Relationships between land use, spatial scale and stream macroinvertebrate communities


Sponseller Department of Biology, Arizona State University, Tempe, Arizona 85287-1501, AZ, U.S.A. E-mail:


1. The structure of lotic macroinvertebrate communities may be strongly influenced by land-use practices within catchments. However, the relative magnitude of influence on the benthos may depend upon the spatial arrangement of different land uses in the catchment.

2. We examined the influence of land-cover patterns on in-stream physico-chemical features and macroinvertebrate assemblages in nine southern Appalachian headwater basins characterized by a mixture of land-use practices. Using a geographical information system (GIS)/remote sensing approach, we quantified land-cover at five spatial scales; the entire catchment, the riparian corridor, and three riparian ‘sub-corridors’ extending 200, 1000 and 2000 m upstream of sampling reaches.

3. Stream water chemistry was generally related to features at the catchment scale. Conversely, stream temperature and substratum characteristics were strongly influenced by land-cover patterns at the riparian corridor and sub-corridor scales.

4. Macroinvertebrate assemblage structure was quantified using the slope of rank-abundance plots, and further described using diversity and evenness indices. Taxon richness ranged from 24 to 54 among sites, and the analysis of rank-abundance curves defined three distinct groups with high, medium and low diversity. In general, other macroinvertebrate indices were in accord with rank-abundance groups, with richness and evenness decreasing among sites with maximum stream temperature.

5. Macroinvertebrate indices were most closely related to land-cover patterns evaluated at the 200 m sub-corridor scale, suggesting that local, streamside development effectively alters assemblage structure.

6. Results suggest that differences in macroinvertebrate assemblage structure can be explained by land-cover patterns when appropriate spatial scales are employed. In addition, the influence of riparian forest patches on in-stream habitat features (e.g. the thermal regime) may be critical to the distribution of many taxa in headwater streams draining catchments with mixed land-use practices.


Maintenance of stream biodiversity in the face of encroaching human development has received much attention in recent years (e.g. Allan & Flecker, 1993; Harding et al., 1998). Furthermore, research has demonstrated that the conversion of forests to pastures and/or residential areas may influence in-stream habitat and macroinvertebrate communities in several ways. Loss of terrestrial vegetation (Swank, Swift & Douglas, 1988) and an increased area of impervious surfaces (Changnon & Demissie, 1996) can influence evapotranspiration and infiltration and alter natural flow regimes (Poff et al., 1997). Many land-use practices increase sediment inputs to streams, altering substratum characteristics and channel morphology, often reducing macroinvertebrate diversity (Lenat & Crawford, 1994; Waters, 1995; Quinn et al., 1997). Removal of streamside vegetation and subsequent increased solar radiation reaching the stream channel can increase temperature (Burton & Likens, 1973; Rutherford et al., 1997; Quinn et al., 1997) and alter thermal regimes that are critical to the life history and ecology of macroinvertebrates (Vannote & Sweeney, 1980; Ward & Stanford, 1982; Quinn et al., 1994). In association with altered catchment hydrology and land use, inputs of inorganic nutrients from terrestrial sources (e.g. Omernik, 1977; Hunsaker & Levine, 1995; Johnson et al., 1997) may interact with increased light availability and stream temperature to enhance in-stream primary production (Webster et al., 1983), resulting in changes in the trophic structure of benthic communities (e.g. Gurtz & Wallace, 1984).

The relative magnitude of land-use effects may depend upon the spatial distribution of land use in catchments (Allan & Johnson, 1997). Catchment hydrology, as well as the availability of inorganic nutrients in streams, is often related to processes that occur across the terrestrial landscape (Omernik, 1977; Swank, Swift & Douglas, 1988; Hunsaker & Levine, 1995). Conversely, the availability of light and organic carbon in streams is often related to processes restricted to the scale of streamside vegetation (Gregory et al., 1991). Several studies have shown that in-stream physico-chemical and biotic features may be constrained by catchment properties operating at different spatial scales. For example, Richards, Johnson & Host (1996) identified catchment-scale characteristics, particularly basin geology and the distribution of arable agriculture, as features mediating channel morphology and stream hydrology. At the same time, land-cover in the riparian corridor had a strong influence on bank erosion and in-stream sediment-related variables. Further, they found that basin-wide features were the best predictors for macroinvertebrate assemblage structure. Conversely, Richards et al. (1997) showed that despite a close relationship between catchment-scale properties and channel structure and hydrology, macroinvertebrate species traits were correlated with local (reach-scale) features. In addition, spatial scales of catchment control may vary temporally. For example, Johnson et al. (1997) found that the strength of relationship between stream-water chemistry and catchment scale versus riparian scale land-cover patterns varied among season and with dissolved chemical constituents in Michigan streams.

