, 1999)

Watershed area (predictor)

Watershed area (km2)

Larger watersheds correlated with better macrobenthos scores

USEPA stream reach GIS data and USGS digital elevation model (1998b,c)

Watershed shape index (predictor)

Horton’s Form Factor (area/square of watershed length)

More compact watersheds (smaller index) correlated with less erosion, and better macrobenthos scores

USEPA stream reach GIS data and USGS digital elevation model (1998b,c)

Watershed slope/relief ratio (predictor)

Watershed elevation change/watershed length from outlet to highest point on perimeter

Less relief correlated with better macrobenthos scores

USEPA stream reach GIS data and USGS digital elevation model (1998b,c)

Topographic complexity (predictor)

Standard deviation of watershed elevation, based on 30-m raster digital elevation model

Less topographic complexity correlated with better macrobenthos scores

USEPA stream reach GIS data and USGS digital elevation model (1998b,c)

Mean elevation (predictor)

Mean elevation of watershed, based on 30-m raster digital elevation model

Greater mean elevation correlated with better macrobenthos scores

USEPA stream reach GIS data and USGS digital elevation model (1998b,c)

Clay content of soil (predictor)

Percent of watershed with soil having at least 25% clay content

Lower clay content in soil correlated with better macrobenthos scores

USGS State Soil and Geographic database (Schwarz and Alexander, 1995)

Table 2.

Mean, standard deviation, maximum and minimum values for the macrobenthic invertebrate index scores and for each of the variables used in the analyses of 73 North Carolina watersheds






N.C. Biotic Index scores





EPT Biotic Index scores





Percent forest (watershed)





Percent forest (riparian)





Percent agriculture (watershed)





Percent agriculture (riparian)





Percent developed (watershed)





Percent developed (riparian)





Mean elevation (m)





Soil clay content (%)





Precipitation (cm)





Watershed area (km2)





Watershed shape index





Watershed relief ratio (m/km)





Macrobenthic Invertebrate Indices

The macrobenthic invertebrate organisms associated with the substrates of rivers, streams, and lakes are considered especially good indicators of water quality (Karr 1999, Metcalfe-Smith 1994, Lenat 1993, Plafkin and others 1989, Hilsenhoff 1987). Because macrobenthic invertebrates are less mobile than other types of aquatic organisms, their often-diverse communities respond to a wide variety of pollutants. These communities can reflect both short-term and long-term trends because many species have life cycles longer than a year (N.C. Department of Water Quality 2001a, Metcalfe-Smith 1994). Additionally, macrobenthic invertebrate community structure reflects nonpoint source pollution from silvicultural activities because these organisms are abundant in low-order streams where much timber harvesting occurs, and because they are sensitive to habitat and water quality changes (Adams and others 1995).

Regression analyses allowed us to consider which landscape characteristics have a significant impact on two metrics describing the tolerance of benthic macroinvertebrates to stream degradation: the North Carolina Biotic Index (NCBI) and an index of Ephemeroptera (mayfly), Plecoptera (stonefly), and Trichoptera (caddisfly) tolerance (EPTBI).

The NCBI was developed by Lenat (1993) to examine the general level of pollution at stream sites; his approach was modeled after the biotic index Hilsenhoff (1987) created for use in Wisconsin streams and rivers. The NCBI rates stream sites based on the water quality tolerance of the macroinvertebrates sampled at the site; Lenat determined each taxon’s level of pollution tolerance by studying more than 2000 macroinvertebrate stream samples collected in North Carolina between 1983 and 1992. Taxa found in higher numbers at degraded sites were given higher tolerance values, because they tolerate higher levels of pollutants; taxa found in greater numbers at less degraded sites were assigned lower tolerance values. The NCBI score for a site measures the degradation tolerance of the benthic macroinvertebrates, relative to their abundance. EPTBI scores (N.C. Division of Water Quality 2001a) are restricted to three invertebrate taxa considered highly sensitive to water quality degradation: the aquatic nymphs of Ephemeroptera and Plecoptera, and the larvae of Trichoptera.

