Incorporating Hourly Convective Cloud Data Into Tropical Cyclone Rapid Intensification Forecasting With Machine Learning

• Published in: Journal of geophysical research. Machine learning and computation Vol. 2; no. 1
• Main Authors: Wu, Qiaoyan, Luo, Tong, Hong, Jiacheng
• Format: Journal Article
• Language: English
• Published: Wiley 01.03.2025
• Subjects: diurnal cycle; machine learning; rapid intensification; tropical cyclone
• ISSN: 2993-5210; 2993-5210
• DOI: 10.1029/2025JH000595

In this study, we developed a machine learning (ML) model to predict the rapid intensification (RI) of North Atlantic tropical cyclones (TCs) using 6‐hourly Statistical Hurricane Intensity Prediction Scheme (SHIPS) predictors and additional data on very deep convective clouds with an infrared brightness temperature below 208 K. The presence of these clouds is considered a precursor to TC RI. The ML model, which incorporates SHIPS data with hourly cloud coverage, outperformed the ML model with 6‐hourly coverage of very deep convective clouds for TCs in the Atlantic basin from 2018 to 2023, as indicated by improvements in the Brier Skill Score by 5.9%, 9.9%, 1.0%, and 11.3%, for RI thresholds of ≥25, 30, 35, and 40 knots in 24 hr, respectively. These results highlight the potential of hourly cloud data, with pronounced diurnal variations, to enhance TC RI forecasting accuracy. Plain Language Summary Improvements in forecasting intensity have accelerated in the past years, yet predicting the rapid intensification (RI) of tropical cyclones (TCs) remains challenging. TC RI has been linked to the presence of very deep convective clouds with infrared brightness temperature below 208 K, which show diurnal variations in areal extent, peaking in the early morning and diminishing in the afternoon. This study explored how a machine learning‐based statistical forecast model, incorporating hourly cloud coverage in addition to 6‐hourly Statistical Hurricane Intensity Prediction Scheme (SHIPS) data, can enhance RI predictive accuracy. The ML model, incorporating both SHIPS data and hourly cloud coverage, demonstrated superior performance compared to the ML model with 6‐hourly coverage of very deep convective clouds over the period from 2018 to 2023. This improvement was evident in the Brier Skill Score, with relative enhancements of 5.9%, 9.9%, 1.0%, and 11.3%, for RI thresholds of ≥25, 30, 35, and 40 knots in 24 hr, respectively. These results highlight the potential of using more frequent hourly cloud data to enhance the accuracy of TC RI forecasting. Key Points Very deep convective clouds with IR BT <208 K, showing clear diurnal variations, signal rapid intensification of TCs Machine learning integrating environmental factors with hourly convective cloud data improves probabilistic forecasting of TC rapid intensification Machine learning integrated hourly convective cloud data improves TC rapid intensification deterministic forecasting and detection accuracy