Flood disaster management using Earth Observation technologies

Floods are among the most frequent and damaging disasters worldwide, affecting millions of people and causing high economic losses. Traditional ground-based monitoring systems often lack sufficient coverage, particularly in vulnerable regions, creating a need for more reliable tools. Satellite-based Earth Observation (EO) has emerged as a vital resource for flood detection, monitoring, and management. Satellites including Sentinel-1, Sentinel-2, Landsat, and commercial constellations like Planetscope, Skysat, COSMO-SkyMed provide valuable datasets to map inundation, assess damage, and guide emergency response. This thesis focuses on the Emilia-Romagna floods of May 2023, one of Italy’s most severe in recent decades, particularly in Spazzate-Sassatelli. Using Sentinel-2 (open-source) and SkySat (commercial) imagery, a modular Python-based algorithm was developed to process images based on K-means clustering machine learning method. Results showed that Sentinel-2 could generate regional maps in less than one minute using cloud-based processing such as Google colab, while SkySat provided finer-scale details within ~4 minutes on local hardware. Accuracy assessments confirmed Sentinel-2’s effectiveness at large scales, though SkySat offered superior precision in urban and narrow floodplain contexts. The Sentinel-2 algorithm offers cleaner and more precise flood detection, while SkySat identifies a wider area but with slightly more false positives. Overall, both deliver comparable performance. Overall, free open data is good for making fast maps of large areas, while very detailed commercial images are better for showing local damage clearly. In the future, flood monitoring should use more automation, machine learning, and real-time systems to respond faster and more effectively.