Predictability of surface water pollution loading in Pennsylvania using watershed-based landscape measurements

We formally evaluated the relationship between landscape characteristics an d surface water quality in the state of Pennsylvania (USA) by regressing tw o different types of pollutant responses on landscape variables that were m easured for whole watersheds. One response was the monthly exported mass of nitrogen estimated from field measurements, while the other response was a GIS-modeled pollution potential index. Regression models were built by the stepwise selection protocol, choosing an optimal set of landscape predicto rs. After factoring out the effect of physiography, the dominant predictors were the proportion of "annual herbaceous" land and "total herbaceous" lan d for the nitrogen loading and pollution potential index, respectively. The strength of these single predictors is encouraging because the marginal la nd cover proportions are the simplest landscape measurements to obtain once a land cover map is in hand; however, the optimal set of predictors also i ncluded several measurements of spatial pattern. Thus, for watersheds at th is general hierarchical scale, gross landscape pattern may be an important influence on instream pollution loading. Overall, there is strong evidence that using landscape measurements alone, obtained solely from remotely sens ed data, can explain most of the water quality variability (R-2 approximate to approx. 0.75) within these watersheds.