Tom Kent, University of Leeds
Using an idealised fluid model to evaluate EnKF performance in the presence of convection
Coauthors
Onno Bokhove, Steven Tobias, Gordon InverarityAbstract:
To aid understanding of and facilitate research into forecast-assimilation systems, computationally inexpensive ‘idealised’ models can be employed that embody some essential features of these systems. One such idealised fluid model has been developed which combines the nonlinearity due to advection in the rotating shallow water equations and switches for the onset of convection and precipitation; in doing so, it captures two important dynamical processes of convecting and precipitating weather systems. Here, we evaluate the performance of the ensemble Kalman filter (EnKF) and ascertain its suitability for convective-scale data assimilation (DA).
When using intermediate-complexity ‘toy’ models for atmospheric DA research, it is important to justify their relevance in the context of Numerical Weather Prediction (NWP). E.g., for meaningful ensemble-based experiments, a well-configured ensemble is key to providing an adequate estimation of forecast error. Furthermore, the observing system should be tuned to give a similar observational influence as in NWP. These properties are examined using the EnKF under different flow regimes (e.g., rotating flow, flow over topography) valid at the convective scale. Interesting cases arise when the strong nonlinearity associated with the thresholds for convection and precipitation leads to a bi-modal ensemble distribution, i.e., regions where some forecasts exhibit convection/precipitation while others do not.
Future work will look into ensemble-based DA algorithms developed for strongly nonlinear systems (e.g., the iterative EnKF), and is expected to further expose the limitations of the traditional EnKF in the presence of highly nonlinear convection and precipitation processes.