Tropical Cyclones Intensity Prediction in the Western North Pacific Using Gradient Boosted Regression Tree Model

As an artificial intelligence method, machine learning (ML) has been widely used in prediction models of high-dimensional datasets. This study proposes ML the Gradient Boosted Regression Tree (GBRT), to predict intensity changes tropical cyclones (TCs) Western North Pacific at 12-, 24-, 36-, 48-, 60-, and 72-h (hr) forecasting lead time model is optimized by Bayesian Optimization algorithm. The predictands are TCs different times, obtained from best track data Shanghai Typhoon Institute (STI) Joint Warning Center (JTWC) 2000 2019. predictors synoptic variables, climatological persistent variables derived reanalysis National Centers for Environmental Prediction (NCEP), sea surface temperature (SST) Oceanic Atmospheric Administration (NOAA). results show that GBRT can capture well succeeding 12-h, 24-h, 36-h, 72-h. Compared with traditional multiple linear regression (MLR) model, better performance predicting changes. MLR R 2 forecast increases average 8.47% 4.45% STI JTWC data. MAE (RMSE) drops 26.24% (25.14%) 10.51% (4.68%) two datasets, respectively. potential future change (POT), during previous 12 h (Dvmax), Initial storm maximum wind speed (Vmax), SST, Sea-Land ratio most significant over Pacific.