Early warning of complex climate risk with integrated artificial intelligence

Abstract

As our climate undergoes significant shifts, both the Earth and human societies are increasingly exposed to disasters and stress. This situation underscores the critical need for robust Early Warning Systems (EWS), which are intricately designed to monitor, assess, and relay information about impending risks and hazards. EWS are vital in promoting resilient and sustainable development, yet they encounter substantial challenges in forecasting hazards and impacts, communicating risks, and in the efficiency of decision-making processes. In this perspective, we examine these challenges and explore the transformative role of integrated Artificial Intelligence Foundation Models (AI FMs), especially focusing on the capabilities of Large Multi-Modal Models (LMMs). We discuss the power of these models in developing a Multi-Hazard Early Warning System (MHEWS), combining Meteorological and Geospatial FMs for impact prediction in a comprehensive approach. Emphasizing a user-centric strategy, this paper highlights the importance of intuitive interfaces and incorporating community feedback to enhance crisis management. Given the complex nature of climate risks, we emphasize the need for causal representations in AI models, to avoid conclusions and predictions based on spurious correlations. Additionally, we introduce the concept of decadal EWSs, which aims to provide longer term, yet spatially resolved forecasts. By leveraging climate ensembles and generative approaches, these advancements aim to provide proactive solutions for evolving climate dynamics and societal responses.