An Input Sampling Scheme to Radar Echo Extrapolation For RNN-Based Models

2022 IEEE Intl Conf on Dependable, Autonomic and Secure Computing, Intl Conf on Pervasive Intelligence and Computing, Intl Conf on Cloud and Big Data Computing, Intl Conf on Cyber Science and Technology Congress (DASC/PiCom/CBDCom/CyberSciTech)(2022)

引用 1|浏览13
暂无评分
摘要
Short-term heavy rainfall can have a significant impact on people's production, life and travel. Numerical Weather Prediction (NWP) is complex. It can predict weather conditions for the next week or even two weeks, but cannot predict the weather in the near hours in a timely manner. The TREC and Optical Flow methods used in the meteorological field are only good at translational transformations, and it is difficult to predict the generation and dissipation of cloud systems. ConvLSTM, PredRNN, and other deep RNN-based methods input only one map at each time step, and do not make good use of the spatial information of several neighboring moments. Inspired by schedule sampling, a new sampling scheme is proposed to allow the model to effectively use the spatial information of the past few times. Also, the advantages of MSE and MAE are combined to change the loss function.
更多
查看译文
关键词
radar echo extrapolation,precipitation nowcasting,sampling scheme,ConvLSTM,spatio-temporal sequence prediction
AI 理解论文
溯源树
样例
生成溯源树,研究论文发展脉络
Chat Paper
正在生成论文摘要