基于时空预测网络的雷达回波外推研究

Radar echo extrapolation research based on spatiotemporal prediction network

短临降水预测由于气象数据体量大、种类繁多,以及大气系统的复杂性,预测难度大。拟构建一个基于时空预测网络的雷达回波外推模型来提高预测性能。该网络旨在将时间特征和空间特征进行解耦,独立提取特征。空间模块通过注意力机制建模时间不变信息,时间模块通过级联的门控机制建模时间依赖。最后,在雷达回波数据集上验证了模型的性能。

Short-term precipitation prediction is difficult due to the large volume and variety of meteorological data,as well as the complexity of atmospheric systems.We propose to construct a radar echo extrapolation model based on a spatiotemporal prediction network to improve the prediction performance.The network aims to decouple temporal and spatial features and extract features independently.The spatial module models time-invariant information through an attention mechanism,and the temporal module models temporal dependence through a cascaded gating mechanism.The performance of the model is validated on a radar echo dataset.