Flood Detection Integrating Spectral Indices (FLOODISI): A Novel Approach to Open-Water Mapping

Open-water mapping of flooded areas using multispectral remote sensing still presents significant challenges, particularly in urban environments, where high spectral mixing leads to substantial commission and omission errors in water classifications. This study proposes FLOODISI (Flood Detection Integrating Spectral Water Indices), a framework that integrates multiple spectral water indices through adaptive thresholding to generate an Integrated Water Map (IWM). The method was applied to Landsat-8/OLI imagery to map an extreme flood event in southern Brazil. The approach evaluates 12 spectral indices and iteratively adjusts threshold values to minimize false positives while preserving true positives. The results indicate that the Normalized Difference Flood Index (NDFI2), using adaptive thresholding, achieved the best individual performance, with an overall accuracy of 91.8%, whereas the IWM increased this value to 93.5%, substantially reducing omission errors and improving open-water detection in urban areas. In comparison, the Random Forest classification achieved an overall accuracy of 95.0%, but exhibited similar precision and specificity, with a slight increase in commission errors and a modest reduction in omission errors relative to the IWM. In general, the integration of multiple spectral indices with adaptive thresholds through FLOODISI improved the robustness of open-water detection by reducing the dependence on individual spectral indices and providing a scalable, reproducible, and computationally efficient solution for rapid open-water mapping of flooded areas.