Uncertainty Quantification in Forecasting using Generative Adversarial Networks

We will explore generative adversarial network models(GANs) to capture spatial dependencies in forecasting uncertainty, e.g. when predicting the next frame in a video or the weather tomorrow.

Capturing the spatial dependencies in forecasting uncertainty is important for many applications. For example, to predict the risk of flooding, it is important to take spatial dependencies in prediction uncertainty into account. If the dependencies are not taken into account, the risk will usually be underestimated, and can have dramatic consequences. In this project we will explore GANs to learn the uncertainty. We will compare the performance with established statistical methods such as the multivariate normal distribution with covariance functions.


The goal of the project is to evaluate the potential of using GANs to quantify uncertainty on forecasting.

Learning outcome

interdisciplinary research
Uncertainty quantification
Machine learning / generative adversarial networks


Hard working and motivated. Interested in learning (the rest can be learned during the thesis work)


  • Michael Riegler
  • Hugo Hammer

Collaboration partners

  • John Bjørnar Bremnes, Norwegian Meteorological Institute
  • Siri Sofie Eide, Norwegian Meteorological Institute

Contact person