Convolutional long short-term memory (ConvLSTM) and convolutional gated recurrent unit (ConvGRU) are 2 widely adopted deep learning models that combine recurrent mechanisms with convolutional operations for spatio-temporal sequences forecasting. To clarify convergence speed and classification ability of the above 2 models, using the same model architecture to predict a same classification problem is in need. This research treats transforms the short-duration heavy rainfall (SDHR) district-level warning problem in Beijing into as a binary classification problem in deep learning, and composite radar reflectivity data of Beijing-Tianjin-Hebei radar network and rainfall data from automatic weather stations in Beijing are used for training and performance evaluation. The results show that, the convergence speed of ConvGRU is approximately 25% faster than that of ConvLSTM. The early-warning performance of ConvLSTM and ConvGRU has the similar trend with region, time, rain intensity, but most of the scores of ConvLSTM are higher, and in a few cases, ConvGRU has higher scores.

deep learning,convolutional long short-term memory,convolutional gated recurrent unit,classification