DFG Project Ensemble Postprocessing
Network members
- Dr. Sándor Baran, University of Debrecen, Hungary
- Dr. Patricia Szokol, University of Debrecen, Hungary
- PD Dr. Jürgen Groß, University of Hildesheim, Germany
- Dr. Roman Schefzik, German Cancer Research Center (DKFZ), Heidelberg, Germany
- Dr. Sebastian Lerch, Karlsruhe Institute of Technology and Heidelberg Institute for Theoretical Studies, Heidelberg, Germany
- Dr. Stephan Hemri, Federal Office of Meteorology and Climatology MeteoSwiss, Switzerland
Network Project
Numerical weather prediction (NWP) models are used to predict the weather. They consist of a system of differential equations that describe the state of the atmosphere as accurately as possible and, by integration over time, provide progons about the state of the atmosphere at time points in the future. Typically, NWP models are run multiple times, each run with different initial conditions and/or different model formulations, to thereby reflect the uncertainties in the initial conditions/model formulations. The result is a prediction ensemble, from each run of the model a deterministic prediction emerges.
Since ensemble predictions are usually poorly calibrated, statistical models are developed to improve the properties of the predictions and quantify the prediction uncertainty.
The goal of the research collaboration in the network is to develop statistical models for postprocessing the ensemble predictions (ensemble postprocessing).
The current plans of the network are
- Further development of existing models for normally distributed weather variables such as temperature, so that they can be applied to other weather variables, e.g. skewed distributions such as wind speed, or mixed distributions such as precipitation.
- Extension of existing models to the multivariate case, i.e. consideration of spatial and temporal dependencies or dependencies between weather variables.
- Application of machine learning methods in the context of ensemble postprocessing, especially for discrete weather variables like cloud cover.
- Implementation of new models, improvement of existing implementations.
Previous publications by network members on the project topic