To predict the large earthquakes by deep neural network modeling and more generative AI modeling, it is required that available augmentation of multivariate timeseries should be developed for the sets of large earthquakes and background characteristic features inferred from data-driven science investigation of monitoring big data in the earth science. In this chapter, the new augmentation method of multivariate timeseries of the global and regional correlated seismicity and local seismicity rates are proposed, being based on generative diffusion augmentation using Gaussian mixed modeling. In addition, after one-step ahead prediction of multivariate timeseries of correlated seismicity and local seismicity rates, the full prediction of one-year ahead timeseries of large earthquake is possibly performed by 2dCNN modeling. Furthermore, it suggests that the latent space mapping method of dense layer before final node in 2dCNN model is available for the identification of the clusters belonging large earthquake and no earthquake labels.