Classification of stream biological condition within the Chesapeake Bay watershed

Abstract

Human land use activities at the landscape scale are increasingly the largest threat to the biological condition of watershed and stream ecosystems. The Chesapeake Bay watershed (CBW), a particularly valuable watershed within the United States, has undergone considerable land use change over the past 400 years and faces many restoration challenges. Using fish indicators of biological integrity (IBIs), and data for land use, land cover, and environmental attributes, five empirical models (CART model, Random Forest, Conditional Tree, Conditional Forest, and ordinal logistic regression) were used to predict the biological condition of 1st-3rd order streams within the CBW. After the models were evaluated using resubstitution and 10-fold cross validation, the highest performing model was identified (Random Forest) and extrapolated to 71,182 stream sites within the CBW using geographic information software (GIS). Of these sites, 49% (35,006 sites) were classified as having “Good” biological condition, 24% (16,826 sites) as having “Fair” biological condition, and 27% (19,350 sites) as having “Poor” biological condition. The variable importance plot generated by the Random Forest (RF) model showed that watershed area (upslope of sampling location, km2) was the most important variable, followed by percentage of impervious surface cover, and percentage of pasture cover. Additionally, the Random Forest’s partial dependence plots showed the marginal effect of each variable on the class probability. As watershed area (km2) increases, there is a higher probability of a “fair” or “good” classification of stream biological condition; with a threshold watershed area of approximately 20 – 25 km2. Also, as the percentage of impervious surface cover increases, there is a greater probability of a poor classification of stream condition (threshold of ~5% impervious surface cover). The results of this study may help environmental and land use managers understand the effects of human land use and make more effective land use decisions to address watershed impairment within the CBW.