R package eppverification 

This R package contains verification tools for the statistical postprocesing of ensemble forecasts. You can find the latest development version here on GitHub.

Local Spatial Bayesian Model Averaging for the Postprocessing of Ensemble Forecasts 

Description: 
Current practise in predicting future weather is the use of numerical weather prediction (NWP) models. These are deterministic numerical simulation models representing the physics of the atmosphere. The models produce so called ensemble forecasts, which often exhibit biases and dispersion errors. Therefore these ensemble forecasts need to be statistically postprocessed, what we call ensemble postprocessing (EPP).

A common postprocessing procedure for an univariate weather variable (e.g. temperature) is the following:
1. Interpolation: Interpolate the ensemble forecasts given at grid points to the observed location, for which one would like to make predictions. This is usually done by the geostatistical method kriging. 
2. Fitting: Employ an univariate postprocessing method like e.g. ensemble model output statistics (EMOS) or Bayesian model averaging (BMA) to the ensemble forecasts based on training data. This yields a fitted predictive distribution. 
3. Prediction: Predict for the observed location with the estimated distribution based on a verification dataset.

In our new approach, which is called local spatial Bayesian Model Averaging (SBMA), ensemble forecasts at grid points are directly employed to obtain predictive distributions at stations. Several fitting methods are proposed and will be compared by using training and verification data.

Members: 
This project is part of my PhD and is a collaboration with Jürgen Groß and Annette Möller of the DFG project "Statistical postprocessing of ensemble forecasts for various weather quantities". 

Results:
t.b.a.