Longwang: Zero-Shot Global Spatiotemporal Precipitation Downscaling with a Latent Generative Prior
This work addresses the need for high-resolution precipitation data in climate impact assessment by enabling zero-shot downscaling without paired data, generalizing across climate simulations and projections.
Longwang introduces a zero-shot latent generative framework for global spatiotemporal precipitation downscaling, generating daily 10 km resolution fields from monthly 100 km inputs. It outperforms standard posterior sampling in reconstructing fine-scale patterns, preserving temporal coherence, and recovering extreme intensities on ERA5 reanalysis.
High-resolution precipitation information is essential for climate impact assessment, yet global climate models remain too coarse to resolve key small-scale processes. Existing machine learning downscaling methods often require paired low- and high-resolution data for supervised learning, are tied to fixed regions or scale factors during inference, and can be computationally expensive to train and run in physical space. Here we introduce Longwang, a zero-shot latent generative framework for global spatiotemporal precipitation downscaling. Longwang learns a context-conditioned latent generative prior and combines it with a physically informed observation operator through posterior sampling, enabling daily O(10 km) precipitation fields to be generated from monthly O(100 km) inputs. On ERA5 reanalysis, Longwang outperforms standard posterior sampling with an unconditional generative prior in reconstructing fine-scale spatial patterns, preserving temporal coherence, and recovering extreme precipitation intensities. The framework further generalizes to historical climate simulations and future climate projections under substantial distribution shift.