The southern Appalachian region of the United States has been subject to a variety of land-use changes during the last century (Benfield, 1995). Between 1880 and 1920, almost all of the timber in the states of Virginia, West Virginia, North Carolina and Tennessee was harvested (Yarnell, 1998). This was followed by agricultural activity, which increased in importance through to the 1960s. Since 1970 human population in the southern Appalachians has increased by nearly 30% and agricultural activity has declined by 31%, leading to a large-scale conversion of agricultural land to either forests or residential areas (SAMAB, 1996). This history has resulted in a mosaic of different land-use patches across the landscape. Current land-cover patterns in many Southern Appalachian catchments are characterized by high gradient headwater areas that are difficult to develop and are often forested, and low gradient areas subject to active agriculture or conversion to urban land (Wear & Bolstad, 1998).

In this study, we investigated the effects of land-use on benthic macroinvertebrate assemblages in southern Appalachian headwater streams. Specifically, we address how differences in land use among catchments influences in-stream physico-chemical features and macroinvertebrate assemblage structure. In addition, we sought to determine how changes in in-stream features (both biotic and abiotic) relate to the spatial arrangement of land use within catchments. To do this, we quantified the relationship between land-cover and in-stream variables at multiple spatial scales. This approach allowed us to evaluate the ecological implications of land-use practices that occur streamside versus those that occur across the catchment.


Site description

The study was conducted in the Upper Roanoke River Basin (URRB) in south-western Virginia (Fig. 1) at the interface of the Appalachian Valley and Ridge and Blue Ridge physiographic provinces. The URRB is topographically and geologically diverse, and is characterized by Precambrian and Cambrian metamorphics and clastics at higher altitude and Cambrian and Ordovician carbonates lower in the valley (Waller, 1976). Nine catchments with second or third order streams were selected for study. Catchments ranged in altitude from 325 to 575 m above sea level, and varied from 278 to 1014 hectares. Six of nine catchments were characterized by some degree of human activity (e.g. agricultural or urban development); the remaining three had no such development. One 50-m stream reach was selected within each catchment as our study site. Benthic habitat at all streams consisted of particles ranging from silt to cobbles, although the relative proportion of these size classes varied among sites.

Figure 1.

 Upper Roanoke River Basin (URRB), showing North and South Fork sub-basins, and selected study sites. Site names correspond as follows: (1) Powers Branch, (2) Sugar Run, (3) Purgatory Tributary, (4) Martins Creek, (5) Little Back Creek, (6) Greenbriar Branch, (7) Mudlick Creek, (8) Barnhardt Creek, (9) Franklin Creek.

Land use quantification

A geographical information system (GIS) (Arcview 3.1; ESRI, Redland, CA, U.S.A.) was used to quantify land cover within each basin. Digital land-cover information for the Roanoke Valley was obtained from a preliminary land-cover map of Virginia, produced through the Virginia Gap Analysis Project (VAGAP) for 1992 (Morton, 1998). This statewide data set was generated from 14 Landsat thematic mapper scenes, and classified using both unsupervised and enhanced supervised methods (Morton, 1998). Pixels provided 30 × 30 m resolution and included information from seven land-use categories: deciduous forest, coniferous forest, mixed forest, shrub/scrubland, herbaceous (mostly agriculture), open water and disturbed (areas lacking vegetation). We further categorized the data into forest and non-forest. The forest category included all forest types, but was dominated by the deciduous class. The non-forest category included both agricultural and urban/suburban areas.

Morton (1998) used aerial videography to acquire reference data and assess the accuracy of the statewide land-cover data set, which was found to be 81%. This level of accuracy may not hold for riparian corridors, which are more difficult to classify, particularly in mountainous areas. Our accuracy assessment of riparian corridors, from 25 locations across five study catchments with the greatest land-cover heterogeneity, found approximately 75% of the streamside pixels were correctly classified. However, of the pixels classified as non-forest, 50% had thin strips (e.g. < 1 m) of riparian vegetation but otherwise lacked forested area. Incorrect classifications encountered in the assessment generally occurred when pixels classified as forest had small residences but were otherwise well forested.

Catchments were delineated using a watershed delineator (ESRI & Texas Natural Resource Conservation Commission, 1997) which uses neighbourhood functions with data from US Geologic Survey (USGS) digital elevation maps to quantify the surface area contributing to drainage through a given point. The outputs were converted to catchment polygons, which included the entire drainage area upstream of the sampling reach. Land-cover derived from all pixels within catchment polygons were used to characterize catchment scale land-use patterns. Within each catchment polygon, a 60-m riparian corridor polygon was created with 30 m of width extending laterally, beginning at the sampling reach and continuing for the entire length of the stream. The influence of riparian land-cover at specific distances upstream of the study reach was assessed by dividing riparian corridors longitudinally into sub-corridor polygons. We used five spatial scales for analysis: (1) the catchment scale, including all of the area within its watershed, (2) riparian corridor scale, extending over the entire length of the stream, (3) 2000 m riparian sub-corridors, (4) 1000 m riparian sub-corridors and (5) 200 m riparian sub-corridors (Fig. 2). Associated catchment, riparian corridor and sub-corridor polygons were overlaid onto the digital land-cover data for each study basin. Percentages for land-cover categories were quantified within all spatial polygons. Regression analysis was used to relate physico-chemical and biological variables to land cover at the catchment, riparian corridor and sub-corridor scale. Percent non-forest values were transformed (arcsine square root) for statistical analyses.

