EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Multidimensional Distributional Neural Network Output Demonstrated in Super‐Resolution of Surface Wind Speed

Harrison J. Goldwyn, Mitchell Krock, Johann Rudi, Daniel Getter, Julie Bessac

Journal of Geophysical Research: Machine Learning and Computation · 2026-06 · 3 (3)

DOI: 10.1029/2025jh000964 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Journal of Geophysical Research: Machine Learning and Computation, online erschienen am 13.05.2026. Verfasst von Harrison J. Goldwyn und 4 weiteren Autor:innen. Die Arbeit nennt 47 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: Advanced Scientific Computing Research.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Accurate quantification of uncertainty in neural network predictions remains a central challenge for scientific applications involving high‐dimensional, correlated data. While existing methods capture either aleatoric or epistemic uncertainty, few offer closed‐form, multidimensional distributions that preserve spatial correlation while remaining computationally tractable. In this work, we present a framework for training neural networks with a multidimensional Gaussian loss, generating a closed‐form predictive distribution over outputs informed by non‐identically distributed training data. Our approach captures aleatoric uncertainty by iteratively estimating the means and covariance matrices, and is demonstrated on a super‐resolution example out‐of‐training‐sample. We leverage a Fourier representation of the covariance matrix to stabilize network training and preserve spatial correlation. We introduce a novel regularization strategy—referred to as information sharing—that interpolates between image‐specific and global covariance estimates, enabling convergence of the super‐resolution downscaling network trained on image‐specific distributional loss functions. This framework allows for efficient sampling, explicit correlation modeling, and extensions to more complex distribution families all without disrupting prediction performance. We demonstrate the method on a surface wind speed downscaling task and discuss its broader applicability to uncertainty‐aware prediction in scientific models.

Förderung

  • Advanced Scientific Computing Research

Zitieren

APA

Goldwyn, H. J., Krock, M., Rudi, J., Getter, D., & Bessac, J. (2026). Multidimensional Distributional Neural Network Output Demonstrated in Super‐Resolution of Surface Wind Speed. Journal of Geophysical Research: Machine Learning and Computation, 3(3). https://doi.org/10.1029/2025jh000964

Vancouver

Goldwyn HJ, Krock M, Rudi J, Getter D, Bessac J. Multidimensional Distributional Neural Network Output Demonstrated in Super‐Resolution of Surface Wind Speed. Journal of Geophysical Research: Machine Learning and Computation. 2026;3(3). doi:10.1029/2025jh000964

BibTeX
@article{Goldwyn2026,
  author = {Goldwyn, Harrison J. and Krock, Mitchell and Rudi, Johann and Getter, Daniel and Bessac, Julie},
  title = {Multidimensional Distributional Neural Network Output Demonstrated in Super‐Resolution of Surface Wind Speed},
  journal = {Journal of Geophysical Research: Machine Learning and Computation},
  year = {2026},
  volume = {3},
  number = {3},
  doi = {10.1029/2025jh000964}
}
Kostenlos starten →