Forecasting formation of a Tropical Cyclone Using Reanalysis Data

TL;DR

This work tackles the challenge of forecasting tropical cyclone genesis by leveraging high-resolution ERA5 reanalysis data and IBTrACS genesis records across six basins. It introduces a CNN–LSTM architecture with TimeDistributed CNN layers and stacked LSTMs to capture the spatiotemporal dynamics of genesis, using input windows of as little as $12$ hours to predict formation up to $60$ hours ahead. Across 5-fold validation, the model achieves high accuracy, precision, and recall, with especially strong performance at $24$–$48$ hours lead times and robust results up to $60$ hours. The findings suggest that reanalysis data contain sufficient information to anticipate TC formation, offering potential for real-time mitigation and further improvements with additional variables and imbalanced sampling strategies.

Abstract

The tropical cyclone formation process is one of the most complex natural phenomena which is governed by various atmospheric, oceanographic, and geographic factors that varies with time and space. Despite several years of research, accurately predicting tropical cyclone formation remains a challenging task. While the existing numerical models have inherent limitations, the machine learning models fail to capture the spatial and temporal dimensions of the causal factors behind TC formation. In this study, a deep learning model has been proposed that can forecast the formation of a tropical cyclone with a lead time of up to 60 hours with high accuracy. The model uses the high-resolution reanalysis data ERA5 (ECMWF reanalysis 5th generation), and best track data IBTrACS (International Best Track Archive for Climate Stewardship) to forecast tropical cyclone formation in six ocean basins of the world. For 60 hours lead time the models achieve an accuracy in the range of 86.9% - 92.9% across the six ocean basins. The model takes about 5-15 minutes of training time depending on the ocean basin, and the amount of data used and can predict within seconds, thereby making it suitable for real-life usage.

Figures (4)

• Figure 1: Location of TC and non TC formation.
• Figure 2: Pictorial representation of reanalysis data.
• Figure 3: Model description for $T = 3$.
• Figure 4: Epoch wise train versus test accuracy for T = 3.