Predicting Beyond Training Data via Extrapolation versus Translocation: AI Weather Models and Dubai's Unprecedented 2024 Rainfall

Abstract: AI models have transformed weather forecasting, but their skill for gray swan extreme events is unclear. Here, we analyze GraphCast and FuXi forecasts of the unprecedented 2024 Dubai storm, which had twice the training set's highest rainfall in that region. Remarkably, GraphCast accurately forecasts this event 8 days ahead. FuXi forecasts the event, but underestimates the rainfall, especially at long lead times. GraphCast's success stems from "translocation": learning from comparable/stronger dynamically similar events in other regions during training via global effective receptive fields. Evidence of "extrapolation" (learning from training set's weaker events) is not found. Even events within the global distribution's tail are poorly forecasted, which is not just due to data imbalance (generalization error) but also spectral bias (optimization error). These findings demonstrate the potential of AI models to forecast regional gray swans and opportunity to improve them through understanding the mechanisms behind their successes and limitations.