Model-based Estimation of Impervious Surface by Application of Support Vector Machines

Due to its various negative consequences impervious surface is increasingly recognized as a key issue in assessing the sustainability of land use – particularly in urban environments. In Germany, a regional or state-wide method for an accurate, fast and cost-effective quantification and assessment of impervious surfaces is still missing. Thus, we researched into the characterization of impervious surfaces based on an automated analysis of Landsat-7 data by means of a model-based estimation with Support Vector Machines (SVM). The approach provides a semi-automated method to model the impervious surface via Landsat data based on an automated training procedure. Additionally, the technique facilitates the supplementation of vector-data in order to enable the consideration of small-scale infrastructure, the aggregation of the gathered information to user-defined spatial units and the combination with socioeconomic data. The study area includes the entire region of the German federal state of Bavaria covering an area of 75.500 km. The analysis targets the degree of impervious surface for the total of residential, industrial and transport areas as well as the area of impervious surface per head. This information is referenced to various administrative and landscape-related units. The validation of the modeled degree of impervious surface based on the Landsat data shows a mean absolute error of 14 percent. The combination of the impervious surface raster and vector information on linear infrastructure such as roads and railways showed a difference of one percent from according reference information. The result of this study documents that the proposed approach is qualified for a highly automated and area-wide mapping of the degree of impervious surface. * T. Esch. University of Wuerzburg, Am Hubland, D-97074 Wuerzburg, Germany [email protected]