https://doi.org/10.1029/2025JH000678

Abdelmoaty, Hebatallah M.; Papalexiou, Simon Michael; Mamalakis, Antonios; Singh, Shivam; Coia, Vincenzo; Hairabedian, Melissa; Szeftel, Pascal; Grover, Patrick (2025) Generative Adversarial Networks for Downscaling Hourly Precipitation in the Canadian Prairies, Journal of Geophysical Research: Machine Learning and Computation, https://doi.org/10.1029/2025JH000678

Developing robust downscaling methods is essential for maximizing the applicability of climate model outputs in engineering design and climate mitigation, particularly in a changing climate. This study evaluates four deep learning model configurations for downscaling, focusing on their structure, functionality, and ability to capture localized convective events in the Canadian prairies. These model configurations aim to downscale coarse-resolution climate model outputs (?200 km) to the finer spatial resolution of regional climate models (?50 km) for hourly precipitation. We introduce advanced metrics to assess the fidelity of precipitation downscaling, examining both marginal statistics and spatiotemporal dependencies. A U-Network (UNET) captures spatial and temporal dependencies efficiently while three generative adversarial networks (GANs) configurations incorporate a critic network to enhance the realism of generated fields. The study also evaluates the effects of a thresholding layer to constrain precipitation values and a convolution long short-term memory layer in the GAN critic to better capture temporal dependencies. Results show that all four model configurations effectively capture spatial dependencies, with the simplest GAN architecture outperforming others in preserving temporal dependencies. Latitudinal correlations are better preserved than longitudinal across all models. While UNET produces overly smoothed fields, GANs generate more detailed outputs when downscaling Coupled Model Intercomparison Project phase 6 projections. By optimizing deep learning models for this region, the study provides key insights into future precipitation trends, enabling the identification of localized storms. These findings are critical for improving infrastructure resilience across catchments in the prairies.

This study compares four deep learning models for downscaling future hourly precipitation data in the Canadian prairies. We focus on studying how well these deep learning models transform coarse climate projections (?200 km) into finer, more detailed fields (?50 km) for hourly precipitation fields. These models include a U-Network (UNET) and three generative adversarial network (GAN) configurations, assessed based on their ability to reproduce spatial patterns, temporal sequences, and statistical accuracy. While UNET captures these aspects satisfactorily, GANs produce more localized and detailed downscaled fields. The simplest GAN configuration preserves temporal patterns well. GANs also succeeded in downscaling Coupled Model Intercomparison Project phase 6 projections, confirming an increasing trend of hourly precipitation under the effect of climate change. The results show that deep learning can help downscale local storms and improve infrastructure planning.

https://doi.org/10.1029/2025JH000678