Author(s): Noor Ahmad Yaqubi; Biswa Bhattacharya; Ioana Popescu
Linked Author(s): Ioana Popescu
Keywords: AI; Extreme rainfall; Rainfall forecasting; Localized extremes; Machine learning
Abstract: This study conducts a systematic global review of AI-based rainfall forecasting techniques with a specific focus on short-duration, localized extreme rainfall events. Using case studies from diverse climatological regions worldwide, the review presents how extreme rainfall is defined, how AI models are trained and evaluated for such events, and which studies test performance under extreme conditions. We compare forecast skill and useful lead time for both normal and high-intensity rainfall, highlighting how model performance changes during extremes. The review also identifies which atmospheric predictors, data sources, and input resolutions are most influential, and how local observations contribute to detecting and forecasting localized extremes. Together, these insights provide a clearer understanding of how AI approaches and configurations can be improved to better capture short-lived, high-intensity rainfall relevant for urban flood management.
Year: 2026