Generate the Forest before the Trees -- A Hierarchical Diffusion model for Climate Downscaling

2025-06-24

AI Summary

Traditional climate downscaling methods incur high computational costs; while recent AI models—such as diffusion models—enable ensemble generation and mitigate over-smoothing, they remain computationally intensive. This paper proposes the Hierarchical Diffusion Downscaling (HDD) model, which introduces a coarse-to-fine hierarchical sampling mechanism and a multi-scale downsampling architecture to substantially reduce per-pixel computation. HDD employs hierarchical denoising training and multi-resolution inputs, preserving 0.25° spatial fidelity while reducing pixel-level computational load by over 50%. Crucially, HDD achieves zero-shot transfer across CMIP6 models—the first such demonstration—enabling seamless multi-model adaptation and scalable ensemble forecasting. Evaluated on ERA5 and CMIP6 data, HDD delivers high fidelity, computational efficiency, and strong generalization, offering a scalable generative solution for local climate planning and impact assessment.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Machine Learning: Deep Neural Architectures and Foundation Models
• Humans and AI: Other Foundations of Human Computation & AI

Application Category

• Economics, Online Markets and Human Computation: Economic ramifications for generative AI infrastructure and applications
• Search and Retrieval-Augmented AI: Efficiency and scalability of Web search engines
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

Downscaling is essential for generating the high-resolution climate data needed for local planning, but traditional methods remain computationally demanding. Recent years have seen impressive results from AI downscaling models, particularly diffusion models, which have attracted attention due to their ability to generate ensembles and overcome the smoothing problem common in other AI methods. However, these models typically remain computationally intensive. We introduce a Hierarchical Diffusion Downscaling (HDD) model, which introduces an easily-extensible hierarchical sampling process to the diffusion framework. A coarse-to-fine hierarchy is imposed via a simple downsampling scheme. HDD achieves competitive accuracy on ERA5 reanalysis datasets and CMIP6 models, significantly reducing computational load by running on up to half as many pixels with competitive results. Additionally, a single model trained at 0.25° resolution transfers seamlessly across multiple CMIP6 models with much coarser resolution. HDD thus offers a lightweight alternative for probabilistic climate downscaling, facilitating affordable large-ensemble high-resolution climate projections. See a full code implementation at: https://github.com/HDD-Hierarchical-Diffusion-Downscaling/HDD-Hierarchical-Diffusion-Downscaling.

Problem

• Reducing computational load in climate downscaling models
• Overcoming smoothing issues in AI-based downscaling methods
• Enabling high-resolution climate data for local planning

Innovation

• Hierarchical Diffusion Downscaling (HDD) model
• Coarse-to-fine hierarchical sampling process
• Lightweight probabilistic climate downscaling