Dissertation · 2025 · Open access

Downscaling of general circulation models using generative AI

Svensson, Markus

DiVA (Luleå University of Technology) · Published 1 January 2025

Open full textCite this paper

In brief

“Downscaling of general circulation models using generative AI” is a 2025 dissertation by Svensson, Markus, published in DiVA (Luleå University of Technology). OpenAlex counts 0 citations as of 21 November 2026. It is open access: the full text is free to read at DiVA (Luleå University of Technology).

Citations
0
Authors
1

Abstract

Global Climate Models (GCMs) are an essential tool for projecting climate change impacts but theircoarse spatial resolution (often 100-200 km) limits their applicability for detailed climate assessments ona regional scale. This thesis develops a downscaling framework which combine a novel conditionalGenerative Adversarial Network (cGAN) with Quantile Delta Mapping (QDM) bias correction to increasespatial resolution by a factor of 10. The cGAN architecture incorporates Residual-in-Residual DenseBlocks and both channel and spatial attention mechanisms, where daily precipitation is the primaryvariable, and temperature and sea level pressure as conditional inputs. Training is conducted on ERA5-Land reanalysis data which have been upscaled to match the resolution of the EC-Earth3 GCM, enablingthe cGAN to learn regional downscaling patterns prior to application on the bias-corrected GCM output.The proposed method has been benchmarked on simple traditional downscaling methods, includingbilinear and nearest-neighbour interpolation. Results indicate that the cGAN significantly outperformstraditional methods, where the Power Spectral Density shows improved performance across alltimescales and spatial maps show both long- and short-term superiority. The cGAN was further testedon Pearson correlation and Kling-Gupta Efficiency, showing values above 0.97 and 0.9 for all cellsrespectively. The QDM bias correction combined with a dry-day adjustment effectively mitigates thecommon drizzling bias in GCM precipitation, showing that the method yields similar number of dry days.However, some limitations remain. Some seasonal biases persist, where there are someunderestimations of rainfall for the more rainfall intense months (July to September), and QQ-plot analysisreveals overestimation of moderate rainfall intensities and underestimation of extremes in certainlocations. The findings highlight the sensitivity to the bias correction stage, where the bias propagates tothe finer resolution.Overall, this work demonstrates that integrating generative AI with statistical bias correction offers acomputationally efficient and operationally viable approach for producing high-resolution, bias-reducedprecipitation datasets from coarse-resolution GCM outputs.

Cite this paper

Formatted from the catalogue record. Check names, capitals and page numbers against the paper before you submit.

APA 7

Markus, S. (2025). Downscaling of general circulation models using generative AI. DiVA (Luleå University of Technology). http://urn.kb.se/resolve?urn=urn:nbn:se:ltu:diva-115015

MLA 9

Markus, Svensson,. “Downscaling of general circulation models using generative AI.” DiVA (Luleå University of Technology), 2025. http://urn.kb.se/resolve?urn=urn:nbn:se:ltu:diva-115015.

Chicago

Markus, Svensson,. 2025. “Downscaling of general circulation models using generative AI.” DiVA (Luleå University of Technology). http://urn.kb.se/resolve?urn=urn:nbn:se:ltu:diva-115015.

Harvard

Markus, S. (2025) ‘Downscaling of general circulation models using generative AI’, DiVA (Luleå University of Technology). Available at: http://urn.kb.se/resolve?urn=urn:nbn:se:ltu:diva-115015.

IEEE

S. Markus, “Downscaling of general circulation models using generative AI,” DiVA (Luleå University of Technology), 2025.

BibTeX

@phdthesis{markus2025downscaling,
  title     = {{Downscaling of general circulation models using generative AI}},
  author    = {Markus, Svensson,},
  year      = {2025},
  url       = {http://urn.kb.se/resolve?urn=urn:nbn:se:ltu:diva-115015}
}

Download for a reference manager: BibTeX (.bib) · RIS for Zotero, Mendeley and EndNote (.ris)

Topic and keywords

Physical Sciences › Environmental Science › Global and Planetary Change › Climate variability and models

  • climate downscaling
  • conditional generative adversarial networks
  • quantile delta mapping
  • bias correction
  • precipitation
  • high-resolution climate data
  • dense blocks
  • attention mechanism

Work with this paper

Most-cited papers in Environmental Science

All highly cited Environmental Science papers

Questions about this paper

How do I cite this paper in APA style?

Markus, S. (2025). Downscaling of general circulation models using generative AI. DiVA (Luleå University of Technology). http://urn.kb.se/resolve?urn=urn:nbn:se:ltu:diva-115015

How many times has this paper been cited?

0 times, according to OpenAlex on 21 November 2026. Google Scholar and Scopus count different sets of sources, so their numbers differ.

Is this paper free to read?

Yes. It is open access at DiVA (Luleå University of Technology): sign in and open the full text from this page.

Who wrote this paper?

Svensson, Markus.

Bibliographic data from OpenAlex (CC0), last checked 21 November 2026. ResearchGuru does not host this paper; once you are signed in, the links open it at the publisher or the repository that does.