Insights from a Convective-Allowing Ensemble Data Assimilation and Prediction System for Lake-Effect Snow

Lake-effect snow (LES) is a cold-season mesoscale convective phenomenon that can lead to significant snowfall rates and accumulations to regions downwind of open bodies of water, including cities east of the US Great Lakes. While operational global numerical weather prediction models are approaching the ability to resolve critical lake-effect processes, high-resolution regional convection-allowing models continue to provide improved forecast skill for these events. However, they often have difficulty recognizing the shape, location, intensity, and spatially sharp gradients of precipitation that result from these very localized snowstorms that often have banded features. These forecast errors can be attributed to both model error and uncertainties in initial and boundary conditions, but only recently has research started to quantify the relative contributions from these sources. In this study, the Ensemble Kalman Filter (EnKF) is coupled with a convection-allowing regional model (WRF) ensemble to analyze the evolution and predictability of a long-lived and high precipitation LES event in December 2013, selected owing to a large number of detailed observations available for verification from the Ontario Winter Lake-effect Systems (OWLeS) field campaign. Ensemble design is found to be an important consideration; while varying model parameterizations can introduce considerable ensemble spread, lateral boundary conditions corresponding to weather conditions beyond the Great Lakes region can be influential for analyses and forecasts longer than one day, requiring an appropriate ensemble perturbation choice. Short-term forecasts initialized from both regional EnKF analyses and current operational global weather models show that there exists a strong forecast dependence on regional initial conditions, impacting the timing and intensity of predicted precipitation as well as band location and orientation as assessed by an object-based verification approach. This lends some insight into the limited timescales of practical predictability of LES events, but proves that a regional DA-ensemble forecast system can still be a skilled and useful tool in LES prediction at short- to medium-range lead times. Additional work using ensemble sensitivity techniques shows a further dependence on initial conditions, particularly in the impact of regional and mesoscale environmental features on the structure and evolution of LES events; regional-scale atmospheric waves, basin-scale circulations, and even unresolved dynamics play comparatively important backstage roles in LES forecast uncertainty. Overall, the improved forecast skill as well as invaluable probabilistic information illustrates the further utility of an ensemble data assimilation forecast system for the eastern Great Lakes region.