InSAR–Based Flood Detection of the 2024 UAE Rainfall Events Using Principal Component Analysis

Extreme rainfall events are becoming more frequent and intense due to climate change. Accurate flood detection is therefore essential to prevent or reduce future flood risks. In the United Arab Emirates (UAE), three consecutive rainfall events in February, March, and April 2024 set unprecedented records, surpassing all observations since rainfall measurements began in 1930. The April 2024 event led to one of the most severe flood episodes in the country’s history, with flash floods reported across most wadi systems. In arid regions, it is often difficult to detect flooded areas using traditional methods that rely solely on optical or SAR images acquired during or after flood events, primarily due to decorrelation effects. To overcome this problem, this study proposes a new approach that generates maps from Interferometric Synthetic Aperture Radar (InSAR) using coherence change detection (CCD) integrated with Principal Component Analysis (PCA) to reduce the impact of decorrelation. This approach evaluates InSAR-CCD using Principal Component Analysis (PCA) for multitemporal Sentinel-1 SLC data to map flood inundation in the UAE. Three coherence layers for pre-flood, peak-flood, and post-flood phases were computed and transformed through PCA to isolate dominant variance patterns linked to inundation. The Feature Preserve Smoothing filter was applied to CCD-PCA to reduce noise and ensure consistent resolution. The method was compared with the Change Difference Threshold (CDT). Results showed that filtered CCD obtained from PCA produced continuous and topographically consistent flood extents in urban plains, wadis, and salt-flat areas (sabkha). The observed coherence loss captured not only standing water but also saturated soil, erosion, and sediment transport. Thus, optical imagery was used to compare and cross-validate the CCD-PCA and CDT by choosing random points on the map to ensure they represented water bodies rather than sediment transport or soil moisture. The filtered CCD derived from PCA showed an overall accuracy (OA) of 0.84 and a Kappa (κ) value of 0.71, while CDT showed an OA of 0.65 and a κ of 0.20. The filtered CCD-PCA product showed perfect sensitivity, and no flooded pixels were missing. The results highlighted the sensitivity and accuracy of flood detection in arid environments using InSAR, which has great potential for flood detection and future mitigation strategies in arid regions.