Abstract: Aiming at  the problem of poor performance exhibited by current deep learning  in extreme weather phenomenon Madden-Julian oscillation (MJO) prediction tasks,  we proposed a time series prediction model based on a combination of dynamic graph neural network and Transformer. Firstly, we mapped the two-dimensional grid of Earth’s land and sea to the nodes of graph structure, 
and proposed  a method of using  multi-attention hybrid sea and land masks for node screening. Secondly, we iteratively updated edge weights based on heat conduction and node similarity measurement to obtain the most accurate climate model information at each time step. Thirdly, we used the maximum extreme value method to extract abnormal node information during different time periods as occurrence points of extreme climate, and strengthened variable weights of these points. Finally, we input above results into a graph neural network for encoding and utilized Transformer for decoding operations  to obtain prediction results. 
Experimental results show that the model can achieve an effective bivariate correlation coefficient (COR) prediction value of up to 39 d as well as an effective root mean square error (RMSE) prediction value of 31 d in prediction, and its performance is superior to existing models.

Key words: spatio-temporal forecasting, graph neural network,  , weather forcasting, time series prediction