Prediction of Tropical Cyclogenesis Based on Machine Learning Methods and Its SHAP Interpretation

Abstract This study trains three machine learning models with varying complexity—Random Forest, Support Vector Machine, and Neural Network—to predict cyclogenesis at a forecast lead time of 24 hr for given tropical disturbances identified by an optimized Kalman Filter algorithm. The overall performance is competent in terms of f1‐scores (∼0.8) compared to previous research of the same kind. An assessment by SHapley Additive exPlanations (SHAP) values reveals that mid‐level (500 hPa) vorticity is the most influential factor in deciding if a tropical disturbance is developing or non‐developing for all three models. Wind shear and tilting are found to hold a certain level of importance as well. These results encourage further experiments that use physical models to explore the dynamical, mid‐level pathway to tropical cyclogenesis. Another usage of SHAP values in this work is to explain how a machine learning model decides if an individual tropical disturbance case will develop, by listing the contribution of each feature to the output genesis probability, illustrated by a case study of Typhoon Halong. This increases the reliability of the machine learning models, and forecasters can take advantage of such information to issue tropical cyclone formation warnings more accurately. Several caveats of the current machine learning application in the studies of tropical cyclogenesis are discussed and can be considered for future research. These can benefit the interpretation and emphasis of certain output fields in the operational dynamical prediction system, which can contribute to more timely cyclogenesis forecasts.