Synthetic Aperture Radar (SAR) is the only spaceborne remote sensing modality capable of reliable flood inundation mapping under cloud cover — the atmospheric condition that invariably accompanies large flood events and renders optical sensors (Sentinel-2, Landsat, PlanetScope) operationally useless during the critical period when emergency managers most need spatial flood extent data. Sentinel-1's C-band SAR constellation, operated by ESA, provides global coverage at 6-day repeat cycle (3-day at mid-latitudes with both Sentinel-1A and 1B operational) with Ground Range Detected (GRD) products available as free open data within 1–3 hours of acquisition.
The primary SAR flood mapping methodology is change detection: comparison of a flood acquisition against a pre-flood reference image acquired under comparable geometric and seasonal conditions. Open water surfaces produce specular reflection that returns minimal backscatter to the SAR sensor (appearing dark in GRD imagery), while vegetated land surfaces produce diffuse volume scattering (appearing bright). A pixel-wise thresholding approach applied to the backscatter difference image (flood minus pre-flood reference) classifies pixels with backscatter decrease exceeding a threshold as inundated.
Threshold selection is the critical methodological decision in SAR flood mapping, with significant impact on flood extent accuracy. This study evaluated four threshold selection approaches across 23 flood events from the Sentinel-1 archive (2014–2024): (1) fixed threshold (–3 dB backscatter decrease); (2) Otsu's method applied to the difference image histogram; (3) Kittler-Illingworth minimum error thresholding; (4) Expectation-Maximization Gaussian Mixture Model fitting. Reference flood extents were derived from post-event aerial photography and high-resolution optical imagery.
Kittler-Illingworth minimum error thresholding achieved the best overall performance (mean F1 score 0.84 across all 23 events) and demonstrated the most consistent performance across diverse flood types (fluvial, pluvial, coastal surge). The fixed –3 dB threshold produced acceptable results (mean F1 0.79) for fluvial flooding but degraded significantly for pluvial urban flooding (mean F1 0.61) where street-level inundation produces backscatter signatures more complex than the simple specular-to-diffuse contrast of open rural floodplains.
For operational flood emergency response, the recommended protocol combines: Sentinel-1 SAR flood mapping (KI thresholding) as the primary spatial extent product; uncertainty envelope from the hydraulic model's ensemble forecast as a spatial prior for probability mapping; and VHR optical imagery (Maxar WorldView, Planet Skysat) acquired post-storm for validation and damage assessment. The Prime Logic Water & Flood Intelligence Platform automates the complete Sentinel-1 GRD ingestion, preprocessing, change detection, and product generation workflow with a target processing time of under 20 minutes from SAR acquisition to web-mapped flood extent delivery.