Effects of Incidence Angle Correction Methods on Flood Detection in Sentinel-1 GRD data

Abstract

The use of Sentinel-1 scenes for water and flood mapping is a common procedure. In this work it is investigated how Incidence Angle (IA) correction methods influence the performance of a stat-ofthe-art rule-based water mapping processor. The three correction methods Cosine Correction (CC),Empirical Range Fit (ERF) correction and Histogram (HIST) correction are implemented and tested. 12 globally distributed scenes are processed with and without IA correction, and the results quantified using reference data created from Sentinel-2 scenes. Each correction method is tested with different settings (varying bin counts and polynomial degrees, different R-value selection methods) and the best setting for each correction method is presented. Three metrics IoU, recall and precision are used for quantitative comparison, while all results are also visual inspected. The HIST correction underperformed compared to the CC and ERF correction, which both yield comparable results. The variation of bin counts had a considerable effect on the HIST correction, but did not have a noteworthy effect on the CC or ERF correction. The CC led to the best results, when using the minDiff R-value selection method combined with a bin count of 17. The overall best results where produced when using the ERF correction method with a 3° polynomial and a bin count of 34. Over all the presented IA correction methods improve the classification results, when used on scenes with enough water surfaces.