Convolutional Neural Network for Convective Storm Nowcasting Using 3-D Doppler Weather Radar Data

Convective storms are one of the severe weather hazards found during the warm\nseason. Doppler weather radar is the only operational instrument that can\nfrequently sample the detailed structure of convective storm which has a small\nspatial scale and short lifetime. For the challenging task of short-term\nconvective storm forecasting, 3-D radar images contain information about the\nprocesses in convective storm. However, effectively extracting such information\nfrom multisource raw data has been problematic due to a lack of methodology and\ncomputation limitations. Recent advancements in deep learning techniques and\ngraphics processing units now make it possible. This article investigates the\nfeasibility and performance of an end-to-end deep learning nowcasting method.\nThe nowcasting problem was transformed into a classification problem first, and\nthen, a deep learning method that uses a convolutional neural network was\npresented to make predictions. On the first layer of CNN, a cross-channel 3D\nconvolution was proposed to fuse 3D raw data. The CNN method eliminates the\nhandcrafted feature engineering, i.e., the process of using domain knowledge of\nthe data to manually design features. Operationally produced historical data of\nthe Beijing-Tianjin-Hebei region in China was used to train the nowcasting\nsystem and evaluate its performance; 3737332 samples were collected in the\ntraining data set. The experimental results show that the deep learning method\nimproves nowcasting skills compared with traditional machine learning methods.\n