Application of Machine Learning for Secondary Eyewall Formation Prediction in HWRF and HMON Models

The frequent occurrence of secondary eyewalls in major hurricanes and their close association with storm intensity change and duration have sparked a great interest in developing secondary eyewall forecasting tools. However, the cause of their formation still needs further investigation even though many competing hypotheses are proposed. Each hypothesis is supported by physical theories and/or observations in case studies. Meanwhile, machine learning does not require a complete understanding of the physical processes of the atmosphere features and is effective in estimating causal effects and uncovering the "hidden insights" by learning from historical relationships and trends in the data. Therefore, in this work, we aim to study how each hypothesized condition in convective activities, sustained wind, rain bands, heat flux, boundary layers, etc., contribute to and/or co-play to affect the formation of the secondary eyewalls through machine learning methods. We adopt the state-of-the-art machine learning algorithms in decision trees and Bayesian networks to work on the data of the major storms from 2017 HWRF and HMON model outputs. 50% of the data is used for training, 25% for validation, 25% for testing the algorithms. Our findings, which include the identification and quantification of the influential factors on SEF, may help achieve a consensus on the causes of SEF and predictions of SEF. Performances of different machine learning algorithms are compared and analyzed as well.