AI Agent Links Climate Events to Real-World Impact

This paper introduces the CDEP Agent, an auditable LLM-agent framework designed to bridge the gap between meteorologically identified compound drought-to-extreme-precipitation (CDEP) events and their documentation in real-world records. Climate science recognizes CDEP events as significant drivers of extreme impacts, yet it's unclear if these events are adequately captured by existing early warning and post-event reporting systems. The CDEP Agent addresses this by linking detected CDEP candidates from meteorological reanalysis data to various sources of hazard and impact evidence, which often have differing spatial scales, temporal resolutions, and reporting conventions. Using California as a case study, the framework identified 408 candidate CDEP events between 2021 and 2025. Each candidate was evaluated against sources like the U.S. Drought Monitor, NOAA Storm Events, and public webpages across five dimensions. The findings were stark: only 34.3% of candidates were corroborated on both hazard components, and a mere 1.5% were explicitly linked to their antecedent drought. This indicates that the majority of meteorologically detected CDEP events are not documented, and their compound nature is almost never recorded, revealing a significant deficiency in current warning and reporting mechanisms. The CDEP Agent provides a crucial tool for climate scientists to validate physical event definitions against actual documented impacts and offers social scientists and disaster-response agencies a provenance-linked evidence base for these often-overlooked compound events.