Hail Storms Recognition Based on Convolutional Neural Network

Abstract

Hail storm happens when hydrometeors are brought up above the frozen height level by strong updrafts. Before the time of hail fall, an abnormal 3D spatial structures of clouds could be captured by weather radar. If this pattern of 3D structure can be identified, hail storms are able to be forecasted. Based on these, an automatic hail recognition algorithm using deep learning method is proposed in this paper. The deep learning model we use is Convolutional Neural Network (CNN), whose inputs are three-dimensional storm cells including nine slices at different altitudes. The slices’ orientation is corrected according to the location of the bounded weak echo recognition (BWER), and the sizes of these slices are normalized using the cell cores as the reference points. Experiment on real weather cases shows that the method proposed in this paper has better performance than the traditional POSH method. It has a higher hit ratio and reduces the false alarm ratio significantly.