Figure 2.

 Examples of GIS polygons used to analyse land-cover patterns at five spatial scales, including the entire catchment, riparian corridor and three riparian sub-corridors extending 200, 1000 and 2000 m upstream from the sampling reach.

Physical and chemical characterizations

Physical and chemical measurements were made for each stream from January to December 1999. Triplicate water samples were collected monthly from each site, passed through glass fibre filters (Whatman GFF, Gelman Type AE, Maidstone, U.K.), and kept frozen until analysed. Samples were analysed for ammonium–nitrogen (NH4–N) using the phenate method (Soloranzo, 1969) and nitrate–nitrogen (NO3–N) by colormetric techniques following reduction by Cd (Wood, Armstrong & Richards, 1967) on a Technicon Auto-analyser (Bran & Luebbe, Buffalo Grove, IL, U.S.A.). Total inorganic nitrogen (TIN) was calculated as the sum of NH4–N and NO3–N. Ortho-phosphate (PO4–P) was analysed as soluble reactive phosphorus (SRP) using the molybdate colorimetric method (Murphy & Riley, 1962; Wetzel & Likens, 1991). Specific conductance was measured seasonally using a YSI Model 30/50- (YSI, Yellow Springs, OH, U.S.A.) conductivity meter. Alkalinity was determined once at each site by standard methods (APHA, 1998). Temperature was recorded monthly until continuous temperature data recorders (HOBO, Polasset, MA, U.S.A.) were installed in March 1999, after which it was recorded hourly through November 1999. Temperature variables used in analyses included the overall mean temperature (T, °C), and maximum summer temperature (Tmax, °C). Discharge was measured at least monthly at each site using velocity determined with an electronic flowmeter (FLOW-MATE, Marsh-McBirney, Frederick, MD, U.S.A.) and cross-sectional area (Gore, 1996). Mean substratum particle size (MPS) was determined using standard granulometry techniques (sensuWolman, 1954) on 100 randomly selected rocks at each site with a USGS gravelometer. We also quantified the percentage of all particles measured at each site small enough to be categorized as coarse gravel, sand and silt (% < 16 mm).

Benthic algal biomass

Epilithic biomass (g m–2) and chlorophyll a (mg m–2) were determined monthly at each site from the surfaces of five rocks collected along each 50 m reach. Upon collection, rocks were packed in ice, and returned to the laboratory for processing. Upper surfaces were scrubbed and the resultant slurry subsampled and collected on preweighed glass fibre filters (Whatman GFF, Gelman Type AE). Filters were incubated in a 90% buffered acetone solution for 24 h (sensuSteinman & Lamberti, 1996). Photosynthetic pigments (chlorophyll a and phaeophytin) were measured using spectrophotometry (at 664, 665 and 750 nm) on a Shimadzu UV-1601 spectrophotometer. An additional subsample was taken to determine epilithic ash-free dry mass (AFDM) by gravimetric techniques (drying at 50 °C for 48 h, followed by ashing at 550 °C for 1 h). Photosynthetic pigment concentration (mg m–2) and AFDM (g m–2) were normalized to surface area determined by covering scraped surfaces with aluminium foil of known mass per unit area.

Invertebrate assemblages

Benthic invertebrates were collected from all sites in April 1999. Five Hess samples (surface area, 0.08 m2 sample–1; mesh size, 250 mm) were taken along each 50 m reach. Samples were preserved in 80% ethanol, and individuals identified to genus following Merritt & Cummins (1996), Stewart & Stark (1993) and Wiggins (1996). Chironomid larvae from all samples within a site were combined, quantitatively subsampled, mounted and identified following Epler (1995) and Merritt & Cummins (1996).

To address changes in benthic community structure among sites, we quantified traditional measures of richness and evenness, and a compositional index commonly used to indicate environmental stress. Indices used to assess diversity included taxon number (S), and the Shannon–Weiner Diversity Index:

inline image

wherepi represents the proportion of individuals found in the ith taxon, with values summed across all taxa (S). We also used the associated evenness measure calculated from H′. This index assumes that maximum diversity occurs when all taxa are equally abundant (i.e. H′=H′max=ln S). Therefore, the ratio of H′/H′max represents a measure of evenness where an assemblage with an equal abundance of taxa would have a value of 1 (Pielou, 1969). As an additional measure of evenness we quantified the percentage of the total numbers accounted for by the five most dominant taxa at each site (% 5 Dominant) (Barbour et al., 1999). Finally, to assess compositional differences among sites, we quantified the taxonomic richness of commonly intolerant taxa (Ephemeroptera + Plecoptera + Trichoptera, EPT), widely used as an indicator of disturbance to stream communities (Lenat & Crawford, 1994; Wallace et al., 1996).

