I collected fish samples and habitat data at 43 sites throughout the upper Roanoke River watershed, Virginia. Sites were separated into three watershed areas size classes: 10 - 15, 20 - 30, and 70 - 80 km². I correlated physical in-stream conditions with proportions of forest, disturbed, and herbaceous/agricultural land at various watershed-scales to determine factors affecting stream habitat. I grouped fishes into metrics commonly used in indexes of biotic integrity and created a multimetric index called the mean metric score to represent fish communities at sites. Fish variables and metric values were compared with stream habitat and watershed variables to determine primary influences on fish communities. I correlated land use at 24 spatial scales, which differed by buffer width and stream network area, with mean metric scores to determine zones of greatest influence on fish communities.

In-stream habitat conditions and amounts of forest, herbaceous/agricultural, and disturbed land varied greatly among sites. Habitat varied due to natural differences among sites, such as elevation and watershed area, and due to land use. Disturbed land use was greatest at lower elevations while forests were more abundant at higher elevations. Substrate size distribution was highly correlated with all three land use types at several spatial scales. Correlations between land use within various buffers and median particle size became stronger as larger proportions of watersheds were included in analysis.

Fish species richness increased from small to large sites by species addition. Species collected at small sites were also collected at large sites, but several species collected at large sites were absent elsewhere. For example, orangefin madtoms and bigeye jumprocks were only collected at three large sites. Fish distribution was a result of several factors such as watershed area, elevation, proportions of pools and of riffles, particle size, and land use within buffers and entire watersheds.

Sites with high mean metric scores were primarily limited to tributaries of the North and South Forks of the Roanoke River. Most sites with low mean metric scores were located near the cities of Roanoke and Salem. Forest and disturbed land use were highly correlated with mean metric scores. Elevation was also highly correlated with mean metric scores but herbaceous/agricultural land use was not. Correlations between percent forest within 24 buffers and mean metric scores were highest for small stream network areas and declined as more land farther from sites was included for analysis. Correlations between disturbed land use and mean metric scores were strong regardless of the area considered. Mean metric scores declined precipitously as disturbed land use within watersheds and buffers increased from 0 to 10 %, but reached a plateau at 10 to 20 % after which increases in disturbed land use did not result in lower mean metric scores.

My results suggest that species addition and ecological shifts from more generalized to more specialized species occur with increased stream size. Forested buffers are important for maintaining ecological integrity, and buffers along sites with adequate integrity should be candidates for riparian restoration. Future development should be concentrated in watersheds that are already developed and reforestation of riparian areas in developed watersheds may reduce the impacts of watershed-level disturbance.