Enhancing floodwater depth mapping in tropical regions using multi-temporal Sentinel-1 SAR, supervised classification, and multi-DEM sensitivity analysis

Abstract Flooding in tropical Southeast Asia often occurs under persistent cloud cover, which limits optical imagery and increases reliance on radar-based monitoring. This study presents a framework for estimating floodwater depth during the October 2020 floods in Quang Tri Province, Vietnam, using multi-temporal Sentinel-1 synthetic aperture radar (SAR) data and supervised classification. Three classifiers—random forest (RFC), maximum likelihood, and minimum distance—were trained on balanced water and non-water samples. RFC delivered the most consistent performance, with testing accuracies above 99% and RMSE values below 0.021. Vertical transmit–Horizontal receive backscatter was the dominant feature, and omission and commission errors remained minimal across dates. Floodwater depth was estimated using the Floodwater Depth Estimation Tool (FwDET) with four Digital Elevation Models (DEMs): a locally corrected 30 m DEM and three global models (SRTM (Shuttle Radar Topography Mission), NASADEM, MERIT (Multi-Error Removed Improved Terrain)). The depth maps are highly sensitive to DEM selection. The reference DEM produced shallow, coherent inundation patterns, whereas SRTM and NASADEM yielded larger, fragmented deepwater zones due to positive elevation bias and stepped surfaces. MERIT reduced these artefacts but lacked fine-scale detail. A slope-stratified analysis confirms that depth differences are greatest in low-slope terrain. Median depth biases exceed 0.25–0.35 m for slopes below 0.1° and fall below 0.1 m in steeper areas. The strong negative slope–depth correlation, consistent across two flood peaks, highlights the influence of DEM structure and floodplain morphology. The findings show that DEM quality is the dominant source of uncertainty in remote-sensing-based depth estimation in tropical lowlands. Integrating reliable SAR-based water classification with terrain-dependent diagnostics provides a clearer basis for interpreting FwDET results. These insights support improved flood hazard assessment in Quang Tri and are transferable to other flood-prone regions reliant on global DEMs.