 |
The introduction to this article provides insufficient context for those unfamiliar with the subject. Please help improve the article with a good introductory style. |
In Numerical Weather Prediction (NWP) systems, data assimilation is the process of objectively adapting the model state to observations in a statistically optimal way taking into account model and observation errors. This is an essential step before a forecast run can be started. An example of how this is implemented at ECMWF can be found here: http://www.ecmwf.int/research/ifsdocs/ASSIMILATION/Chap1_Overview2.html The most common types of assimilation in NWP are:
- 3D var
- 4D var
- Kalman filter
- Optimal Interpolation
Most methods are implemented using some cost function, so adding a penalty to observations that are far away from the value predicted by the model. This cost function is scaled by the estimated observation error. By summarising all cost functions for all observations, and then varying the model state to find a minimum in this function, a model state is found that fits best to all observations. This is then used as starting point for calculating forecasts.