The NCBI is considered a more reliable indicator of stream chemistry and habitat quality than of in-stream sediment (N.C. Division of Water Quality 2001a). The EPTBI is believed to be a better measure the impact of sediment in streams (David Lenat, personal communication), but is considered a less reliable measure of water quality, especially at high elevation sites and sites at which low numbers of EPT organisms are collected.

For this study, we assembled a complete list of NCDWQ benthic macroinvertebrate sampling sites and scores from across North Carolina, available from the Biological Assessment Unit (NC Division of Water Quality 2002a), and from the latest basinwide water quality assessment reports, which include all the macrobenthic index results collected since 1983 (NC Division of Water Quality 1999, 2000a, 2000b, 2000c, 2000d, 2000e, 2000f, 2001b, 2001c, 2002b). From this list of 252 regularly sampled sites, we selected 73 by excluding those with watersheds extending into other states, those with watersheds larger than 2500 km2 and those without both NCBI and EPT Biotic Index invertebrate samples taken between 1990 and 1994. Additionally, when watersheds associated with the sampling sites were nested, we selected one of the watersheds at random to include in the analyses. The 73 watersheds (Figure 1) included in the analyses cover an area of 37,704.6 km2, or 29.9% of North Carolina’s total land area.
Figure 1.

Seventy-three North Carolina watersheds were delineated and analyzed in this study. A risk characterization example was conducted on an additional watershed, Swift Creek, in the central part of the state.

Data Analyses

We conducted a series of simple linear regressions (SAS Institute Inc. 2000) to examine the direction and strength of the relationship between the landscape variables (transformed when necessary for a normal distribution of model residuals) and the two macrobenthic invertebrate tolerance indices. Because the NCBI and EPTBI measure tolerance for degraded stream conditions, a positive regression coefficient in the model indicated that the landscape variable has a negative correlation with water quality, whereas a negative coefficient has a positive relationship.

We next conducted a stepwise multiple regression analyses (Draper and Smith 1998) to analyze the proportion of variability in the stream invertebrate tolerance indices attributable to the most statistically significant land form and land cover variables at the watershed scale. We used an α-level of 0.05 as the significance level for the partial F entry test and for the partial F removal test. The resulting coefficient of multiple determination (R2) for each regression equation indicates the proportion of variability in the invertebrate tolerance indices attributable to the landscape variables included in the model. Before running regressions on multiple variables, however, we eliminated landscape variables that exhibited high (>0.8) multicollinearity with other variables (Table 3). These were the percent of watershed agricultural land cover (almost the inverse of the percent watershed forested), the percent of riparian zones developed (correlated with the development of the entire watershed), and the percent of riparian zones in agricultural land cover (highly negatively correlated with the percent of riparian zones forested).
Table 3.

Matrix of the correlations among landscape variables included in the analyses (underlined numbers indicate correlations greater than 0.8)


% WS For.

% Rip. For.

% WS Ag.

% Rip. Ag.

% WS Dev.

% Rip. Dev.

% Clay

WS Area

WS Shape

WS Topo.

WS Mn. El.

% WS For.












% Rip. For.











% WS Ag.










% Rip. Ag.









% WS Dev.








% Rip. Dev.







% Clay






WS Area





WS Shape




WS Topo.



WS Mn. El.


WS Relief

*Not statistically significant at p = 0.05.

Ag = agricultural, Dev. = developed, For. = forested, Mn. El. = mean elevation, Rip. = riparian, Topo. = topography, WS = watershed.


The simple regression analysis found statistically significant relationships between several landscape variables and the NCBI and EPTBI scores; the exceptions were the percent of riparian zones developed, watershed area, watershed shape, and soil clay content (Table 4). Two of the three watershed land cover variables — percent agricultural and percent forested — exhibited somewhat strong relationships. The percent of agriculture land cover at the watershed scale had a positive relationship with the indices, meaning that it was negatively correlated with aquatic ecological integrity. The percent of forest was correlated with better stream conditions. For both land cover types, the relationship was strongest with the NCBI. The percent of developed land cover also had a statistically significant relationship with the indices, but the r2 value was considerably lower. It showed a stronger relationship with the EPT Biotic Index, and was correlated with degraded stream conditions.
Table 4.

