is a technique to enhance the direct model output from numerical weather prediction (NWP) models. It corrects for systematic biases of the model and accounts for local
influences which are not (completely) resolved in the gridbox representation of the model output. Besides it is used to produce forecasts for parameters which are not calculated in the NWP model at all. Think
of the probability of precipitation, visibility, probability of lightning, etc.
Statistical postprocessing uses a statistical representative dataset (usually 3-4 seasons) of both model calculations and local observations. Model output parameters which are thought to have predictive
potential for the observed parameter are fed into a statistical package. Statistical methods such as multiple lineair regression or logistic regression are then used to calculate the "best" fit on a training dataset.
Subsequently this fit is used in forecasting mode and applied to the current model forecasts.
Another statistical method, that we use is
, which combines forecasts from different models, and leads to a complete probability distribution of e.g. the forecasted temperature or precipitation.
Our research has led to a modified version of
BMA and a paper
about this was published in Monthly Weather Review in November 2010.
Other publications can be found in the menu on the right.
- Calibration of ensemble forecasts
- Probabilistic forecasts of severe weather, e.g. thunderstorms
- Extreme precipitation forecasts for water boards
- Verification of (severe) weather forecasts and warnings
Other tasks and activities:
Expert centre in statistical postprocessing.
Incorporation of ensemble predictions in the forecasting process.
Development and implementation of new statistical postprocessing
techniques in (severe) weather forecasting.
Development and maintenance of guidance forecasts.
Verification and quality control of guidance forecasts. Reports to
internal and external partners (ECMWF), with underlying
question: "what is the added value of each step in the