ClimateAR: Multi-Scale Autoregressive Generative Modeling for Climate Forecasting

Accurate climate forecasting provides critical support for decision-making in agriculture, energy, and disaster preparedness. Current deterministic models often fail to capture climate uncertainty, while existing generative approaches oversimplify the system by neglecting key spatiotemporal dependencies and cross-scale interactions. To address these limitations, we introduce ClimateAR, an AutoRegressive generative model for probabilistic seasonal-to-interannual Climate forecasting. The framework incorporates two novel components: (1) an aligned tokenizer that bridges and aligns heterogeneous simulation and real-world data to improve transferability across domains, and (2) a mixed-scale conditioning mechanism that captures multi-scale climate interactions for robust probabilistic forecasting. Extensive evaluations on the ERA5 reanalysis dataset show that ClimateAR achieves state-of-the-art performance, improving anomaly correlation skill by 37.56\% on average compared to leading baselines.

Accurate climate forecasting provides critical support for decision-making in agriculture, energy, and disaster preparedness. However, the climate system itself is highly uncertain, and traditional methods often provide a precise prediction. To address this issue, we propose ClimateAR, which generates multiple possible future scenarios nstead of directly predicting a specific value. Furthermore, we specifically consider the relationship between global climate phenomena and changes in local variables (such as temperature) within the climate system and design a corresponding method (technically called multi-scale autoregression). We also incorporate a large amount of simulation data to train the model and enhance its capabilities. Experiments show that it significantly outperforms existing methods on multiple climate prediction tasks.