Reducing False Alarms in Urban Flood Detection: An Enhanced NDWI (ENDWI) with Hybrid Max Fusion on Sentinel-2 Data

Abstract. Although optical satellite-derived water indices have significantly advanced urban flood detection, accurately distinguishing flooded from non-flooded pixels while minimizing false positives caused by spectral confusion in built-up areas remains a considerable challenge. This study proposes and evaluates the Enhanced Normalized Difference Water Index (ENDWI) in comparison with seven established water indices to reduce false alarms in complex urban environments. The approach was applied to a flash flood event in Al-Lith Governorate, a coastal urban area along the Red Sea in Saudi Arabia, selected as the case study because of its recurrent vulnerability to intense rainfall and rapid-onset flooding. Sentinel-2 imagery acquired two days after the event served as the core methodology for this study. Validation was performed using WorldView-4 high-resolution imagery obtained within two days of the event, based on 1,262 ground-truth points (559 flooded and 703 non-flooded) generated within polygons to ensure consistency with the Sentinel-2 spatial resolution. Analysis of the raw indices revealed that the Automated Water Extraction Index for shadows (AWEIsh_raw) achieved the highest area under the receiver operating characteristic (ROC) curve (AUC = 0.836), followed by the Normalized Difference Water Index (NDWI_raw) (0.813) and ENDWI_raw (0.784), positioning ENDWI among the top three performers. Following Otsu thresholding, ENDWI_otsu yielded the highest overall accuracy (79.41 %) and the lowest false alarm rate (10.95 %). A novel hybrid maximum fusion of ENDWI_raw and AWEIsh_raw further enhanced results, attaining an overall accuracy of 82.65 %, producer’s accuracy of 94.50 %, F1-score of 76.73 %, and Kappa coefficient of 0.637 after thresholding, with only 21 false positives (false alarm rate = 2.99 %). Overall, ENDWI exhibited robust and consistent performance across individual applications, post-thresholding, and hybrid fusion with AWEIsh, establishing it as a reliable and effective tool for accurate urban flood mapping.