Evaluating Stochastic Parameter Perturbations in Convection-Permitting Ensemble Forecasts of Lake-Effect Snow

Lake-effect snowstorms, such as those occurring downwind of the Laurentian Great Lakes, can produce large, highly localized, snowfall accumulations that are challenging to simulate and forecast accurately. Effective forecasts need to include both expected snowfall and measures of forecast uncertainty. One type of numerical weather prediction forecast uncertainty for these events arises from the uncertainties associated with the parameterization of important sub-grid processes such as planetary boundary layer and surface layer turbulence (PBL/SL) and cloud and precipitation microphysics (MP). One way to represent the impact of physics scheme uncertainty is to design convection-permitting ensemble forecasts that use stochastic parameter perturbation (SPP) to vary individual uncertain parameters within physics schemes. The goal of this research is to evaluate and improve the utility of SPP for representing uncertainty in ensemble forecasts of lake-effect snow, with a focus on PBL/SL and MP parameterizations.

We focus on a snowfall event observed during the Ontario Winter Lake-effect Systems (OWLeS) field campaign that occurred from 10–12 December 2013. Snowfall was produced by a long-lake-axis-parallel snow band extending from Lake Ontario over the Tug Hill Plateau, where snowfall totals exceeded 100 cm. We run nested simulations of the event down to 1.33-km horizontal grid spacing using the Weather Research and Forecasting (WRF) model configured similarly to the operational High-Resolution Rapid Refresh model. All simulations use the Thompson-Eidhammer aerosol-aware MP scheme and Mellor-Yamada-Nakanishi-Niino PBL/SL scheme, both of which are configured to include SPP. To understand the impacts of SPP, a suite of 20-member SPP ensemble simulations are run, including ensembles where SPP is applied only to PBL/SL or MP, or where SPP applied to multiple schemes concurrently. To compare the effects of SPP relative to other sources of ensemble spread, additional simulations are run where perturbations to initial and boundary conditions (ICs/BCs) are applied instead of SPP, or where SPP and IC/BC perturbations are applied together. Results from the ensemble forecasts are evaluated against observations from OWLeS, including snowfall (from manual observations, gauges, and radar estimates), snow band characteristics from scanning and profiling radars, and atmospheric profiles from rawinsondes.

Results indicate that SPP perturbations produce substantial spread in simulated precipitation, despite having only modest impacts on the free tropospheric synoptic scale flow. They accomplish this by modulating lake-atmosphere fluxes, boundary layer characteristics, precipitation growth processes, and hydrometeor terminal fall speeds. The spread and skill of simulated precipitation from an ensemble using SPP alone is comparable to that from ensemble that uses IC/BC perturbations alone. The specific physical pathways whereby different SPP perturbations generate spread in simulated lake-effect precipitation are examined and discussed.