Efficient Kilometer-Scale Precipitation Downscaling with Conditional Wavelet Diffusion

2025-07-02

AI Summary

Existing global precipitation products (e.g., IMERG) offer only ~10 km spatial resolution, insufficient for kilometer-scale hydrological modeling and extreme weather analysis. To address this, we propose a wavelet-domain conditional diffusion generative model that super-resolves low-resolution precipitation fields to 1 km. Crucially, our method models high-frequency wavelet coefficients across multiple scales, explicitly capturing fine-scale precipitation structures and effectively suppressing generation artifacts. Unlike pixel-space diffusion models, ours achieves a 10× spatial super-resolution factor while preserving physical consistency and perceptual realism in the generated precipitation fields. Moreover, it accelerates inference by 9×. This work establishes a new paradigm for high-fidelity, computationally efficient precipitation downscaling in Earth system science.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Machine Learning: Deep Generative Models & Autoencoders
• Natural Language Processing: Generation

Application Category

• Search and Retrieval-Augmented AI: Efficiency and scalability of Web search engines
• Web Mining and Content Analysis: Large pretrained models with web data
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

Effective hydrological modeling and extreme weather analysis demand precipitation data at a kilometer-scale resolution, which is significantly finer than the 10 km scale offered by standard global products like IMERG. To address this, we propose the Wavelet Diffusion Model (WDM), a generative framework that achieves 10x spatial super-resolution (downscaling to 1 km) and delivers a 9x inference speedup over pixel-based diffusion models. WDM is a conditional diffusion model that learns the learns the complex structure of precipitation from MRMS radar data directly in the wavelet domain. By focusing on high-frequency wavelet coefficients, it generates exceptionally realistic and detailed 1-km precipitation fields. This wavelet-based approach produces visually superior results with fewer artifacts than pixel-space models, and delivers a significant gains in sampling efficiency. Our results demonstrate that WDM provides a robust solution to the dual challenges of accuracy and speed in geoscience super-resolution, paving the way for more reliable hydrological forecasts.

Problem

• Achieve kilometer-scale precipitation downscaling from 10 km to 1 km
• Improve speed and accuracy of precipitation super-resolution models
• Generate realistic high-resolution precipitation data for hydrological modeling

Innovation

• Wavelet Diffusion Model for 10x super-resolution
• Conditional diffusion in wavelet domain
• Focus on high-frequency wavelet coefficients