The Development of a Consensus Machine Learning Model for Hurricane Rapid Intensification Forecasts with Hurricane Weather Research and Forecasting (HWRF) Data

Abstract

This study focused on developing a consensus machine learning (CML) model for tropical cyclone (TC) intensity-change forecasting, especially for rapid intensification (RI). This CMLmodelwas built upon selected classical machine learning models with the input data extracted from a high-resolution hurricane model, the HurricaneWeather Research and Forecasting (HWRF) system. The input data contained 21 or 34 RI-related predictors extracted from the 2018 version of HWRF (H218). This study found that TC inner-core predictors can be critical for improving RI predictions, especially the inner-core relative humidity. Moreover, this study emphasized that the importance of performing resampling on an imbalanced input dataset. Edited Nearest Neighbor and Synthetic Minority Oversampling Technique improved the Probability of Detection (POD) by ∼10% for the RI class. This paper also showed that the CML model has satisfactory performance on RI predictions compared to the operational models. CML reached 56% POD and 46% False Alarm Ratio (FAR), while the operational models had only 10 to 30% POD but 50 to 60% FAR. The CML performance on the non-RI classes was comparable to the operational models. The results indicated that, with proper and sufficient training data and RI-related predictors, CML has the potential to provide reliable probabilistic RI forecasts during a hurricane season.

Predicting rapid intensification of tropical cyclones in the western North Pacific: a machine learning and net energy gain rate approach

“…(2015); henceforth referred to as KRD15). Ko et al (2023) explored the application of a consensus machine learning (CML) model in TC intensity change forecasting and indicated the CML exhibits better performance on RI predictions compared to the operational models such as SHIPS, GFS. Narayanan et al (2023) proposed a simple deterministic binary classification model based on the cooccurrence of environmental parameters (MCE) to predict an RI event.…”

“…Before the applying ML methods in TC intensity forecasting, it is known that the statistical-dynamical-based forecast models using climatological, persistence, and numerical model predictors provide the highest skill in intensity (Goldenberg et al, 2015;Kim et al, 2018). Yamaguchi et al, 2018;Xu et al, 2021;Ko et al, 2023). The statistical-dynamical model developed by Kim et al (2018) showed the smallest mean absolute errors at short lead time (up to 24 h) for TC intensity prediction compared to operational dynamical forecast models (Kim et al, 2018).…”

“…The MLP model correctly predicted more RI events than other operational TC intensity models as well as outperformed the statistical-dynamical models such as SHIPS, DSHIPS and LGEM by 5-22% in simulating real-time operational forecasts (Xu et al, 2021). A Consensus Machine Learning (CML) model with the input data extracted from HWRF for TC intensity change, especially for RI reached 56% the probability of detection (POD) and 46% the false alarm ratio (FAR), while the operational models (GFS, HWRF, SHIPS) had only 10-30% POD but 50-60% FAR (Ko et al, 2023).…”