GAN–argcPredNet v1.0: a generative adversarial model for radar echo extrapolation based on convolutional recurrent units

[摘要] Precipitation nowcasting plays a vital role in preventingmeteorological disasters, and Doppler radar data act as an important inputfor nowcasting models. When using the traditional extrapolation method it is difficult tomodel highly nonlinear echo movements. The key challenge of the nowcastingmission lies in achieving high-precision radar echo extrapolation. In recentyears, machine learning has made great progress in the extrapolation ofweather radar echoes. However, most of models neglect the multi-modalcharacteristics of radar echo data, resulting in blurred and unrealisticprediction images. This paper aims to solve this problem by utilizing thefeatures of a generative adversarial network (GAN), which can enhance multi-modal distribution modeling,and design the radar echo extrapolation model GAN–argcPredNet v1.0. The modelis composed of an argcPredNet generator and a convolutional neural networkdiscriminator. In the generator, a gate controlling the memory and output isdesigned in the rgcLSTM component, thereby reducing the loss ofspatiotemporal information. In the discriminator, the model uses adual-channel input method, which enables it to strictly score according tothe true echo distribution, and it thus has a more powerful discrimination ability.Through experiments on a radar dataset from Shenzhen, China, the resultsshow that the radar echo hit rate (probability of detection; POD) and critical success index (CSI)have an average increase of 21.4 % and 19 %, respectively, compared with rgcPredNetunder different intensity rainfall thresholds, and the false alarm rate(FAR) has decreased by an average of 17.9 %. We also found a problem during the comparison of theresult graph and the evaluation index. Therecursive prediction method will produce the phenomenon that the predictionresult will gradually deviate from the true value over time. In addition,the accuracy of high-intensity echo extrapolation is relatively low. This isa question worthy of further investigation. In the future, we will continueto conduct research from these two directions.

[学科分类] 土木及结构工程学