HFIP Monthly Seminar

Feb 16, 2022, 2:00 - 3:00pm ET



Title Presenter Presentation
2:00 - 3:00pm The Development of a Consensus Machine Learning Model for Hurricane Rapid Intensification Forecasts with HWRF Data (abstract)

This study focuses on developing a consensus machine learning model (CML) model for tropical cyclone (TC) intensity probabilistic forecasting, especially for rapid intensification (RI). This CML model is built based on six classical machine learning models with the input data extracted from a high-resolution hurricane model, the Weather Research and Forecasting (HWRF) system. The input data contains 20 to 34 RI-related predictors extracted from the 2018 version of HWRF (H218). This study found that TC inner-core predictors can be critical for improved RI predictions. Inner-core relative humidity is identified as one of the most influential predictors in our input dataset. Moreover, the importance of performing resampling on an imbalanced input dataset is also emphasized in this paper. Edited Nearest Neighbor and Synthetic Minority Oversampling Technique used for resampling can improve Probability of Detection (POD) by about 10% for the RI class. This paper also shows that the CML model built based on the rebalanced input data has satisfactory performance on RI predictions. CML can reach about 47% POD but with less than 50% false alarm ratio (FAR), while the HWRF system had only 15% POD and 40% FAR. The CML model was further tested with the HWRF data from 2019. The performance was slightly degraded compared to the results with H218, possibly due to limited training data. The results indicate that, with proper and sufficient training data, CML has the potential to provide reliable probabilistic RI forecasts during a hurricane season.

Mu-Chieh Ko (Laura) and Xiaomin Chen (bio)