| Geographic Information Systems (GIS) continue to be used more frequently and for a broader variety of applications. Careful consideration of the characteristics of underlying datasets that are incorporated in GIS models, particularly with respect to their accuracy for specific applications, is increasingly important. In this study, we evaluated the accuracy of two land cover datasets, the National Land Cover Dataset (2006) and the Gap Analysis Program (GAP) dataset at scales typical for Midwestern forest land ownership. We also evaluated the applicability of the Soil and Water Assessment Tool (SWAT 2005) for headwater streams in forested areas in central Iowa, including streams in urban, grazed, and preserved forests. For the landcover datasets, overall accuracy for Level I classification ranged from 59% for NLCD 2006 to 71% for the GAP dataset. Accuracy was relatively high for row crops (83% for both NLCD 2006 and for GAP) and developed areas (70% for NLCD 2006; 100% for GAP). Neither dataset generated optimal results for overall classification. Overall, the GAP dataset produced fewer errors for the areas we studied. For our evaluation of the SWAT 2005 model, we used the R2 coefficient and Nash-Sutcliffe Efficiency (NSE) statistic to characterize model performance using a multi-site approach for the set of nine streams. For calibration of discharge, R2 values ranged from 0.45 to 0.85, and NSE ranged from 0.41 to 0.84. Values of these statistics were lower for validation (R 2 of 0.07 to 0.72, NSE from -3.63 to 0.13). Model performance was variable for total suspended solids (calibration R2 from 0.01 to 0.80, and NSE from -0.55 to -0.04; validation R2 from 0.004 to 0.90, and NSE from -1.45 to 0.27). Overall, the SWAT model showed potential for prediction of discharge from small streams in forested areas, however, it did not perform as well for prediction of suspended solid concentration under our study conditions. |
