A novel spatial-temporal-decoupled masked pre-training framework (STD-MAE) that effectively captures spatial and temporal heterogeneity to enhance the performance of downstream spatiotemporal predictors.
