Stream conditions have widely deteriorated due to anthropogenic disturbance, yet large-scale predictive-based models addressing this issue are limited.  We developed a watershed classification system for the Chesapeake Bay basin that uses improved data on watershed geography to predict stream impairment.  Stream conditions were quantified using hydrologic metrics, water quality metrics, and biological indicators.  We analyzed data from Smithsonian and USGS monitored watersheds to select hydrological and water quality metrics.  Biological condition was quantified with indices for benthic macroinvertebrates and fish reported by federal (US EPA) and state (Maryland, Virginia) stream assessment surveys.  We applied multiple linear regression, ordination and decision tree models to relate stream metrics to independent variables describing natural watershed attributes and anthropogenic influences.  The independent variables were summarized from digital maps of land use, land cover, and other geographic factors.  We also included variables derived from county budgets of net anthropogenic phosphorus inputs (NAPI) and net anthropogenic nitrogen inputs (NANI).  NAPI and NANI provide indices of nutrient pollution potential.  The best models derived from these analyses were applied to predict stream conditions in all first- to third-order streams of the Chesapeake basin.

Results/Conclusions

There were distinct hydrologic responses to the presence of impervious surface or cropland in a watershed, but the responses varied with physiography and climate. Average annual net phosphorus input (NAPI) was 4.52 kg P per hectare.  Around 15% of NAPI and 34% of NANI to smaller watersheds were discharged in streams, while 10% of NAPI to major river basins was discharged to Chesapeake Bay.  The strength of multiple regressions against watershed predictors (e.g., forest cover, developed land, and net nutrient inputs) varied among physiographic provinces, and the models explained 28 to 93% of the variation among watersheds in stream phosphorus discharge and 76 to 97% of the variation in nitrogen discharge.  Regression analysis of Maryland biological data identified impervious surface and tree cover as the principal predictors of stream condition (explaining 65% of variability in benthic integrity and 62% of variability in sensitive taxa).  In similar analyses of assessment data from the entire Chesapeake basin, elevation and impervious surface percentage were the strongest indicators of condition.  The basin-wide model classified 39% of all first to third order streams in the Chesapeake drainage in good condition, 16% in fair, and 46% in poor.  Our study demonstrates strong effects of land use on stream conditions; but the effects vary among physiographic province.