Analysis of the Atmospheric Duct Existence Factors in Tropical Cyclones Based on the SHAP Interpretation of Extreme Gradient Boosting Predictions

The atmospheric duct (AD) is an anomalous structure in which electromagnetic waves can make transhorizon propagation. ADs often occur in the formation, development and disappearance of tropical cyclones (TCs). In this work, the eXtreme Gradient Boosting (XGBoost) model is used to predict TC ducts and a relatively high accuracy of 81.3% is obtained. Shapely additional explanations (SHAP) values of the features including TC parameters and local meteorological parameters are employed to interpret XGBoost model predictions of the TC ducts existence. Furthermore, the importance ranking of the features is revealed, among which the distance between dropsondes and TC eyes is the most important. In addition, the detailed relationships between the AD existence and the features are presented. Hence, this work can not only improve the knowledge of the relationship between TC ducts and the features, but also be of great value to the ducts prediction.