Variable Importance Measures based on Ensemble Learning Methods for Convective Storm Tracking

Convective storms are hazardous weather events that cause significant damages to life and property. Advanced devices allow observing the storms in the operations, especially Doppler radars, which can provide a wide observation area with high spatiotemporal resolution. Consequently, meteorological experts should make more considerable effort to derive analysis results quickly and accurately. Recently, machine learning-based methods have been presenting solutions to relieve the burden. There are essential issues to apply machine learning methods, such as model selection, feature extraction, and feature selection. In this paper, we focused on the feature selection for implementing convective storm tracking method based on ensemble methods. By actual observation radar data and derived variable importance measures, it is possible to conclude that the reduced features make better performances than others.