GraphCast and AIFS AI Models Detected Dubai Floods 8 Days Before They Occurred

The artificial intelligence models GraphCast and AIFS detected signals of the exceptional rainfall that struck Dubai in April 2024 eight days before it occurred, according to a study published in Science Advances.

The study showed that the two models successfully predicted the timing, location, and intensity of the event eight days in advance, despite the fact that their training data did not include any similar event in Dubai. In contrast, a third model called FuXi predicted the event but significantly underestimated the actual rainfall amount.

The study was led by Chiang Sun, a researcher at the School of Atmospheric Sciences at Nanjing University and the principal author of the research, who described the rainfall as a stringent test of artificial intelligence's ability to forecast rare and unprecedented climate events in specific regions, particularly given that the models had not encountered a similar event in Dubai during the training phase.

The researchers relied on reanalysis data from the European Centre for Medium-Range Weather Forecasts (ECMWF) as a reference for comparison. The storm, which struck in mid-April 2024, caused widespread disruption to transportation and severe damage to Dubai's infrastructure.