| Runoff from urban areas is a leading source of nonpoint source pollution in estuaries, lakes, and streams. The extent and type of impervious land cover are considered to be critical factors in evaluating runoff amounts and the potential for environmental damage. Land cover information for watershed modeling is frequently derived using remote sensing techniques, and improvements in image classification are expected to enhance the reliability of runoff models. In order to understand potential pollutant loads there is a need to characterize impervious areas based on land use. However, distinguishing between impervious features such as roofs and roads using only spectral information is often challenging due to the similarity in construction materials. Since spectral information alone is often lacking, spatial complexity measured using fractal dimension was analyzed to determine its utility in performing detailed classification. Fractal dimension describes the complexity of curves and surfaces in non-integer dimensions. Statistical analysis demonstrated that fractal dimension varies between roofs, roads, and driveways. Analysis also observed the impact of scale by determining statistical differences in fractal dimension, based on the size of the window considered and the ground sampled distance of the pixels under consideration. The statistical differences in fractal dimension translated to minor improvements in classification accuracy when separating roofs, roads, and driveways. |
