Downscaling Extreme Precipitation with Wasserstein Regularized Diffusion

2024-10-01

AI Summary

Global coarse-resolution (30–55 km) precipitation datasets are inadequate for high-fidelity flood risk assessment and climate adaptation planning. To address this, we propose the first generative downscaling method based on a Wasserstein-regularized diffusion model, elevating CPC/ERA5 precipitation data to 1 km resolution. Our approach enforces Wasserstein distance constraints throughout the entire denoising trajectory, ensuring alignment between generated and observed extreme-value distributions and substantially mitigating systematic biases. Furthermore, it incorporates multi-source reanalysis and observational data as physically informed conditioning inputs, guaranteeing dynamical and thermodynamical consistency in the outputs. Experiments demonstrate that our method outperforms state-of-the-art approaches across three critical dimensions: reconstruction accuracy, fidelity of extreme precipitation events, and realism of spatial structures. Notably, it accurately reproduces millimeter-scale peak intensities and fine-scale rainband structures associated with tropical cyclones and cold fronts.

Technology Category

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

Application Category

• Web Mining and Content Analysis: Web data generation and simulation
• Economics, Online Markets and Human Computation: Economic ramifications for generative AI infrastructure and applications
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking

Abstract

Understanding the risks posed by extreme rainfall events necessitates both high-resolution products (to assess localized hazards) and extensive historical records (to capture rare occurrences). Radar and mesonet networks provide kilometer-scale precipitation fields, but with limited historical records and geographical coverage. Conversely, global gauge and blended products span decades, yet their coarse 30-50 km grids obscure local extremes. This work introduces Wasserstein Regularized Diffusion (WassDiff), a generative downscaling framework that integrates diffusion modeling with a distribution-matching (Wasserstein) regularizer, suppressing bias throughout the entire generative denoising process. Conditioned on 55 km CPC gauge-based precipitation and the 31 km ERA5 reanalysis, WassDiff generates 1 km precipitation estimates that remain well-calibrated to targets across the full intensity range, including the extremes. Comprehensive evaluations demonstrate that WassDiff outperforms existing state-of-the-art downscaling methods, delivering lower reconstruction error and reduced bias. Case studies further demonstrate its ability to reproduce realistic fine-scale structures and accurate peak intensities from extreme weather phenomena, such as tropical storms and cold fronts. By unlocking decades of high-resolution rainfall information from globally available coarse records, WassDiff offers a practical pathway toward more accurate flood-risk assessments and climate-adaptation planning.

Problem

• Downscaling coarse precipitation data to high-resolution for accurate flood-risk assessment
• Generating realistic fine-scale rainfall structures from extreme weather events
• Reducing bias in precipitation estimates using Wasserstein Regularized Diffusion

Innovation

• Wasserstein Regularized Diffusion for downscaling
• Generates 1 km precipitation from coarse data
• Improves accuracy in extreme rainfall modeling