Detecting Hail from Space: A satellite-based computer vision system for global, size-aware detection and risk assessment

Background

Hailstorms represent a significant meteorological hazard, inflicting substantial economic losses on agriculture, infrastructure, and property worldwide. The detection and monitoring of hail events are crucial for mitigating these impacts, supporting insurance claims, and informing disaster response and risk modeling. Traditionally, the field has relied on ground-based radar systems—especially dual-polarization Doppler radars—to infer hail presence. However, these systems are geographically limited, primarily covering developed regions such as the United States and parts of Europe, while vast areas in the developing world, mountainous regions, and remote territories remain underserved. Furthermore, global climate models and regional simulations, while useful for broad-scale prediction, lack the spatial and temporal resolution necessary for precise, event-level hail detection. This gap in observational capability has been highlighted by organizations such as the Intergovernmental Panel on Climate Change (IPCC), which notes low confidence in observed and projected hailstorm trends, underscoring the pressing need for improved, globally consistent hail detection technologies.

Current approaches to hail detection suffer from several critical shortcomings. Ground-based radar systems provide only coarse spatial resolution—often on the order of one square kilometer or larger—making it difficult to localize hail impacts with the precision needed for building-level assessment or detailed climatological analysis. These systems also depend on indirect proxies, such as radar reflectivity signatures and empirical relationships, rather than direct observation of hail at the surface, introducing uncertainty and potential for false positives or negatives. Additionally, hail reports collected from the ground are inherently biased toward populated areas, resulting in incomplete spatial representation and underreporting in rural or inaccessible regions. Satellite-based optical sensors, while offering broader coverage, are hampered by cloud cover and daylight constraints, further limiting their utility for real-time hail detection. As a result, there remains a significant unmet need for a solution that delivers high-resolution, direct, and globally consistent hail detection, overcoming the spatial, temporal, and observational limitations of existing methods.

Technology

This technology utilizes Synthetic Aperture Radar (SAR) satellite imagery, specifically leveraging Sentinel-1’s dual-polarization (VV and VH) data at approximately 10-meter resolution, to enable near real-time, global detection of hailstorms and assessment of hail risk. The process begins by identifying anomalies in SAR images through differencing hail-affected scenes with reference “clean” background scenes, effectively highlighting surface changes caused by hail. Advanced computer vision techniques—including random walk-based image segmentation, morphological operations (dilation, erosion, median filtering), and Density-Based Spatial Clustering of Applications with Noise (DBSCAN)—are then applied to isolate and refine spatially coherent hail damage zones. The system further enhances these zones through feature-preserving upscaling, resulting in contiguous, meter-scale hail impact maps. Additionally, it aggregates historical detection data onto a global grid to generate a hail risk rating, classifying regions by recurrence and severity of hail events. Individual regions can be classified by relative hailstone size, allowing for accurate identification of possible economic impact.

This solution is highly differentiated due to its ability to deliver direct, high-resolution, and globally consistent hail detection, overcoming the significant limitations of traditional ground-based radar and proxy-based methods. Unlike conventional systems, which are restricted by coarse spatial resolution, limited geographic coverage, and indirect inference of hail presence, this approach provides meter-scale precision and operates independently of weather, daylight, or ground infrastructure. Its validation against extensive ground-based hail reports and precipitation datasets demonstrates high detection accuracy and the capability to reveal previously undocumented hail events, even in remote or radar-sparse areas. The integration of a multi-year SAR archive enables robust long-term climatological analysis and risk modeling, while the system’s scalability and adaptability make it a transformative tool for insurance, disaster response, and global severe weather monitoring.

Benefits

• Provides near real-time, meter-scale resolution global hail detection regardless of weather or daylight conditions
• Delivers direct observational confirmation of hail impacts with high spatial precision (~10 meters)
• Offers truly global coverage, including remote and radar-sparse regions lacking ground-based radar infrastructure
• Enables detailed mapping of hail damage at building or street level for improved damage assessment and insurance claims validation
• Generates a long-term, consistent archive of hail data supporting climatological studies, risk modeling, and policy development
• Incorporates an automated hail risk rating system based on historical detection frequency for rapid risk assessment
• Utilizes advanced computer vision algorithms for accurate anomaly detection and spatial clustering, enhancing detection efficiency (~79%)
• Supports disaster response and recovery efforts by providing timely and precise hailstorm impact information

Applications

• Insurance claims validation
• Disaster response coordination
• Global hail risk mapping
• Agricultural damage assessment
• Infrastructure vulnerability analysis

Additional info

This system employs Sentinel-1 SAR imagery, specifically dual-polarized data, to detect hailstorms. It extracts anomalies by differencing scenes, then applies random walk segmentation, morphological operations, and DBSCAN clustering. Feature-preserving upscaling refines results. This delivers near real-time, meter-scale global hail detection and generates a historical hail risk rating.