Sign of regression coefficient values from the simple regression models, the associated coefficients of determination (r2), and the implication for water quality for the relationships between the landscape variables and the macrobenthic invertebrate tolerance indicesa

Landscape variable




Water quality







% Watershed forest




% Riparian forest




% Watershed agricultural (sqrt)





% Riparian Agricultural ([y + 1]0.25)





% Watershed developed (ln + 1)





% Riparian developed (ln + 1)*





Rainfall (ln)




% Clay*





Watershed area (ln)*





Watershed shape*





Watershed topographic complexity (ln)




Watershed mean elevation (x0.25)




Watershed relief ratio (ln)




*Not statistically significant at p < 0.05.

aThe riparian variables are for 100-foot streamside zones. When a variable was transformed to ensure a normal distribution in model residuals, the transformation is included in parentheses.

EPTBI = Ephemenoptera, Plecoptera, Trichoptera Biotic Index; NCBI = North Carolina Biotic Index.

The land form feature exhibiting the strongest correlation with invertebrate indices was topographic complexity. Mean elevation and relief ratio, which are closely related to topographic complexity, demonstrated relationships nearly as strong. These variables have a negative relationship with macrobenthic invertebrate tolerance to stream degradation, indicating that greater topographic complexity, relief ratio, and mean elevation are each associated with better water quality. Nonpoint pollution effects on aquatic ecosystems resulting from land use changes may be mitigated or exacerbated by the terrain, or topographic complexity may simply be a better indicator of human disturbance than the land cover variables. Specifically, because greater topographic complexity was related to better stream conditions, steep terrain in some areas may prevent agricultural and urban development that could result in degraded water quality. Conversely, terrain with flatter topography may have been more extensively and permanently altered by human activities, including the hydrologic modification that results from ditching and tiling in agricultural landscapes. Topographic complexity, relief ratio, and mean elevation were not included in the multiple regression analyses because of their high positive correlation with watershed forest cover and high negative correlation with watershed agricultural land use (Table 3); land cover variables are of greater interest to policymakers than landscape characteristics that cannot be altered.

Rainfall had a statistically significant relationship with macrobenthic invertebrate community structure. This correlation was also negative, meaning that greater amounts of rainfall were more likely to be accompanied by better water quality and aquatic habitat conditions.

Multiple Regression Results

The stepwise regression models explained 56.3% of the variability in the NCBI and 50.3% in the EPT Biotic Index (Table 5). One land cover variable (watershed percent forested) and one land form variable (watershed shape) were common to both indices. In both regression models, watershed percent forested was positively related to intolerant macroinvertebrate community structure, and therefore with better water quality. The opposite is true with watershed shape, a dimensionless measure of watershed elongation, defined as
Table 5.

Multiple regression results analyzing the relationships between the landscape variables and the macrobenthic invertebrate indicesa

North Carolina Biotic Index


pr > F






Landscape variable

p value

Reg. coeff.

Stand. coeff.

Percent watershed forested




Watershed shape




EPT Biotic Index


pr > F






Landscape variable

p value

Reg. coeff.

Stand. coeff.

Percent watershed forested




Watershed shape




Watershed area (In)




aThe standardized (stand. coeff.) regression coefficient estimates indicate the relative importance of the landscape characteristic relative to the other landscape variables in the regression model. When a variable was transformed to ensure a normal distribution in model residuals, the transformation is included in parentheses.

where A is the area of the watershed, and l is the length of the watershed from its outlet to the farthest point parallel to the main river channel. Higher values indicate greater roundness in watershed shape, whereas lower values occur with greater elongation parallel to the main channel. Greater circularity in watershed shape, therefore, had a positive relationship with tolerant macroinvertebrate taxa, and a negative relationship with water quality. This was expected, because runoff from rounder watersheds tends to concentrate and reach the mouth more quickly and with greater erosive power (Ward 1995, Brooks and others 1997). The only other significant variable — watershed area — was included in the EPT model. Like watershed percent forested, this variable is positively correlated with better water quality.

