EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Bridging Global Climate Solutions and Local Realities: Evaluating Neural Networks for High-Resolution Downscaling

Daria Taniushkina, Aleksander Lukashevich, Valeriy Shevchenko, Daniil Sukhorukov, Yulia Melkozerova, Vladislav Pedashenko, Vsevolod Morozov, Roman Sultimov, Ilya Makarov, Aleksandr Volkov, Yury Maximov

IEEE Access · 2026 · 14 · 47863-47878

DOI: 10.1109/access.2026.3673693 ↗

Kurz gefasst

Fachartikel in der Zeitschrift IEEE Access, 2026. Verfasst von Daria Taniushkina und 10 weiteren Autor:innen. Die Arbeit wurde bisher 1-mal zitiert (Crossref) und nennt 96 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: MoscowIndependent Research Institute of Artificial Intelligence, Ministry of EconomicDevelopment of the Russian Federation.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Effective climate adaptation and resilience require high-resolution localized climate data; however, widely used global climate models (GCMs) operate at coarse spatial resolutions and are unable to address community-scale needs. Even the most advanced GCMs often reduce the world’s largest cities to a single grid cell, largely ignoring their complex terrains and microclimates. This limits community-level risk assessment and adaptation planning. This paper introduces a unified, open-source framework for neural downscaling of climate data that converts coarse-resolution GCM outputs into finer-scale projections. Using standardized inputs, preprocessing, data splits, and latitude-aware evaluation metrics, we evaluate nine state-of-the-art neural architectures across a range of climate variables and geographically diverse regions using three public datasets. Empirical results show that transformer-based models (ViT, HAT) achieve the highest predictive accuracy overall, while generative models (ESRGAN and diffusion) produce realistic fine-scale structure and better describe climate in complex terrain. We release all code, datasets, and evaluation tools to support reproducible comparison and practical deployment.

Förderung

  • MoscowIndependent Research Institute of Artificial Intelligence
  • Ministry of EconomicDevelopment of the Russian Federation

Zitieren

APA

Taniushkina, D., Lukashevich, A., Shevchenko, V., Sukhorukov, D., Melkozerova, Y., Pedashenko, V., Morozov, V., Sultimov, R., Makarov, I., Volkov, A., & Maximov, Y. (2026). Bridging Global Climate Solutions and Local Realities: Evaluating Neural Networks for High-Resolution Downscaling. IEEE Access, 14, 47863-47878. https://doi.org/10.1109/access.2026.3673693

Vancouver

Taniushkina D, Lukashevich A, Shevchenko V, Sukhorukov D, Melkozerova Y, Pedashenko V, et al. Bridging Global Climate Solutions and Local Realities: Evaluating Neural Networks for High-Resolution Downscaling. IEEE Access. 2026;14:47863-47878. doi:10.1109/access.2026.3673693

BibTeX
@article{Taniushkina2026,
  author = {Taniushkina, Daria and Lukashevich, Aleksander and Shevchenko, Valeriy and Sukhorukov, Daniil and Melkozerova, Yulia and Pedashenko, Vladislav and Morozov, Vsevolod and Sultimov, Roman and Makarov, Ilya and Volkov, Aleksandr and Maximov, Yury},
  title = {Bridging Global Climate Solutions and Local Realities: Evaluating Neural Networks for High-Resolution Downscaling},
  journal = {IEEE Access},
  year = {2026},
  volume = {14},
  pages = {47863--47878},
  doi = {10.1109/access.2026.3673693}
}
Kostenlos starten →