Rank–abundance curves (sensu;Hawkins, Murphy & Anderson, 1982; Marsh-Mathews & Mathews, 2000) were generated for invertebrate assemblages at each site. Curves consisted of lognormal proportional abundance (pi) for each taxon plotted against the corresponding taxonomic abundance ranking. The slope of this line (Δlog pi/n, where n=number of taxa) was used as an integrative measure of taxon richness and evenness. Diversity is maximized when the slope of the curve approaches zero. Slopes become increasingly negative when the proportional abundance (pi) of the more dominant taxa increases (i.e. evenness decreases), or when taxon richness decreases. We used the absolute value of the slope for all analyses; therefore, larger values represent steeper rank–abundance curves, and thus, less diverse assemblages.

Data analysis

Bivariate regression analysis was used to relate land cover variables (e.g. % non-forest) at each spatial scale to the in-stream physico-chemical variables and macroinvertebrate indices measured at all sites. In addition, macroinvertebrate indices were used as dependent variables in stepwise multiple regression analyses with in-stream physico-chemical parameters. Variables that failed to meet a 0.05 probability value were removed from multiple regression models. Regression analysis was used to test rank–abundance curves for linearity and the slope of the lognormal plots was compared among sites using methods described in Zar (1996) with a Bonferroni correction for multiple comparisons. All regression analyses were performed on SAS 7.0 (SAS Inc., Cary, NC, U.S.A.).


The magnitude and spatial arrangement of non-forested land varied considerably among drainage basins at the catchment, riparian corridor and sub-corridor scales (Table 1). In addition, the proportion of non-forested area within any given riparian corridor often varied greatly when compared with that of the associated catchment. For example, in four basins (Table 1, sites 1, 2, 3 and 6), the percentage non-forest in the entire riparian corridor was approximately 25% of that observed at the scale of the entire catchment; indeed sites 2 and 3 were nearly an order of magnitude lower. At the same time, the percentage non-forest within the riparian corridor increased 2.8 (site 4), 1.3 (site 5), 1.4 (site 8) fold at three other sites. Within riparian corridors, the distribution of non-forested area varied in proximity to the study reach (Table 1). In basins with low non-forest cover at both the catchment and total riparian corridor scales, non-forest cover in sub-corridors was also low. For example, at sites 1–3, six of nine riparian sub-corridors were entirely forested and a maximum of only 3% was non-forest (Table 1). For basins less forested at broader spatial scales, the percentage non-forest in sub-corridors often increased substantially. For the four basins with the least forest cover at the catchment scale,% non-forest in the sub-corridors averaged 58% and was as great as 93%. Finally, longitudinal changes in forest cover within riparian corridors varied greatly among basins. For instance, at site 9. 20% of land cover was non-forest at both the catchment and entire riparian corridor scales. However, if only the first kilometre of the riparian corridor was considered, 80% of the landscape was non-forest. Conversely, site 6 was classified as 34% non-forest at the catchment scale, but only 19 and 4% was non-forest at the entire riparian corridor and 1000 m sub-corridor scales, respectively.

Table 1.   Land-cover patterns for the nine study basins. Values are percent non-forested land within each scale category. Sub-corridor values are the percent of non-forested land within a polygon of 60 m width and progressively greater distances upstream from the sampling reach Thumbnail image of

Physical responses

Variation in several in-stream physical properties was related to catchment features and/or land-cover patterns at multiple spatial scales (Table 2). For example, stream discharge ranged from 15.3 L s–1 (site 3) to 55.4 L s–1 (site 5), varied temporally to a similar extent among streams with annual CVs ranging from 46 to 83% and was closely related to basin area (r2=0.85, n=9, P=0.004). Mean stream temperature varied by only 4.3 °C among sites, but maximum temperature varied by 6.7 °C, with the greatest mean temperature and highest maximum temperature both at site 7. Mean stream temperature decreased with increasing altitude (r2=0.78, n=9, P=0.002), and multiple regression indicated that the combination of altitude and the percentage non-forest at the entire riparian corridor scale explained 93% of the variation in temperature (n=9, P=0.02). Maximum stream temperature increased with the percentage of non-forested land at the entire riparian corridor scale (r2=0.79, n=9, P=0.001) and at the sub-corridor scales of 2000 and 1000 m (r2=0.71, n=9, P=0.005; r2=0.71, n=9, P=0.005, respectively). However, the strongest relationship between maximum stream temperature and the percentage non-forest was found at the 200 m riparian sub-corridor scale (r2=0.81, n=9, P=0.001). The mean size of coarse substratum ranged from 44.8 mm at site 8 to 72.4 mm at site 2, and decreased with the percentage of non-forested land at the catchment (r2=0.56, n=9, P=0.02), riparian corridor (r2=0.74, n=9, P=0.003) and 2000, 1000 and 200 m and sub-corridor (r2=0.80, n=9, P=0.001; r2= 0.61, n=9, P=0.013; r2=0.74, n=9, P=0.003, respectively) scales. The percentage of particles < 16 mm ranged from 17.1 at site 3 to 44.4 at site 8, and was only weakly related to the percentage non-forest within 200 m sub-corridor polygons (r2=0.52, n=9, P=0.03).

