Early Warning Systems & AI

A critical next step for EWS is to move beyond just hazard prediction to impact-centric forecasting. AI models achieve this by using weather forecasts as predictors for impact-related quantities, trained on historical datasets that map weather to its consequences. For instance, AI can predict the impact of weather extremes on vegetation status using satellite imagery, aligning past weather data with satellite products. Beyond traditional meteorological and geospatial data, deep neural networks can incorporate diverse predictors such as social media posts, archived reports, and radio news, which are typically ignored by current EWS. This holistic approach allows for a more comprehensive understanding and prediction of the real-world effects of climate events, informing better preparedness and response strategies.

The integration of AI into EWS introduces critical concerns regarding biases and distribution shifts. AI models, especially foundation models, can implicitly learn from large datasets, but this also means they can perpetuate or even amplify biases present in the training data, such as geographical disparities in observation data or societal biases related to ethnicity, gender, and age. A key challenge is ensuring that AI models respect causal mechanisms to avoid spurious predictions and can generalize to extreme events not seen in training data, which is vital for climate change scenarios. The FATES (Fairness, Accountability, Transparency, Ethics, and Sustainability) principles are emphasized, advocating for user engagement in design, open-source models, transparent standards, and international cooperation to mitigate risks and ensure equitable access, particularly for the Global South and disadvantaged communities. This fosters trustworthy and effective AI-based EWS that serve all.