Parameter estimation for boundary-layer turbulence parameterizations over heterogeneous terrain

Accurate representation of turbulent exchange in the mountain boundary layer is particularly challenging for numerical weather prediction models. The use of common planetary boundary layer schemes, which invariably assume flat and homogeneous terrain, results in significant model errors and can even decrease the forecast skill over hills and mountains.

We seek to improve the accuracy of planetary boundary layer parameterization schemes over heterogeneous terrain using ensemble-based parameter estimation. Parameter estimation within a data assimilation framework offers a way to reduce model errors by constraining model parameters with atmospheric observations. We consider an idealized modelling environment structured as Observing System Simulation Experiments, consisting of a large-eddy simulation, providing a simulated truth, and a single column model ensemble, where the only model error source is the planetary boundary layer parameterization. This way, we eliminate additional error sources, such as initial-condition error, which could have a detrimental impact on parameter estimation. We attempt to estimate parameters in planetary boundary layer schemes affecting vertical turbulent mixing by assimilating appropriate synthetic surface observations and vertical profiles from the large-eddy simulation run. We demonstrate that, by appropriately configuring the data assimilation system, parameter estimation drives the estimated parameters to converge toward optimal values, and at the same time reduces systematic errors in atmospheric state simulations. This approach holds promise for improving the accuracy of numerical weather prediction models, especially over heterogeneous terrain. It will also make it possible to assimilate many meteorological observations made over mountains, which are often rejected by operational assimilation systems due to large discrepancies of the observed and modelled climatology.