Longwang: Zero-Shot Global Spatiotemporal Precipitation Downscaling with a Latent Generative Prior

2026-05-17

AI Summary

This work proposes a zero-shot latent-space generative downscaling framework to address the limitations of global climate models, which struggle to accurately represent fine-scale precipitation processes due to coarse resolution. Unlike existing machine learning approaches that rely on paired training data, fixed regions, and high computational costs, the proposed method integrates a context-conditioned latent prior with a physics-informed observation operator, enabling the first unpaired generation of global daily precipitation fields at ~10 km resolution. The framework supports inference at arbitrary spatiotemporal scales, exhibits strong generalization under distributional shifts, and substantially outperforms conventional techniques in reconstructing fine structural details, preserving temporal consistency, and recovering extreme precipitation intensities. It has been successfully applied to both historical climate simulations and future projection scenarios.

Abstract

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.

Problem

• precipitation downscaling
• zero-shot learning
• spatiotemporal resolution
• global climate models
• high-resolution precipitation

Innovation

• zero-shot
• latent generative prior
• spatiotemporal downscaling
• posterior sampling
• climate precipitation