As climate change progresses, extreme weather and climate events are becoming more complex. Often, it is not a single hazard that causes major damage, but a combination—such as drought followed by heatwaves, or storms coinciding with heavy rainfall. These 'compound events' can lead to unexpected and widespread impacts on people, ecosystems, and infrastructure. However, most current forecasting systems are not well equipped to anticipate such interactions.

This talk will present ways to improve early warning of compound climate risks by combining two types of approaches: data-driven methods like machine learning, and process-based knowledge from physical climate science. Together, these methods can help detect warning signals early and better understand how different climate factors interact to produce dangerous outcomes.

The presentation will also highlight the importance of building specialized datasets that focus on past compound events—recording when, where, and how different hazards combined and what their impacts were. These datasets are essential for developing better prediction tools and for testing how well models perform under complex real-world conditions.

By integrating these innovations, the talk will outline a path toward more effective early warning systems that can support timely decisions and reduce the risks of future climate-related disasters.