Application of a Radar Echo Extrapolation‐Based Deep Learning Method in Strong Convection Nowcasting.

Strong convection nowcasting has been gaining importance in operational weather forecasting. Recently, deep learning methods have been used to meet the increasing requirement for precise and timely nowcasting. One of the promising deep learning models is the convolutional gated recurrent unit (ConvGRU), which has been proven to perform better than traditional methods in strong convection nowcasting. Despite its encouraging performance, ConvGRU tends to produce blurry radar echo images and fails to model radar echo intensities that have multi‐modal and skewed distributions. To overcome these disadvantages, we tested the structural similarity (SSIM) and multiscale structural similarity (MS‐SSIM) indexes as loss functions. The SSIM and MS‐SSIM loss functions are composed of luminance, contrast, and structure and provide more information about the intensity, grade, and shape of the radar echo, which can reduce blurring. Due to multi‐layer downscaling, MS‐SSIM extracted more radar echo characteristics, and its extrapolation was the most realistic and accurate among all of the loss function schemes. Only the MS‐SSIM scheme successfully predicted strong radar echoes after 2 h, especially those at the rainstorm level. Plain Language Summary: The Convolutional Recurrent Neural Network model has been considered a pioneering work in the field of nowcasting. However, this model tends to predict blurred radar echo images and fails to model the radar echo intensity that has a multi‐modal and skewed distribution. In this paper, we adopt two measures of structural similarity (SSIM and MS‐SSIM) as loss functions to reduce the blurring problem caused by the mean square error (MSE) loss function for precipitation nowcasting. We have shown that the MS‐SSIM scheme is more efficient in capturing the spatiotemporal correlations, and has the best prediction performance, especially for heavier precipitation. Even for rotating moving radar echoes, the MS‐SSIM scheme has the best forecast performance Key Points: Convolutional gated recurrent unit model was selected to solve the problem of strong convective nowcastingUsing the structural similarity index and multiscale structural similarity (MS‐SSIM) index as a loss function can effectively reduce the blurring problem of extrapolation resultsOnly the MS‐SSIM scheme successfully predicted the strong radar echo after 2 h, especially for the radar echo at the rainstorm level [ABSTRACT FROM AUTHOR]