Evaluation of the applicability of a probabilistic terrain descriptor to improve the thematic accuracy of DLR’s TerraSAR-X based Flood Service (TFS)

Abstract

The study evaluated whether improvements to the thematic accuracy of automatic flood classifications based on synthetic aperture (SAR) backscatter analysis alone could be observed with the inclusion of ancillary data. In particular, prior work that assessed the utility of the height above nearest drainage (HAND) index as a post-classification mask was extended upon. Evaluations were conducted with two sets of freely available, medium-resolution GDEM products, for a range of scales within three open source geospatial analysis environments. The investigation was based on extensive literature review, correspondences with researchers, and preliminary trials in the identified test environments, with particular focus on the TerraView/ TerraHidro. The findings demonstrated that (1) HAND indices could be pre-calculated at local and continental scales; (2) comparable HAND indices were calculated with HydroSHEDS and SRTM Plus data, but the availability of the HydroSHEDS CON product significantly reduce computation times at the pre-processing stage; and (3) variable improvements to thematic accuracy were observed with the application of HAND indices at fixed threshold values to automatically generated flood classification layers. The evaluations were conducted over Europe (1), Germany (1, 2), and subsets of flood scenes that occurred in Germany, Albania, and Serbia (3). Results were assessed qualitatively and quantitatively with image statistics and the computation of correlation matrices. Workflows in Python were prepared for the HAND calculation within two of the test environments to support automation and further testing.