A Comparison of Approaches to Impervious Surface Characterization

Impervious surface estimates derived from different modeling approaches and data types were compared to highly accurate and precise planimetric calibration and validation data. A linear regression model using population density and percent land cover class per unit area produced the most accurate results, with an overall RMSE of 3.2% for the 82 census tracts studied. Two methods using just land cover-based impervious coefficients yielded results with an RMSE of 4.8% and 5.4%, while the subpixel estimates had an RMSE of 5.8% and 7.5%.