PrecipJEPA: JEPA-Regularized Future-State Prediction with Motion-Source Rendering for Precipitation Nowcasting

2026-09-30

AI Summary

This study addresses the challenges of modeling radar echo evolution and the loss of high-intensity structures in long-term precipitation nowcasting by introducing Joint Embedding Predictive Architecture (JEPA) regularization for the first time. Methodologically, it employs spatiotemporal token encoding alongside a parallel motion source renderer. The encoder is directly supervised via masked historical JEPA and further enhanced through an auxiliary pathway to strengthen its representation of observational history, thereby enabling task-driven, precise prediction of future states. Evaluated on the SEVIR and MeteoNet datasets, the proposed approach achieves remarkable improvements of 118.6% and 35.1% in Critical Success Index (CSI) at the highest thresholds, respectively, substantially outperforming existing baselines.

Abstract

Long-term precipitation nowcasting requires modeling radar-echo evolution while preserving localized high-intensity structures. Recent radar-specific studies motivate location-aware prediction and separating echo displacement from intensity change. However existing encoders learn historical representations mainly from final forecast errors. We propose PrecipJEPA, which couples a structured forecasting path with an auxiliary path that enriches its encoder from observed radar history. In the forecasting path, an online encoder first converts the observations into spatiotemporal tokens. The Task-Driven Future-State Predictor (TFP) combines these tokens with a recent-dynamics summary and spatiotemporal queries to construct future radar states. The Parallel Motion-Source Renderer (PMSR) decodes these states into motion and source-sink fields that transform the latest observation into future frames. During joint training, the History-Masked JEPA (H-JEPA) operates on the auxiliary path to predict masked historical features from visible context, directly supervising the same online encoder from the observed sequence. Experiments on SEVIR and MeteoNet show that PrecipJEPA improves highest-threshold CSI by 118.6% and 35.1%, respectively, over the strongest baselines, while maintaining the highest mean CSI throughout the 3-hour forecast.

Problem

• Precipitation Nowcasting
• Radar Echo Prediction
• Long-term Forecasting
• Representation Learning

Innovation

• Precipitation Nowcasting
• JEPA
• Motion-Source Rendering
• Spatiotemporal Prediction
• Self-Supervised Learning