Extreme temperature events are no longer isolated anomalies appearing once in a generation. Called Hankelformer, the model combines structured time-series augmentation with contrastive learning to improve the prediction of non-stationary and extreme events. The second innovation is a dual-stream contrastive learning framework. By requiring the original and augmented views to agree at the representation level, contrastive learning encourages Hankelformer to capture more invariant characteristics of the dynamics. Contrastive learning appears to provide the mechanism that aligns these views, allowing the network to benefit from the extra temporal structure without becoming confused by it.