ENSO-based Ensemble Learning Approach for Tropical Cyclone Intensity Estimation

Estimation of tropical cyclones (TCs) intensity is crucial for disaster reduction and risk management. This study aims to estimate TC intensity using ensemble machine learning models. We utilized four Random Forest (RF) models to predict TC intensity. The input factors include TC’s location, central pressure, distance to land, landfall in the next six-hour, storm speed, and storm direction, which are extracted from the International Best Track Archive for Climate Stewardship Version 4 (IBTrACS V4). This dataset was divided into four sub-datasets based on the ENSO phases (Neutral, El Niño, and La Niña). Inputs for the RF models were taken from each sub-dataset separately. Results show that central pressure has the greatest effect on TC intensity estimate with the maximal root mean square error (RMSE) of 4.92 knots (equivalent to 2.50 m/s) and accuracy of 94-96 %.