Downscaling climate projections to 1 km with single-image super resolution

2025-09-24

AI Summary

Low spatial resolution (e.g., 12.5 km) of current climate prediction datasets impedes local-scale decision-making. To address this, we propose a single-image super-resolution statistical downscaling method tailored to climate variables: a deep learning model is trained on high-resolution gridded observational data and then transferred to enhance low-resolution climate forecast fields to 1-km resolution. To overcome the challenge of evaluating downscaled outputs in the absence of high-resolution ground truth, we introduce a novel evaluation framework driven by in-situ meteorological station observations, ensuring consistency between downscaled results and original inputs in key climate statistics. Experiments on daily mean temperature demonstrate substantial improvement in spatial detail without introducing bias in climate metrics, thereby balancing physical plausibility and practical applicability.

Technology Category

• Machine Learning: Semi-Supervised Learning
• Computer Vision: Representation Learning for Vision
• Reasoning under Uncertainty: Graphical Models

Application Category

• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics
• User Modeling, Personalization and Recommendation: Metrics for user behavior and evaluating success
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

High-resolution climate projections are essential for local decision-making. However, available climate projections have low spatial resolution (e.g. 12.5 km), which limits their usability. We address this limitation by leveraging single-image super-resolution models to statistically downscale climate projections to 1-km resolution. Since high-resolution climate projections are unavailable for training, we train models on a high-resolution observational gridded data set and apply them to low-resolution climate projections. We propose a climate indicator-based assessment using observed climate indices computed at weather station locations to evaluate the downscaled climate projections without ground-truth high-resolution climate projections. Experiments on daily mean temperature demonstrate that single-image super-resolution models can downscale climate projections without increasing the error of climate indicators compared to low-resolution climate projections.

Problem

• Downscaling climate projections to 1 km resolution
• Training super-resolution models without high-resolution climate data
• Evaluating downscaled projections using climate indicator assessment

Innovation

• Single-image super-resolution models downscale climate projections
• Training uses high-resolution observational gridded dataset
• Climate indicator-based assessment evaluates without ground-truth