基于时空图卷积的强对流降水临近预报研究
A study on the proximity prediction of strong convective precipitation based on the spatio-temporal graph convolution
方巍 1薛琼莹 2陶恩屹 2齐媚涵2
作者信息
- 1. 南京信息工程大学计算机学院/数字取证教育部工程研究中心,南京 210044;中国气象局交通气象重点开放实验室/南京气象科技创新研究院,南京 210041;南京信息工程大学江苏省大气环境与装备技术协同创新中心,南京 210044
- 2. 南京信息工程大学计算机学院/数字取证教育部工程研究中心,南京 210044
- 折叠
摘要
降水临近预报对于强对流天气的预报具有重要的支撑作用.气象业务中主要采用雷达回波外推方法解决此问题.然而,现有方法通常缺乏从序列雷达数据中有效学习的能力,导致预测精度不佳.为了解决这一问题,本文提出了一种改进的时空图卷积模型ASTGCN(A Spatio-Temporal Graph Convolution Neural Network)用于强对流降水的临近预报.利用时空图卷积网络,有效地捕获相邻雷达帧之间的时间依赖性.此外,利用注意力机制和自动编码器来增强模型捕获时空相关性的能力.结果表明,该模型可以从数据中发现隐藏的图结构,从而捕获隐藏的空间关系.与现有模型(Transformer)相比,该模型的临界成功指数(CSI)提高了 28％,表明其在强对流降水临近预报方面具有优越的性能.
Abstract
Precipitation nowcasting plays an important supporting role in forecasting severe convective weather.In meteorological services,the radar echo extrapolation method is mainly used to solve precipitation nowcasting problems.However,existing methods often lack the ability to effectively learn from sequential radar data,resulting in poor prediction accuracy.In order to solve this problem,this paper proposed ASTGCN(A Spatio-Temporal Graph Convolution Neural Network)for nowcasting of severe convective precipitation.Efficiently capture the temporal dependence between adjacent radar frames using a spatio-temporal graph convolutional network.In addition,an attention mechanism and an autoencoder were utilized to enhance the model's ability to capture spatio-temporal correlations.Experimental results show that the model can discover hidden graph structures from data and thereby capture hidden spatial relationships.Compared with the existing model(Transformer),the Critical Success Index(CSI)of this model is improved by 28％,indicating its superior performance in the nowcasting of severe convective precipitation.
关键词
强对流降水临近预报/深度学习/ASTGCN模型/注意力机制/雷达回波外推Key words
nowcasting of severe convective precipitation/deep learning/ASTGCN model/attention mechanism/radar echo extrapolation引用本文复制引用
基金项目
国家自然科学基金资助项目(42075007)
苏州大学计算机信息处理技术重点实验室开放项目(KJS2275)
南京气象科技创新研究院北极阁开放研究基金资助项目(BJG202306)
江苏省研究生科研与实践创新计划项目(KYCX23_1388)
出版年
2024
国家科技期刊平台
NSTL
万方数据