Towards high resolution data assimilation and ensemble forecasting

Abstract

Due the increase in computational power of supercomputers the grid resolution of high resolution numerical weather prediction models is now reaching the 1 km scale. As a result, mesoscale processes related to high impact weather (such as deep convection) can now explicitly be resolved by the models. Compared to synoptic scale systems (~1000 km) nonlinearity plays a much more prominent role at the mesoscale. Furthermore the relation between observations of these mesoscale processes and the model state is strongly nonlinear.

The most successful synoptical scale data assimilation method is known as incremental 4D-VAR and makes use of linearised versions of both the model dynamics and the observation operator that maps the model state to the observed values. Due to the use of these linear approximations there are concerns that the incremental 4D-VAR approach will be defeated by these nonlinearities when the resolution of the model is increased to the km scale. This thesis introduces a new method to account for nonlinearities in the estimation of the optimal state of the atmosphere.

Due to the success of ensemble prediction systems (EPSs) for the medium-range several numerical weather prediction consortia have developed, or are in the process of developing, EPSs for the short-range (SREPS). The main goal of these short-range ensemble prediction systems is to aid forecasters in situations of severe weather. In medium-range EPSs several methods are being used to create initial condition perturbations. At the European Centre for Medium-Range Weather Forecasts (ECMWF) part of the initial condition perturbations are created by taking linear combination of singular vectors. This thesis introduces a new final time norm for the singular vector calculation based on the convective available potential energy (CAPE) and describes the properties of these singular vectors