Table 2.   Physical, chemical and biological variables measured at all study streams. Data are mean ± SE derived from monthly values (n = 12), except specific conductance (n = 4), and altitude, catchment area, substratum size and alkalinity which were determined once for each site Thumbnail image of

Chemical responses

Chemical variables were also related to catchment features or land-cover patterns at the catchment scale. Alkalinity ranged from 16 mg L–1 at site 1 to 98 mg L–1 at site 9 and decreased with catchment altitude (r2=0.63, n=9, P=0.01). In addition, specific conductance (μs cm–1) varied among sites from 47.3 at site 1 to 261.8 at site 9, and was strongly related to alkalinity (r2=0.95, n=9, P=0.0001) and altitude (r2=0.67, n=9, P=0.007). Total inorganic N varied among sites (CV=80%), with minimum values of 0.105 mg L–1 (site 5) and maximum values of 0.932 mg L–1 (site 6), and increased with the percentage of non-forested land at the catchment scale (r2=0.59, n=9, P=0.02). Soluble reactive phosphorus did not differ greatly among sites (CV=38%) and did not correspond to any physical features of the catchment, or indeed land-cover at any spatial scale.

Biological responses

Based on field observation, algal assemblages were generally dominated by diatoms, although filamentous algae (e.g. Cladophora sp.) were found in abundance in streams draining catchments with extensive agricultural or urban development (e.g. sites 8 and 9). Mean annual epilithic chlorophyll a concentration ranged from a minimum of just 7.0 mg m–2 at site 3 to 121.1 mg m–2 at site 8. Chlorophyll a values were lowest for sites 1–3, where averages ranged from 7.02 to 27.9 mg m–2; concentrations were higher at the remaining sites with averages ranging from 54.8 mg m–2 (site 6) to 121.1 mg m–2 (site 8) (Table 2). Bivariate regression analysis did show a relationship between chlorophyll a concentration and mean stream temperature (r2=0.66, n=9, P=0.008). However, multiple regression analysis incorporated only maximum temperature into the model, explaining 73% of the variability in chlorophyll a, (n=9, P=0.003). Epilithic biomass ranged from 5.3 g m–2 at site 3 to 33.8 g m–2 at site 9, and was closely related to epilithic chlorophyll a concentration (r2=0.80, n=9, P=0.001).

Macroinvertebrate density varied among sites from 6974 m–2 at site 3 to 15 730 m–2 at site 9 (Table 3). Multiple regression analysis with density as the dependent variable incorporated only epilithic biomass into the model, accounting for 87% of the variability among sites (n=9, P=0.0003). A total of 77 macroinvertebrate taxa were identified from the nine sites. Taxon richness ranged from 24 at site 8 to 54 at site 3 with five EPT taxa at site 8 and 22 EPT taxa at sites 2 and 3 (Table 3). Results of the comparison of rank–abundance slopes indicated significant among site differences (P < 0.001, after Bonferroni correction), with three general groups emerging that represent high, medium and low diversity (Fig. 3). All rank–abundance curves fit linear models (r2 > 0.9, P > 0.001) and slope values ranged from –0.0470 at both sites 1 and 2 to –0.1100 at site 7. Stepwise multiple regression analyses between macroinvertebrate indices and in-stream physico-chemical variables generally incorporated a single independent variable into the models (Table 4). For example, the slope of rank–abundance curves increased with epilithic chlorophyll a concentration (r2=0.64, n=9, P=0.009). Total taxon richness (S) and EPT richness decreased with maximum stream temperature (r2=0.54, n=9, P=0.02, r2=0.75, n=9, P=0.003). Similarly, diversity (H′) and evenness (H′/Hmax) also decreased with maximum stream temperature (r2=0.72, n=9, P=0.004; r2=0.83, n=9, P=0.002, respectively). The percent 5 dominant taxa ranged from 55% at site 3 to 93% at site 8, and increased with increasing maximum stream temperature (r2=0.75, n=9, P=0.003).

Table 3.   Benthic macroinvertebrate indices for each study site. Density data are mean ± SE for five Hess samples from each 50 m study reach. Sites are presented in groups with other sites having statistically similar rank-abundance slopes Thumbnail image of
Figure 3.

 A. Rank–abundance plots for three macroinvertebrate assemblages representative of the three discrete rank–abundance diversity groups, ●=Site 7, ◆=Site 9, ○=Site 3. Symbols show raw data points and lines represent the modelled linear equations. All plots fit linear models with r2 > 0.91, P < 0.001. B. Absolute value of rank–abundance slopes for all study sites. Bars with shared lettering are not significantly different (P < 0.001, after Bonferroni correction).

