Addressing model error related to surface fluxes by estimating roughness lengths in COSMO-KENDA with the augmented state approach
Convection permitting models are highly prone to model error due to the complexity of important subgridscale processes. If model error is not well represented in the ensemble, the corresponding error covariance matrix overestimates the accuracy of the prior ensemble and underestimates the true covariances. As a result, observations are not used optimally which can lead to filter divergence. Covariance inflation techniques are helpful, but they are not designed to target specific model errors, leading to suboptimal results.
In this study we aim to represent and reduce model error by perturbing and estimating appropriate model parameters in an augmented state data assimilation framework. This is done using the operational convection-permitting COSMO model and data assimilation system KENDA employed atthe German weather service. We focus on the estimation of the horizontally varying roughness length, which directly influences the computation of surfaces fluxes. By projecting all uncertainties related to surface fluxes onto the estimated roughness length parameter, we hope to represent and reduce the associated model error and thereby improve the representation of clouds.
The augmented state is fed to data assimilation system KENDA, which employs the LETKF to produce initial conditions. The presented experiments are conducted for the two week period starting from 28.05.016, which includes a weak forcing and strong forcing period. We discuss the effects of different perturbation structures for the roughness length and compare the results to appropriate reference experiments as well as to Meteosat Second Generation visible (VIS) and near-infrared (NIR) satellite reflectance observations.
Ms. Yvonne Muriel Ruckstuhl Ludwig Maximilian University of Munich (LMU)