NDFI-SC: a robust near-real-time flood-mapping workflow using an SDWI-derived stable reference

Flood mapping requires workflows that are automated and robust under cloud-prone conditions. Recent reference-guided adaptive thresholding studies have improved threshold selection in change-detection-based flood mapping, but they still depend on usable optical reference data and may become constrained when cloud-free observations are unavailable. To address this limitation, this study develops NDFI-SC, a robust flood-mapping workflow implemented on Google Earth Engine that combines normalised difference flood index (NDFI) change detection with a Sentinel-1 Dual-Polarisation Water Index (SDWI)-derived stable reference and consistency-driven threshold calibration. Unlike existing optical-reference-guided approaches, NDFI-SC derives the stable reference directly from Sentinel-1, thereby reducing the dependence of threshold calibration on cloud-free optical observations. Near-date Dynamic World data are incorporated for inundated land-cover reporting. It is evaluated across three complementary scenarios: cross-method comparison in China, applicability under optically constrained conditions in Vietnam, and land-cover inundation reporting in Australia. Results show that NDFI-SC suppresses false alarms, reaches a critical success index (CSI) of 0.809 and a false alarm ratio (FAR) of 0.141 in China, shows good agreement with Copernicus Global Flood Monitoring in Vietnam, and supports near-real-time (NRT) reporting of land-cover types in Australia. These results indicate that NDFI-SC provides an NRT framework for flood delineation and impact assessment.