Self‐Supervised Classification of Weather Systems Based on Spatiotemporal Contrastive Learning

The classification of weather systems provides a simple description atmospheric circulations and bridges the gap between large-scale conditions local-scale environmental variables. However, existing methods are challenged due to lack labels inaccurate similarity measures data samples. In this letter, we propose self-supervised Spatiotemporal Contrastive Learning (SCL) framework for without manual labels. particular, operate both spatial temporal augmentation on multivariate meteorological fully explore context information stability in accordance with synoptic nature. For assessment comparison results, apply statistical method based analog forecasting using results as input. experimental demonstrate that proposed SCL model outperforms traditional methods.