For both indices, watershed percent forest was the relative strongest predictor of macroinvertebrate community assemblage, as noted by the standardized regression coefficients. In other words, the more forested a North Carolina watershed was, the more likely it was to have a benthic macroinvertebrate community assemblage indicative of higher-quality stream habitat. For the EPTBI, the strength of the relationship between forest cover and water quality was somewhat reduced by the shape of the watershed.

Physiographic Region Results

Simple regression analysis indicated that the importance of forest cover, both at the watershed and riparian zone scale, varied by physiographic region (Table 6). Percent forest cover at the watershed scale was a significant predictor of macrobenthic community assemblage in the Piedmont for both indices, indicating that more forest was correlated with better water quality. Forest cover in riparian zones was a significant predictor for the NCBI in the Piedmont, and for both indices in the Coastal Plain. Forest cover was not a significant water quality predictor at either scale in the Southern Appalachians, most likely because 90% of both watersheds and riparian zones were forested; the correlation between forest cover at the two scales was also very high in this region (Table 6). The correlation was smaller in the other two regions, because the mean percent of forest in riparian zones was considerably higher than the mean percent of forest cover for entire watersheds.
Table 6.

The number of watersheds analyzed by physiographic region, the correlation between percent forest cover at the watershed and riparian zone scales, the mean and standard deviation of forest cover at the two scales, and the coefficients of determination (r2) describing the relationship between forest cover and the two indices of macrobenthic invertebrate community assemblage

Physiographic region


Watershed–riparian forest correlation

% Watershed forest

% Riparian forest











Southern Appalachians






















Coastal Plain











*Not statistically significant at p < 0.05.

For abbreviations see Table 4.


Results of this study indicate that land cover in North Carolina is a significant predictor of macrobenthic invertebrate tolerance of water quality degradation. The amount of forest and agricultural cover at the watershed scale explained the most variability in macrobenthic invertebrate assemblage, although the same variables at the riparian scale (within 30 m of streams) were also important. Development throughout the watershed was less important, but still statistically significant, whereas the amount of development in riparian areas was not.

These results are consistent with similar studies. Roy and others (2003), working in the Etowah River basin in Georgia, found that forested land cover at the watershed and riparian scale, and urban land cover for entire watersheds, explained about 30% of the variability in several macrobenthic invertebrate indices. Weigel and others (2003) linked the percent of forest at the riparian and watershed scales with intolerant macroinvertebrates. Research by Richards and others (1996) in southwestern Wisconsin suggested that macroinvertebrate community assemblages were strongly influenced by land-use characteristics — positively by the amount of forested wetlands, and negatively by row-crop agriculture. Sponseller and others (2001), using data from the Roanoke River basin of southern Virginia, indicated that local, upstream patches of riparian forest might have a positive effect on macroinvertebrate community structure. Harding and others (1998), also using stepwise regression, found that invertebrate richness and measures of diversity were significantly greater in forested streams than in agricultural streams in the Little Tennessee River and French Broad River basins of western North Carolina. Lenat and Crawford (1994) concluded that land use in North Carolina Piedmont watersheds strongly influenced invertebrate communities. The forested stream in their study had high invertebrate taxa richness, a low biotic index, and many unique species, resulting in a “good” water quality classification. Measurements of the same criteria indicated moderate stress (fair water quality) at an agricultural site and severe stress (poor water quality) at an urban site.

