Wasserstein GAN-Based Precipitation Downscaling with Optimal Transport for Enhancing Perceptual Realism

2025-07-23

AI Summary

High-resolution precipitation forecasting is critical for mitigating localized heavy-rainfall hazards, yet conventional numerical weather prediction models face fundamental limitations in physical representation and computational accuracy at fine scales. To address this, we propose a precipitation downscaling method grounded in Wasserstein Generative Adversarial Networks (WGAN) and optimal transport theory. Our approach leverages the discriminator’s output as an interpretable, perception-aligned fidelity metric—overcoming the well-known misalignment between traditional statistical metrics (e.g., PSNR, SSIM) and human visual assessment. By incorporating an optimal transport-based cost function into adversarial training, the method preserves physical consistency while substantially enhancing fine-scale structural detail and visual realism of generated precipitation fields. Experiments demonstrate that although conventional metrics are marginally lower, the discriminator score exhibits strong agreement with expert evaluation, effectively detecting anomalies and data artifacts. This establishes a novel paradigm for quality control of high-resolution precipitation products.

Technology Category

• Computer Vision: Generative Adversarial Networks (GANs) for Vision
• Natural Language Processing: Generation
• Machine Learning: Calibration & Uncertainty Quantification

Application Category

• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics
• Web Mining and Content Analysis: Web data generation and simulation
• User Modeling, Personalization and Recommendation: Metrics for user behavior and evaluating success

Abstract

High-resolution (HR) precipitation prediction is essential for reducing damage from stationary and localized heavy rainfall; however, HR precipitation forecasts using process-driven numerical weather prediction models remains challenging. This study proposes using Wasserstein Generative Adversarial Network (WGAN) to perform precipitation downscaling with an optimal transport cost. In contrast to a conventional neural network trained with mean squared error, the WGAN generated visually realistic precipitation fields with fine-scale structures even though the WGAN exhibited slightly lower performance on conventional evaluation metrics. The learned critic of WGAN correlated well with human perceptual realism. Case-based analysis revealed that large discrepancies in critic scores can help identify both unrealistic WGAN outputs and potential artifacts in the reference data. These findings suggest that the WGAN framework not only improves perceptual realism in precipitation downscaling but also offers a new perspective for evaluating and quality-controlling precipitation datasets.

Problem

• High-resolution precipitation prediction is challenging
• WGAN improves perceptual realism in downscaling
• Optimal transport enhances fine-scale precipitation structures

Innovation

• WGAN for precipitation downscaling
• Optimal transport enhances realism
• Critic scores evaluate perceptual quality