Table 4.   Results of stepwise multiple regression analyses between macroinvertebrate indices and in-stream physico-chemical variables measured at each site (n = 9). Variables were removed from the model that did not meet a probability value of 0.05. R–A Slope = absolute value of rank–abundance slopes Thumbnail image of

The structure of macroinvertebrate assemblages corresponded to the variance in land-cover patterns among catchments (Table 5). Macroinvertebrate density was most closely related to the percentage of non-forest land at the 2000 m sub-corridor scale. Rank–abundance slopes increased with the percentage of non-forested land at the entire riparian corridor, and at 2000 and 1000 m sub-corridor scales, but were most strongly associated with land-cover patterns at the 200 m sub-corridor scale. In general, other measures of diversity, evenness and composition most closely corresponded to the percentage non-forest at the 200 m sub-corridor scale. Shannon’s evenness decreased with the percentage non-forest, only at the 200 m sub-corridor scale. Taxon richness, diversity, the percent five dominant taxa and EPT taxa richness were most strongly related to land cover within 200 m sub-corridors, but were also significantly related to land-cover patterns in other sub-corridors. In addition, EPT taxa richness was related to the percentage non-forest at catchment scale and was the only macroinvertebrate index that corresponded to land-cover patterns at all spatial scales evaluated.

Table 5.   Regression coefficients (r2) for significant bivariate regressions (n = 9, P < 0.05) between macroinvertebrate indices and the percentage non-forest at five spatial scales. R–A slope = absolute value of rank-abundance slopes. ns = Not significant Thumbnail image of


The observed variability in many of the chemical constituents studied in the URRB streams can be attributed to physical features (e.g. altitude) or land-cover patterns at the catchment scale. The relationships between alkalinity, specific conductance and altitude probably reflect shifts in catchment geology (Waller, 1976). In this part of the URRB, catchments at high altitude are dominated by clastic materials (both granitic and sedimentary) and have relatively low alkalinity and specific conductance when compared with low altitude streams that are influenced more by soluble parent lithology. The three catchments nearest the town of Salem, Virginia (sites 7, 8 and 9) include both clastic and carbonate materials, and subsequently had the highest alkalinity and specific conductance concentrations among sites. Variation in stream-water TIN was only related to the percentage non-forest at the catchment scale. This is consistent with many studies that have shown land-cover at the catchment scale to be a good predictor of in-stream nutrient concentration, particularly nitrate (Omernik, 1977; Close & Davies-Colley, 1990; Hunsaker & Levine, 1995; Johnson et al., 1997).

In-stream physical variables were also closely related to land-cover patterns, particularly at the riparian corridor and sub-corridor scales. Mean stream temperature, although related to altitude, was also influenced by land-cover patterns within riparian corridors. Moreover, maximum temperature was strongly related to the percentage non-forest at the scale of the 200 m riparian sub-corridor. High mean and maximum stream temperature may result from solar radiation reaching the stream channel (Burton & Likens, 1973; Beschta & Taylor, 1988; Rutherford et al., 1997), and/or from run-off heated by impervious surfaces in residential areas (Galli, 1991). The local influence of land-cover on stream thermal regimes (i.e. at the 200 m sub-corridor scale) supports the idea that temperature can change quickly (with respect to longitudinal distance) when streamside vegetation is removed in headwater catchments. Burton & Likens (1973) found summer stream-water temperature fluctuations of 4–5 °C, alternating between 50 m reaches where riparian vegetation had been experimentally removed or left intact in Hubbard Brook streams. Similarly, Storey & Cowley (1997) showed that high temperature in pasture streams of New Zealand return to ‘forested control’ levels after running through approximately 300 m of remnant riparian forests.

Substratum characteristics were also influenced by catchment land use. Mean substratum particle size was inversely related to the percentage non-forest at all spatial scales, but the strongest relationships were observed at the 2000 m sub-corridor scale. The percentage of particles <16 mm also corresponded to land-cover patterns, but only at the 200 m sub-corridor scale. A reduction in mean substratum particle size is associated with a greater frequency of fine particles, which frequently result from sedimentation (Waters, 1995). Sediment inputs to streams have been attributed to many types of land-use practices (e.g. agriculture, silviculture, road construction), particularly when these occur adjacent to the channel (Lenat et al., 1981; Waters, 1995; Richards et al., 1996). Several studies have also reported that intact streamside vegetation inhibits the delivery of sediments to streams. Peterjohn & Correll (1984) found that 19 m of riparian forests removed up to 90% of the particulate materials moving overland from agricultural areas. Similarly, Robinson, Ghaffarzadeh & Cruse (1996) showed that the initial 3 m of riparian forests removed more than 70% of sediment run-off to Iowa streams.

