Using a Stochastic Parameter Perturbation to Represent Process-Based Uncertainty in a Microphysics Parameterization

Most regional NWP model ensemble systems are under-dispersive, producing unreliable and overconfident ensemble forecasts (Buizza et al. 2005, Romine et al. 2015, Jankov et al. 2017, 2019). Besides initial and boundary condition uncertainties, other sources of error are contained within a model's physical parameterization schemes. Radiation, cumulus, boundary-layer, land-surface, and microphysics parameterizations have a plethora of internal constants and diagnosed variables known to vary widely and yet these are typically set to an average value obtained from the literature. The stochastic parameter perturbation (SPP) method in the WRF model was created by Berner et al. (2009, 2015, 2019) to perturb various parameters within physical parameterization schemes and has been used by Jankov et al. (2017, 2019) to introduce perturbations to the land-surface and planetary boundary-layer scheme and by Griffin et al. (2019) to perturb the microphysics scheme. In these studies, ensemble forecasts produced using the SPP and other perturbation methods were compared and verified against weather observations such as near-surface and upper-level variables, radar reflectivity, quantitative precipitation and satellite brightness temperatures.

In this talk, we will present the method and outcomes from the SPP applied to three parameters within the Thompson and Eidhammer (2014) aerosol-aware microphysics scheme in the WRF model. The parameters were specifically chosen as they involve perhaps the most highly uncertain internal elements in the scheme: 1) the size distribution of cloud droplets; 2) the size distribution of graupel/hail; and 3) the activation of cloud condensation and ice nuclei.