Development of a Short‐Term Forecasting Model for Tropical Cyclone Tracks Based on Ventilation Flow and Artificial Intelligence Algorithms

Abstract While research on forecasting tropical cyclone (TC) intensity via statistical methods is relatively abundant, studies on TC track forecasting using such methods remain scarce. Existing statistical TC track forecasts primarily rely on time‐series factors of TC motion; however, their efficacy diminishes significantly during TC track deflections, largely because most statistical methods lack physically meaningful factors governing TC motion. To address this, this study focuses on short‐term TC track forecasting by examining two physically meaningful factors—ventilation flow and steering flow—as predictors. It clarifies ventilation flow calculation procedures, develops a forecasting model integrating these flows and other factors, and uses 1990–2021 JTWC, ERA5, and SHIPS data sets. Analyzing their characteristics across varying radii, vertical layers, and calculation methods (with/without subtracting TC translation speed), performance is evaluated via linear regression and 6 artificial intelligence (AI) models. Results show ventilation flow (without subtracting translation speed) aligns better with actual TC movement; mid‐upper and whole‐layer average airflow correlates more strongly with TC motion. Incorporating steering/ventilation flow into time‐series and SHIPS factors improves forecasts. Among AI models, XGB performs best though some exhibit overfitting. This work provides a new indicator for short‐term TC track forecasting, clarifies ventilation flow's utility, and suggests integrating it into large AI meteorological models to enhance TC track simulation.