Among site variability in chlorophyll a and epilithic standing crop was also related to in-stream physico-chemical changes induced by land-cover patterns within riparian corridors. Chlorophyll a was closely related to thermal regimes in URRB streams, probably reflecting the influence of both stream temperature and light availability on benthic algae. Several studies have demonstrated a positive relationship between stream temperature and algal growth (Bothwell, 1988; Duncan & Blinn, 1989; Suzuki & Takahashi, 1995) and others have attributed greater algal biomass in forested regions to local increases in light availability (Lowe, Golladay & Webster, 1986; Hill & Harvey, 1990; Quinn et al., 1997). Interestingly, while many studies have linked algal abundance and nutrient availability (Fairchild, Lowe & Richardson, 1985; Grimm & Fisher, 1986; Biggs, 1995), variability in algal biomass in this study was not related to in-stream concentrations of TIN or SRP. However, neither TIN nor SRP were low enough at our study sites to be considered limiting (Grimm & Fisher, 1986; Lohman, Jones & Baysinger-Daniel, 1991). The interplay between thermal regime and nutrient availability was demonstrated by Bothwell (1988); in the absence of nutrient limitation, stream temperature explained 90% of the annual variability in benthic algal growth.

Changes in algal cover among sites had a strong influence on the benthos, as macroinvertebrate density was highest at sites with the highest algal biomass and biofilm standing stocks. This relationship is most likely driven by differences in chironomid abundance, which ranged in two orders of magnitude among sites. Many studies have indicated that invertebrate density and production may increase with greater algal biomass (e.g. Dudley, Cooper & Hemphill, 1986; Behmer & Hawkins, 1986; Richards & Minshall, 1988; Death & Winterbourn, 1995). Furthermore, studies in impacted stream ecosystems (mostly agricultural) have documented high macroinvertebrate density, often associated with numerical dominance by environmentally tolerant taxa (Lenat & Crawford, 1994; Harding & Winterbourn, 1995; Quinn et al., 1997). High density at some sites in the URRB may also be related to the presence of filamentous algae. Dudley, Hemphill & Cooper (1986) suggested that the abundance of Cladophora sp. might influence invertebrate density directly as a food source or by accumulating other food resources (e.g. detritus, smaller epiphytes), or indirectly by increasing habitat availability. In our study, highest algal standing crops, and macroinvertebrate density were measured at sites 8 and 9, where the epilithon was commonly dominated by filamentous green algae.

The use of rank–abundance curves allowed us statistically to assess differences in macroinvertebrate assemblage structure among sites. We identified three groups of streams that differed in the number and relative proportion of benthic taxa. Sites with low diversity and high dominance had the steepest rank–abundance slopes (sites 7 and 8) and were characterized by large numbers of only a few chironomid taxa, particularly members of the Orthocladius/Cricocladius group. The group with intermediate rank–abundance slopes (sites 4, 5 and 9) had either slightly higher overall diversity, in conjunction with high dominance by Orthocladius/Cricocladius chironomids (site 9), or had relatively diverse chironomid assemblages (sites 4 and 5). The group with rank–abundance slopes approaching zero had the most diverse macroinvertebrate assemblages, including a greater abundance of more sensitive taxa (e.g. EPT taxa), and frequently more than 10 chironomid taxa. As was the case with macroinvertebrate density, rank–abundance slopes increased among sites with algal biomass, probably reflecting the numerical dominance of Orthocladius/Cricocladius chironomids at sites with the highest algal cover (e.g. sites 7 and 8).

Other macroinvertebrate indices corresponded well to the three groups that emerged from the comparison of rank–abundance slopes, suggesting these curves adequately described assemblage structure in URRB streams. In addition, results from multiple regression analyses indicated that variation in index values among sites was most closely related to the thermal regimes, with diversity, evenness and EPT taxa richness being lowest at sites with the highest maximum stream temperature. Results from these analyses should be interpreted with some caution, however (see MacNally, 2000). Although each multiple regression incorporated only maximum stream temperature into the model, it is probable that other instream features (e.g. substratum size, flow regime, etc.) interact with the thermal regime to generate the habitat template (sensuSouthwood, 1977; Poff & Ward, 1990) upon which macroinvertebrate assemblages are structured.

The influence of temperature on the life history and ecology of aquatic insects is well documented (Vannote & Sweeney, 1980; Ward & Stanford, 1982; Sweeney & Vannote, 1986; Quinn et al., 1994). Ward & Stanford (1982) suggested that large diel fluctuations in stream temperature might lead to increased macroinvertebrate diversity by generating acceptable thermal conditions, and the potential for niche segregation among a wide range of benthic taxa. In the current study, this relationship between thermal regime and diversity may not hold if maximum stream temperatures become too high, especially for thermally sensitive taxa such as many stoneflies (Vannote & Sweeny, 1980; Sweeney, Vannote & Dodds, 1986; Quinn et al., 1994). Sweeney (1993) suggested that changes in temperature of 3–5 °C in Piedmont streams lacking riparian cover is sufficient to deleteriously alter larval recruitment and growth for many mayfly taxa. In addition to this, reduced macroinvertebrate diversity has been attributed to high maximum temperatures in both large regulated rivers (Fraley, 1979) and New Zealand pasture streams (Quinn & Hickey, 1990; Quinn et al., 1997; Storey & Cowley, 1997). In the URRB, both mean and maximum temperature in streams draining forested catchments (sites 1–3) were ∼3–6 °C lower than streams draining catchments with current land use activities (sites 4–9). Moreover, sites 1–3 also had the most diverse and even macroinvertebrate assemblages among streams in this study.

