High-latitude auroral and cloudiness occurrence from automatic image classification

Partamies, Noora; Syrjäsuo, Mikko

doi:10.5194/angeo-44-1003-2026

Articles | Volume 44, issue 2

https://doi.org/10.5194/angeo-44-1003-2026

© Author(s) 2026. This work is distributed under
the Creative Commons Attribution 4.0 License.

https://doi.org/10.5194/angeo-44-1003-2026

© Author(s) 2026. This work is distributed under
the Creative Commons Attribution 4.0 License.

Articles | Volume 44, issue 2

Regular paper

|

01 Oct 2026

Regular paper |

| 01 Oct 2026

High-latitude auroral and cloudiness occurrence from automatic image classification

Noora Partamies and Mikko Syrjäsuo

Data sets

KHOnet2026: Full-colour auroral image classification method - training data & results Partamies, N., and Syrjäsuo, M. https://doi.org/10.5281/zenodo.19653026

AuroraX – an open data platform for aurora science E. Donovan et al. https://doi.org/10.5281/zenodo.16583708

Aurora Cloud Sensor III Data - University College London - Kjell Henriksen Observatory A. Aruliah and I. McWhirter https://doi.org/10.5281/zenodo.14931122

Model code and software

UNISvalbard/KHOnet2026: v1.0 (Version v1.0) M. Syrjäsuo https://doi.org/10.5281/zenodo.23041597

Short summary

We developed a method to prune colour all-sky images into classes of clear or cloudy skies with and without aurora using supervised learning and pre-trained convolutional neural network. We investigate a 10-year database of auroral images taken from Svalbard. The method accuracy is well over 90 %, and the results show that about 2/3 of auroral images are cloudy with the cloudiest month being November. Aurora are most often observed in the morning hours independent on the solar activity.