23AI An Updated Consensus Machine Learning Modeling Using the HAFS dataset for Hurricane Rapid Intensification (RI) Prediction

In this project, a rapid intensification (RI) index for tropical cyclones (TCs) is developed based on consensus machine learning (CML) that is trained with the data from NOAA’s new, operational Hurricane Analysis and Forecast System (HAFS) model. The CML-based RI index (CML-RII) provides guidance for the probability of RI (TC maximum sustained winds increase by at least 30 kt in a 24-h period). Text and graphical products will be created to convey CML-RII information to forecasters and the TC community. Furthermore, the CML-RII is validated against an RI index based on the Statistical Hurricane Intensity Prediction Scheme (SHIPS-RII) and the HAFS model intensity forecasts. The CML-RII described here was originally tested on the Hurricane Weather Research and Forecasting (HWRF) model, and this work advances the CML-RII by training it with HAFS retrospective forecasts and updating its core machine learning models. The HAFS dataset spans three years of TC forecasts in the North Atlantic basin, including 34 RI-related predictors (e.g., wind shear across different layers and relative humidity in the TC inner core and environment). Because RI is a rare event on the spectrum of TC intensity change, the Boulder-Line Synthetic Minority Oversampling TEchnique (BL-SMOTE) is applied to up-sample RI to create a balanced-class dataset that improves the accuracy of the CML-RII. This updated CML-RII uses XGBoosting instead of decision trees, and an additional artificial neural network is used for weight assignment across the core ML models.