Automated identification and tracking framework for hazard cataloging

Event identification and tracking tools have become essential in weather and climate research due to the rapid growth of weather and climate model data. These algorithms automate the detection of extreme weather-related events and hazardous phenomena and their corresponding features, thereby enhancing predictive capabilities. Our study introduces an open and adaptable framework designed for the automated identification and tracking of various hazard types from a Lagrangian perspective, spanning multiple spatial and temporal scales and providing insights into the dynamics and behavior of hazardous events. This framework is versatile and applicable to a range of scenarios from historical hazard analyses to future climate scenarios evaluation. Additionally, this work facilitates the investigation of weather patterns that give rise to the hazards and enables the assessment of their impacts. The algorithm relies on connected components and distance thresholds to identify and track events in space and time, permitting event merging and splitting dynamics over time. Furthermore, it characterizes events based on diverse spatio-temporal features such as duration, volume, intensity, and trajectory. The framework includes visualization tools for displaying event trajectories. It enables statistical analyses of detected time series derived from historical records and climate model outputs (e.g., identifying trends, exploring correlations between feature pairs, and analyzing feature distributions). To prove the framework's efficacy and versatility, we applied it to detect major hazards including heatwaves, cold spells, heavy precipitation events, hydrological droughts, and even combinations of these events identified as compound events. Leveraging ERA5 reanalysis, satellite data, and radar data we constructed an extended historical catalog of these hazards in France and conducted statistical analyses to characterize event properties and structures. Subsequently, applying the framework to Cordex data allowed us to assess potential future hazard scenarios.