Integrating Diurnal Pulsing Signatures for AI‐Driven Tropical Cyclone Intensity Prediction

Abstract Although artificial intelligence (AI) models are increasingly applied to tropical cyclone (TC) intensity prediction, they predominantly rely on instantaneous meteorological variables (e.g., satellite infrared images) while often overlooking more informative temporal evolution signals, such as convective diurnal variation. The diurnal pulse (DP), a daily‐scale expansion of TC upper‐level cold clouds, effectively captures the spatiotemporal evolution of TC convection and is tightly linked to TC intensity changes, particularly rapid intensification (RI). By incorporating the DP as a temporal evolution signal, AI models achieved significantly improved prediction skill for TC intensification rates and RI probabilities, outperforming those using only conventional instantaneous convective indicators. Permutation tests further confirm that the DP predictors contribute more pronouncedly to forecast skill than instantaneous convective proxies. These findings highlight the untapped potential of incorporating physics‐informed, continuous temporal evolution variables into AI models to improve TC intensity forecasting, especially for challenging RI events.