Large area impervious surface mapping on the basis of optical and SAR data

Abstract

ABSTRACT: The aim of this paper is to present an approach based on both optical and SAR data, in order to quantify urban impervious surface (IS) as a continuous variable on regional scales independent of other sources of geoinformation. On the basis of single-polarized TerraSAR-X data settlement areas were identified by a semi-automated, object-oriented classification approach. The settlement footprints then served as the reference area for the impervious surface estimation, which is based on a Support Vector Regression model. The method developed was applied to SPOT 5 data from 2005 and 2009, covering almost the whole area of Can Tho province in the Mekong Delta, Vietnam. In addition, a change detection analysis was carried out and average IS statistics on commune level were computed. Settlement areas were correctly identified with overall accuracies between 89% and 94%, while the comparison of modelled to the training IS values gave an absolute mean error below 15%. The results demonstrate that the approach is suitable for large area impervious surface mapping thereby focusing exclusively on settlement areas.