In our statewide analysis, the percent of forest cover at the watershed scale and in riparian zones were highly correlated enough (0.776) that the two have similar value as predictors of macroinvertebrate tolerance for water quality degradation. Correlations among landscape features and indicators of water quality, however, have been found to vary among physiographic regions (Bryce and others 1999, Herlihy and others 1998). This was the case with our North Carolina study (Table 6), where watershed and riparian forest cover was very highly correlated for the Southern Appalachians, but not for the Piedmont and Coastal Plain, where the percent of forest was higher within 30 m of streams than for entire watersheds. Not surprisingly, the usefulness of these land-cover variables also varied as predictors of macroinvertebrate assemblage. Forest cover was not significant at either scale in the mountains, whereas whole-watershed forest cover was a better predictor in the Piedmont, and riparian zone forest cover was a better predictor in the Coastal Plain. This difference may be explained in part by the somewhat larger percentage of watershed-wide forest cover in the Piedmont (68.37% compared to 59.57% in the Coastal Plain), and by the greater amount of agricultural land cover in the Coastal Plain watersheds (38.25% compared to 25.9% for Piedmont watersheds). Because agriculture was a more widespread land use in the Coastal Plain, riparian forest, where it existed, may have helped to minimize the impact of agriculture-related nonpoint pollution in the form of sedimentation (Cooper and others 1987) and nutrient runoff (Peterjohn and Correll 1984, Lowrance and others 1985, Jones and others 2001).

The statewide stepwise regression models explained between 50.3% and 56.3% of the variability in the EPTBI and NCBI macroinvertebrate index scores (Table 5). Watershed forest cover was the most important variable in both models, which is not surprising, considering the strong relationship between forest cover and index scores indicated by the simple regression results (Table 4). More surprising was the inclusion in the models of watershed shape, a dimensionless measure of watershed roundness that had a positive relationship with tolerant macroinvertebrate taxa and a negative relationship with water quality. Circularity in watershed shape, therefore, appears to somewhat lessen the positive impact of forest cover on water quality. Watershed roundness tends to concentrate runoff, which reaches the watershed outlet with greater erosive power (Ward 1995, Brooks and others 1997), perhaps resulting in stream and riparian habitat less hospitable to intolerant macroinvertebrates. Additionally, the velocity of runoff is greater with more circular watersheds, resulting in a shorter delivery time of diffuse pollutants to the mouth of the watershed, and a lessened ability of the watershed to “self-purify”itself of a portion of the pollutant load (Ha and Bae 2001).

The inclusion of watershed area in the EPTBI model may indicate sensitivity of EPT data to the greater cross-sectional area of the channel that occurs with larger watersheds (Richards and others 1996). This would be consistent with the river continuum concept (Vannote and others 1980), which predicts that the influence of riparian zones and other physical characteristics in a river system decreases as the size of the river increases.

The regression models generated by this study could assist in better understanding which landscape characteristics are statistically likely to have the greatest impact on stream invertebrate assemblages. Additionally, reducing uncertainty is a key component of ecological risk assessment (Suter 1993), and models with fewer parameters are less likely to contain excessive uncertainty from data collection and variance in the regression model (Draper and Smith 1998).

Risk Characterization

Although ecological risk assessment has been used mainly to estimate the risks posed by chemicals introduced into the environment (Boroush 1998), it is also an important and promising methodology to aid in efforts to control water pollution (Chen and others 1993). Ecological risk assessment is an integrative approach that balances the complexity of scientific analysis with land managers’ need for clear and simple answers about the condition of the watershed and the actions needed to achieve certain objectives (Zandbergen 1998).

Results of the current ecological risk assessment suggest that North Carolina watersheds with the least forest cover, and the most agricultural land cover, are most at risk for degraded water quality and steam habitat conditions. Several other land use and land form characteristics had statistically significant relationships with macrobenthic invertebrate tolerance of water quality degradation, including watershed-wide development and riparian zone forest and agricultural land cover. Only watershed shape (for both indices) and watershed area (for the EPTBI), however, were statistically significant components in the models that best predict macroinvertebrate community assemblage.

A useful approach for characterizing risk of potential land management options is to use regression models to “simulate” land use activities, such as change in land cover. The equations allow managers and others to determine the expected existing condition of water quality at any freshwater stream sampling site in North Carolina, by measuring only a few landscape characteristics associated with the site’s catchment. As an example, consider the 463.89 km2 Swift Creek watershed (Figure 1). Located in the Piedmont on the southern edge of the rapidly expanding Raleigh urban area, this is one of several headwater catchments in the Neuse River basin. The Swift Creek watershed has a watershed shape index of 0.208; in 1992, the watershed was 63.77% forested. Using the coefficients in Table 5 and the observed Swift Creek data, the predicted NCBI value is 5.604 (±0.295). The actual value of 5.91 is just outside the 95% confidence interval. When simulating a decrease of forest from 63.77% to 50%, the resulting predicted NCBI score would be 6.265 (±0.35). That result is consistent with the decrease in water quality expected to accompany a decrease in forest cover.