Macroinvertebrate diversity decreased among catchments with the percentage of non-forested land, although only when appropriate spatial scales were used in analyses. Several studies comparing macroinvertebrate assemblages among basins characterized by different land-use practices (e.g. native forest versus pasture) or parent geology have documented predictive ability at the catchment-scale (Lenat & Crawford, 1994; Harding & Winterbourn, 1995; Richards et al., 1996; Quinn et al., 1997). However, in our study, the relationship between land-cover and macro-invertebrate assemblage structure was strongest at the 200 m sub-corridor scale. This suggests that changes induced by local, near-stream development are sufficient to alter community structure, regardless of land-cover patterns found further upstream. For example, the catchment of site 5 was over 90% forested when classified at the catchment scale, yet the first 200 m of riparian corridor upstream of the study reach was classified as ∼25% non-forested as a result of recent (∼5–15 year) residential development. Assemblage structure at site 5 had low taxon richness (40 taxa), and EPT taxa richness (15 taxa), suggesting that local land-cover patterns had a strong influence on habitat quality. Similarly, studies investigating the importance of road construction (Lenat et al., 1981) and run-off from agricultural areas (Lemly, 1982) have documented the influence of local, streamside development on macroinvertebrate assemblages, including shifts in numeric dominance to intolerant taxa like chironomids. Sponseller & Benfield (2001) found that riparian land-cover within 1 km of the study sites was the strongest landscape predictor for leaf breakdown rate in URRB streams. Current results suggest that community structure may be more sensitive to local land-use disturbances than ecosystem processes that incorporate both biotic and abiotic components, which appear to be organized at broader spatial scales. This supports the idea that functional and taxonomic responses to disturbance do not necessarily correspond to one another (e.g. Wallace, Vogel & Cuffney, 1986).

The spatial relationship between land-cover and macroinvertebrate indices also suggests that local, upstream patches of riparian forest may have an ameliorative effect on macroinvertebrate assemblages. For example, the stream at site 6 drains a catchment subject to a mixture of land-use practices (both agriculture and residential), yet several hundred metres of riparian corridor remain forested upstream of the sampling site. The macroinvertebrate assemblage at site 6 was fairly diverse (51 taxa), and analysis of rank–abundance curves grouped this stream with those completely lacking urban or agricultural development. Storey & Cowley (1997) compared macroinvertebrate assemblages in upstream pasture reaches with downstream sites, after streams ran through approximately 600 m of native riparian forest. They found that the presence of riparian forests minimised the effects of conversion to pasture, as several physical parameters (e.g. temperature) as well as assemblage structure in downstream sites were more similar to forested control sites than upstream agricultural sites. The influence of riparian forest patches on in-stream habitat features is likely to be critical to the distribution of many macroinvertebrate taxa in catchments subject to mixed land-use disturbances (see also Sweeney, 1993).

Our inability to predict macroinvertebrate assemblage structure from land-cover at the catchment or entire riparian corridor scale may relate to our lack of historical land-cover data. Harding et al. (1998) found that land-cover patterns at the catchment scale from 1950, compared with land-cover patterns from 1970 and 1990, was the single strongest predictor of current biodiversity in North Carolina streams. In this study, we have assessed current land-cover patterns at multiple spatial scales, but have not addressed the temporal dynamic of land use. The degree to which current assemblage structure in these streams reflects land-use history is unknown. For example, the catchment at site 3 is without any current human activities; yet old barbed wire fencing across the stream in several sections, as well as the dominance of sycamore (Plantanus occidentalis L.) in the riparian forest suggests that the area was farmed in the recent past. Interestingly, the macroinvertebrate assemblage at site 3 had a much higher percentage of chironomids than the other two basins lacking current agricultural or urban development (sites 1 and 2). In addition, we are unable to date the various patches of residential development within and among catchments. Some of the study catchments (e.g. sites 8 and 9) are dominated by relatively old neighbourhoods (∼30–60 years), while others (e.g. sites 5 and 7) show signs of much more recent development (∼5–20 years). The time required for macroinvertebrate communities to recover from disturbances associated with residential or agricultural land use, the community-wide consequences of additional land-use disturbances during recovery from previous development (e.g. residential development in old pasturelands) and the relative stage of assemblage recovery at the time of sampling remain unknown. Investigating land-use effects at both the appropriate spatial and temporal scales will be critical for a full understanding and prediction of macroinvertebrate communities in developing catchments.


This work was supported by the NSF/EPA Waters and Watersheds Program and the National Science Foundation LTER Program at Coweeta Hydrologic Laboratory (DEB 96-32584). We also thank members of the Virginia Tech Stream Team for assistance in the field and laboratory. Matt McTammany provided assistance with GIS applications.