Abstract/Summary

Data assimilation estimates the initial conditions of a weather forecast model by bringing together data from observations, and a forecast from a previously known atmospheric state. The forecast error covariance matrix is part of the assimilation and is very important in the way that the assimilation treats the data. This article shows how these error covariances for large-scale weather systems are represented using balance relationships. An example of how this method can be improved at large-scale is introduced, and contemporary issues are raised concerning how it can be adapted to model error covariances of small-scale phenomena, such as convection, where the balance approach breaks down.