Abstract

Data assimilation techniques combine observations and prior model forecasts to create initial conditions for numerical weather prediction (NWP). The relative weighting assigned to each observation in the analysis is determined by the error associated with its measurement. Remote sensing data often have correlated errors, but the correlations are typically ignored in NWP. As operational centres move towards high-resolution forecasting, the assumption of uncorrelated errors becomes impractical. This thesis provides new evidence that including observation error correlations in data assimilation schemes is both feasible and beneficial. We study the dual problem of quantifying and modelling observation error correlation structure. Firstly, in original work using statistics from the Met Office 4D-Var assimilation system, we diagnose strong cross-channel error covariances for the IASI satellite instrument. We then see how in a 3D-Var framework, information content is degraded under the assumption of uncorrelated errors, while retention of an approximate correlation gives clear benefits. These novel results motivate further study. We conclude by modelling observation error correlation structure in the framework of a one-dimensional shallow water model. Using an incremental 4D-Var assimilation system we observe that analysis errors are smallest when correlated error covariance matrix approximations are used over diagonal approximations. The new results reinforce earlier conclusions on the benefits of including some error correlation structure. i Declaration I confirm that this is my own work and the use of all material from other sources has been properly and fully acknowledged.

Citations