A random forest-based method for reconstructing pressure fields in tropical cyclones from wind data

Tropical cyclones (TCs) are among the most destructive weather phenomena, causing severe social and economic damage through extreme rainfall, strong winds, and storm surges. Wind and pressure fields, which define TCs’ structure and impact, are spatial variables whose accurate estimation is essential for improved analysis and prediction. However, high-precision observations are limited by cost and coverage, and numerical or reanalysis models are computationally intensive. Parametric models based on radial symmetry also struggle to represent the asymmetry and diversity of real TCs. This study proposes a random forest model that estimates TC pressure fields by learning the nonlinear relationship between wind and pressure. Trained on ERA5 reanalysis data, the model captures spatial patterns such as pressure asymmetry, land–sea contrasts, and shows high explanatory power and low error, especially for intense TCs and within the gale-force wind radius. Comparison with parametric and other data-driven models confirms its superior spatial accuracy. An experiment using satellite-based two-dimensional wind data demonstrates the model’s ability to estimate pressure fields from real time inputs, supporting its use in disaster response and early warning.