Near-Real-Time Global Probabilistic Flood Mapping via Local Thresholds and Cloud Computing

This study introduces a novel approach to creating probabilistic flood maps by employing local thresholding techniques with near-real-time Sentinel-1 satellite imagery and comprehensive global land cover data. The methodology's accuracy was evaluated across eleven distinct flood events spanning multiple continents. The process encompasses several key steps: (1) extracting Synthetic Aperture Radar (SAR) backscatter samples for various polarizations and land covers, including inundated areas; (2) fitting Gamma distributions to derive both the probability density function and cumulative distribution function for each land cover and polarization type; (3) utilizing the shape, location, and scale parameters from each Gamma distribution to calculate probability thresholds as percent points (percentiles, or the inverse of the CDF); and (4) applying these local thresholds to SAR imagery to generate detailed probabilistic flood maps. The flood events and the mapping tool are made accessible to the public via the Google Earth Engine platform.