Deep learning, a machine learning method based on neural networks and large amount of data, has been applied to various pattern recognition tasks in recent years. We have developed automatic detection methods of tropical cyclones and their precursors (Matsuoka et al., 2018) and stationary fronts (Matsuoka et al., 2019). In these works, deep convolutional neural networks are trained to detect characteristic atmospheric patterns from a cloud-resolving simulation/reanalysis data. In this talk, I will present the details of the methodology, results, and future prospects including application to analyze large-ensemble future projection data.
Deep learning for detecting tropical cyclones and stationary fronts