Ensemble Forecasts and Regional EnKF Data Assimilation for Prediction of Lake-Effect Snow

Lake-effect snow (LES) is a cold-season phenomenon that can bring intense snowfall rates and extreme accumulations in short periods of time. These events result when a warmer body of water destabilizes colder arctic air above, and can be convective in nature, typically developing banded features. While LES is a common occurrence around the wintertime Great Lakes, challenges remain in providing accurate numerical forecasts of the timing, location, and amounts of precipitation during these events, all of which can vary strongly on small scales. This work seeks to more deeply understand and quantify the predictability of these meso-gamma scale storms through the use of a regional convection-allowing ensemble and data assimilation using the Ensemble Kalman Filter (EnKF).

The numerical experiments highlight an event which occurred during the Ontario Winter Lake-effect Storms (OWLeS) field campaign of 2013-2014, which provided additional sources of verification data. Using the Penn State WRF-EnKF data assimilation and prediction system, several ensembles are constructed, populated by perturbing initial conditions using climatological background errors, varying the choices of physical parameterization schemes, and/or using perturbed initial and boundary conditions from a global model ensemble. Ensemble performance, including overall ensemble mean error, ensemble spread, and spread growth, are shown to be strong functions of ensemble design, especially in choices of lateral and lake surface temperature boundary conditions. These results can be linked to the type and size of perturbations, the size of the domain of the limited-area WRF model, the synoptic dynamics that aid in driving LES events, and convective-scale instabilities. Further analysis shows that regional data assimilation has a positive impact on the simulations when compared to observations and precipitation products, reducing RMSE in low-level temperature and wind fields and improving the timing of LES band initiation, although this influence is most pronounced in the short range.

The results of this analysis will be presented, as well as challenges for both research and operations. Overall, a well-designed ensemble shows promise for enhancing short-term prediction of these extremely localized but high-impact events, and provides a useful platform for future work.