Sensitivity of an orographic drag parametrisation scheme to empirical parameters revealed by ensemble simulations

Mountains affect the general circulation of the atmosphere on multiple spatial scales, some of which are too small to be explicitly resolved by weather and climate models. To represent the drag exerted by unresolved mountains and unresolved gravity waves, orographic drag parametrisations use statistics of unresolved orography to calculate the sub-grid-scale (SGS) drag. Processes such as large scale wave breaking and flow-blocking become resolved at today's model resolutions, but small-scale drag and turbulent orographic form drag remain in the SGS regime. This poses the open problem of a correct partitioning between resolved and unresolved orographic drag.

Recent studies have used ensemble data assimilation methods, in particular joint state and parameter estimation, to improve the representation of SGS boundary-layer turbulence in models. Inspired by those studies, we aim to to evaluate the sensitivity of an orographic drag scheme to its empirical parameters to identify candidates for effective parameter estimation experiments. Using the Weather Research and Forecasting (WRF) model, we conducted numerical experiments of mountain waves over complex terrain with a grid spacing of 10 km using the GSL drag scheme. This parametrisation represents large-scale gravity wave drag, flow blocking drag, turbulent orographic form drag, and small-scale gravity wave drag.

Using ensemble experiments with perturbed empirical parameters, we evaluate the correlation between individual parameters and the model state. The parameters that display the highest ensemble correlation with the model state have the greatest impact on the behaviour of the orographic drag parametrisation, making them candidates for parameter estimation. Our preliminary results refer to two parameters in the scheme that affect low-level wave breaking and the separation between blocking and non-blocking states, and illustrate their ensemble correlations with the model state.