Downscaling Precipitation with Bias-informed Conditional Diffusion Model

TL;DR

Coarse Global Climate Models fail to capture local precipitation patterns, hindering impact assessments. The authors introduce a bias-informed conditional diffusion framework for statistical downscaling, incorporating gamma correction to address long-tail precipitation and Bias-aware Guided Sampling (BGS) to reduce residual biases, conditioned on low-resolution inputs and topography. On PRISM data, the method achieves accurate 8× downscaling and outperforms deterministic baselines such as SRCNN, with improvements amplified by including topography and BGS. This approach provides a scalable, data-driven means to generate high-resolution, bias-corrected precipitation fields for localized climate analysis, with code and data releases facilitating adoption.

Abstract

Climate change is intensifying rainfall extremes, making high-resolution precipitation projections crucial for society to better prepare for impacts such as flooding. However, current Global Climate Models (GCMs) operate at spatial resolutions too coarse for localized analyses. To address this limitation, deep learning-based statistical downscaling methods offer promising solutions, providing high-resolution precipitation projections with a moderate computational cost. In this work, we introduce a bias-informed conditional diffusion model for statistical downscaling of precipitation. Specifically, our model leverages a conditional diffusion approach to learn distribution priors from large-scale, high-resolution precipitation datasets. The long-tail distribution of precipitation poses a unique challenge for training diffusion models; to address this, we apply gamma correction during preprocessing. Additionally, to correct biases in the downscaled results, we employ a guided-sampling strategy to enhance bias correction. Our experiments demonstrate that the proposed model achieves highly accurate results in an 8 times downscaling setting, outperforming previous deterministic methods. The code and dataset are available at https://github.com/RoseLV/research_super-resolution

Figures (2)

• Figure 1: Overall pipeline of the proposed framework: In the training phase, a noise-corrupted high-resolution image, a low-resolution image, and a topography image are concatenated and input into a U-Net model. The U-Net is trained to predict the noise in the corrupted high-resolution precipitation image. In the prediction phase, the trained U-Net iteratively denoises precipitation images, transforming pure noise into high-resolution precipitation. Our bias-informed sampling strategy quantifies and reduces the bias between the corrupted high-resolution precipitation and the low-resolution input at each denoising step.
• Figure 2: One example of $16 \times 16 \rightarrow 128 \times 128$ downscaling of precipitation.