Cooperative Institute for Research in the Atmosphere

Evaluation of Tropical Cyclone Track and Intensity Forecasts from Artificial Intelligence Weather Prediction (AI-WP) Models

Weather prediction (WP) models based on artificial intelligence (AI) have proliferated over just the past few years. This study evaluates the utility of AI-based weather prediction for tropical cyclone track intensity forecasting. Four AI-WP models are evaluated for northern hemisphere 2023 tropical cyclones from May-November using National Hurricane Center verification procedures. Results show that the track forecasts are comparable to those from the best physically based NWP models. However, the intensity forecasts have no skill relative to even the simplest statistical models, due to an extreme low bias in the prediction of the maximum wind. The low intensity bias is explained by consideration of the least-squares minimization of a misplaced idealized vortex between successive forecast times in the AI-WP models. These results show that the utility of current AI-WP models is highly variable, depending on what phenomena are being predicted.