Because the regression model explains only 56% of the variability in the NCBI score, the result will not predict an exact score for a given location. It will, however, offer a value that allows decision-makers to ascertain how severe a change in water quality they risk relative to the watershed’s original condition.

This risk characterization approach could be further improved with analyses predicting the amount of land-use change that will occur during a given time frame in the area for which policymakers or land managers are making management and policy decisions. Those values could easily be incorporated into the vulnerability model framework to better assess the possible change in macroinvertebrate index scores, and therefore the expected change in water quality and aquatic ecological integrity.


This ecological risk assessment process generated vulnerability model equations that can provide a basis for quantitatively comparing, ranking, and prioritizing risks to water quality. Such an assessment can be useful in cost-benefit and cost-effectiveness analyses of alternative management options (US Environmental Protection Agency 1998a). Specifically, the model equations offer a useful approach for characterizing the risk of potential land management options through the “simulation” of land use activities.

There are limits to the value of the empirical approach used to assemble these vulnerability model equations. Such statistical models require large empirical databases and identify correlations, but do not generally demonstrate a cause-and-effect relationship. Although they are important tools for estimating uncertainties, they have limited value for making predictions across scales of biological organization and for untested stresses (Gentile and Slimak 1992). This assessment, however, was limited to only one scale of biological organization — that of benthic macroinvertebrate assemblage — and focused on nonpoint source pollution stresses that researchers have long and consistently understood to negatively impact those communities. Additionally, the goal of this research was to predict variability in macroinvertebrate community structure, not to establish specific cause-and-effect relationships.

This project yielded several interesting results about the relationship between landscape characteristics — including land-use — and benthic macroinvertebrate community structure. This study yielded three general results: (1) Forested land cover, at both the watershed and riparian scales, was a statistically significant predictor of benthic macroinvertebrate communities that are less tolerant of stream degradation, and that indicate a greater level of aquatic ecological integrity and better water quality. The opposite was the case for agricultural land cover at the watershed and riparian scales, and developed land cover in riparian zones. (2) One land cover characteristic (watershed percent forested) and one land form feature (watershed shape) were consistently the most important and most statistically significant variables in explaining macroinvertebrate variability in statewide multiple regression analyses. (3) The importance of forest cover in predicting macrobenthic invertebrate community assemblage varied by the physiographic region in which a watershed was located. The amount of forest cover in riparian zones was a significant predictor of intolerant macroinvertebrate taxa in the Coastal Plain. In the Piedmont, watershed forest cover was a significant predictor for both the NCBI and the EPTBI, and riparian forest cover was a significant predictor for the NCBI. Forest cover was not a good predictor in the Southern Appalachians.

Based on these findings, it appears that water quality and stream ecological integrity, as measured by benthic macroinvertebrate community structure, may be most at risk in North Carolina watersheds that have less forest and more agricultural land cover, and are more circular in shape. Maintaining forested riparian buffers may be a more successful policy strategy for maintaining water quality in the Coastal Plain than in the rest of the state, whereas attempting to maximize the amount of forestation for entire watersheds may be a better approach in the Piedmont.


We thank Dr. George Hess for his comments on the manuscript; Dr. Marcia Gumpertz, Amy Nail, and Mark Atlas for their constructive statistics advice; Dr. Jim Gregory for instruction on watershed hydrology and stream ecology; and David Lenat for his insights about the use of the benthic macroinvertebrate indices. Finally, we appreciate the helpful comments from Alan Herlihy and two anonymous reviewers. This project was supported by the US Environmental Protection Agency through Science to Achieve Results (STAR) grant 2000-STAR-K3 (EPA Agreement Number R828784). It has not been subjected to EPA review and therefore does not necessarily reflect the views of EPA, and no official endorsement should be inferred.

Copyright information

© Springer-Verlag New York, Inc. 2004