Mapping Global Floods with 10 Years of Satellite Radar Data

Abstract

Floods cause extensive global damage annually, making effective monitoring essential. While satellite observations have proven invaluable for flood detection and tracking, comprehensive global flood datasets spanning extended time periods remain scarce. In this study, we introduce a novel deep learning flood detection model that leverages the cloud-penetrating capabilities of Sentinel-1 Synthetic Aperture Radar (SAR) satellite imagery, enabling consistent flood extent mapping in through cloud cover and in both day and night conditions. By applying this model to 10 years of SAR data, we create a unique, longitudinal global flood extent dataset with predictions unaffected by cloud coverage, offering comprehensive and consistent insights into historically flood-prone areas over the past decade. We use our model predictions to identify historically flood-prone areas in Ethiopia and demonstrate real-time disaster response capabilities during the May 2024 floods in Kenya. Additionally, our longitudinal analysis reveals potential increasing trends in global flood extent over time, although further validation is required to explore links to climate change. To maximize impact, we provide public access to both our model predictions and a code repository, empowering researchers and practitioners worldwide to advance flood monitoring and enhance disaster response strategies.

Figures (8)

• Figure 1: Global Flood Map | Aggregated global flood extent map as detected by our deep learning model applied to 10 years of Sentinel-1 SAR data (October 2014 - Sep 2024). Blue areas indicate locations where flooding was detected at least once during this period, shown at 250-meter resolution. Darker gray areas represent the exclusion mask, indicating regions where flood detection may be unreliable due to urban development, steep terrain, or arid conditions. Areas without color showed no flooding during the observation period. This map highlights historically flood-prone regions identified by cloud-penetrating SAR data.
• Figure 2: Comparison of Flood Extent Maps for Ethiopia | Comparison of flood extent mappings over Ethiopia from multiple satellite sources. Blue areas show SAR flood detections from our deep-learning model (2014-2024), while orange areas represent historical flood extent from MODIS and Landsat optical and near-infrared imagery (1984-2021). Green areas indicate agreement between SAR and optical datasets. The map reveals both consistencies and complementarity between detection methods, with notable flood-prone areas identified along the Shabelle, Ganale, and Awash rivers. Darker gray areas indicate regions where SAR flood detection may be unreliable due to terrain or land cover characteristics.
• Figure 3: Detailed Flood Detection Comparison in Key Ethiopian Regions | Comparison of flood extent mappings in two flood-prone regions of Ethiopia: (a) Semera in the Awash River Basin and (b) Dolo Ado along the Ganale River. Blue areas show Sentinel-1 SAR flood detections from our deep-learning model (2014-2024), orange areas represent MODIS/Landsat flood extent (1984-2021), and green indicates agreement between datasets. Both regions demonstrate increased flood detection capabilities from our method, with significant amounts of flooding detected only by our model (blue). Darker gray areas indicate regions where flood detection may be unreliable.
• Figure 4: Semera Flood Map | (a) Overlay of cropland and flood extent maps near Semera, Ethiopia. Analysis reveals that approximately 19% of cropland near Semera is within historically flood-affected zones according to our flood map. (b) Detailed view of a specific area in the northwest part of the overview, illustrating the overlap between cropland (yellow) and flood zones (blue). The high-resolution flood map allows for the identification of specific fields at risk, demonstrating the utility of this mapping approach for agricultural planning.
• Figure 5: Kenya Flood Map, Spring 2024 | Composite flood extent map of Kenya during the 2024 floods, overlaid with cropland data. The map highlights flood-affected areas, and we estimate that approximately 75,000 hectares of cropland were impacted. This estimate aligns closely with official government statistics, which reported 68,000 hectares affected. The map showcases the utility of SAR data for real-time disaster response and assessment.