Stochastic modelling of natural processes
Contact person: Thordis Thorarinsdottir
Keywords: machine learning, statistical modelling, extremes, climate risk
Research group: Statistics and Data Science, Hydrology and Water Resources
Department of Mathematics, Department of Geosciences
We are interested in methodological research in statistics and machine learning for modelling of randomness and quantification of uncertainty in natural processes. The modelling may have a particular emphasis on assessing physical climate risks due (increasing frequency/occurrence of) acute weather events, such as flooding, wildfires, extreme heat, storm surges and droughts. Some of these events are widespread or caused by many different factors, calling for modelling frameworks that apply to multivariate and compound extremes. Alternatively, the work may have a generative focus, where the aim is to build stochastic generative models that are able to accurately account for extreme events. The modelling commonly calls for knowledge-informed statistical and machine learning models that combine statistical theory and environmental process understanding to overcome data deficiencies.
Methodological research topics:
- Generative modelling
- Probabilistic modelling
- Distributional loss functions
- Model evaluation
- Multivariate and compound extremes
External partners:
- Norwegian Computing Center (NR)
- Norwegian Meteorological Institute (met.no)
Mentoring and internship will be offered by a relevant external partner
