Flood mapping using cloud computing platform and machine learning -

Floods remain one of nature's most destructive forces, causing extensive infrastructure damage and loss of human lives globally. The lack of advanced flood risk modelling and real-time monitoring systems hampers the development of effective early warning mechanisms, leaving resource-constrained regions, particularly in Southern Africa, ill-prepared for flood mitigation and community protection. This study addresses the critical need for improved flood risk assessment by leveraging Synthetic Aperture Radar (SAR) data from Sentinel-1 integrated within the Google Earth Engine (GEE) cloud computing platform. The research methodology involved comprehensive mapping of flood-prone zones through integration of multiple data sources including satellite imagery, digital elevation models, land cover maps, and historical flood records to delineate areas affected by flooding during February 2023 in Mpumalanga province, South Africa. Results revealed that flood inundation extent was highest in Ehlanzeni district, followed by Gert Sibande district, while Nkangala district demonstrated comparatively lower flooding likelihood. The increased vulnerability of Ehlanzeni and Gert Sibande districts was primarily attributed to their low relief topography (0-500 m.a.s.l.), contrasting with less flood-prone districts characterized by higher elevation (>500 m.a.s.l.). Rainfall analysis confirmed that areas identified as flood-prone through SAR-derived inundation mapping received relatively high precipitation, establishing precipitation as the primary flooding driver. Land use/land cover analysis revealed significant shifts between grasslands, bare land, forest, and built-up areas, particularly in flood-prone districts. The findings underscore the effectiveness of GEE cloud computing platforms coupled with multi-source Earth Observation data for accurate provincial-level flood risk determination. This research provides a robust, cost-effective methodology particularly beneficial for resource-constrained regions where flood monitoring capabilities are limited. The results are critical for developing targeted flood mitigation interventions and establishing improved early warning systems.