A Deep Learning‐Based Methodology for Precipitation Nowcasting With Radar. Issue 2 (15th February 2020)

• Record Type: Journal Article
• Authors: Chen, Lei; Cao, Yuan; Ma, Leiming; Zhang, Junping
• Is Part Of: Earth and space science. Volume 7:Issue 2(2020)
• Page Start: n/a
• Page End: n/a
• Publication Date: 2020-02-15
• Subjects: deep learning -- precipitation nowcasting -- convolutional LSTM -- group normalization; Space sciences -- Periodicals; Geophysics -- Periodicals; 500.5
• DOI: 10.1029/2019EA000812
• Languages: English
• ISSNs: 2333-5084

Abstract

Abstract: Nowcasting and early warning of severe convective weather play crucial roles in heavy rainfall warning, flood mitigation, and water resource management. However, achieving effective temporal‐spatial resolution nowcasting is a very challenging task owing to the complex dynamics and chaos. Recently, an increasing amount of research has focused on utilizing deep learning approaches for this task because of their powerful abilities in learning spatiotemporal feature representation in an end‐to‐end manner. In this paper, we present convolutional long short‐term memory with a layer called star‐shape bridge to transfer features across time steps. We build an end‐to‐end trainable model for the nowcasting problem using the radar echo data set. Furthermore, we propose a raining‐oriented loss function inspired by the critical success index and utilize the group normalization technique to refine the convergence performance in optimizing our deep network. Experiments indicate that our model outperforms convolutional long short‐term memory with the cross entropy loss function and the conventional extrapolation method. Plain Language Summary: From the viewpoint of deep learning, precipitation nowcasting using radar echo could be regarded as a sequence‐to‐sequence learning problem with spatial correlations. In this study, we leverage the ability of convolution operation (for spatial information) and long short‐term memory (for memory of temporal dynamics) by combining additional residual connections, the normalization technology, and an appropriate loss function to improve precipitation nowcasting. Key Points: A novel deep learning neural network is proposed for precipitation nowcasting Group normalization is shown to be effective in training our model with high resolution radar echo images An appropriate loss function is used to improve the relevant prediction performance