Eric Loken-February 23

Spread and Skill in Mixed- and Single-Physics Convection-Allowing Ensembles

Speakers

Loken, Eric
Ph.D. Student

Start

February 23, 2018 - 3:30 pm

End

February 23, 2018 - 4:00 pm

Address

120 David L. Boren Blvd., Room 5600, Norman, OK 73072   View map

Spread and Skill in Mixed- and Single-Physics Convection-Allowing Ensembles

Increasing computing power has led to the rise of convection-allowing ensembles, which have been used to improve the prediction of all modes of severe weather. However, the vast majority of convection-allowing ensembles remain under-dispersive, especially for fields related to convection. Previous work has suggested a variety of methods to increase ensemble spread, including perturbing initial and lateral boundary conditions and using multiple models and physics parameterizations. By comparing mixed- and single-physics ensemble subsets from the 2016 Community Leveraged Unified Ensemble (CLUE), this talk will investigate how the inclusion of multiple microphysics and boundary layer schemes influences forecast spread and skill at a variety of spatial scales and forecast hours. Forecast spread is analyzed for 2-m temperature, 2-m dewpoint temperature, 500 hPa geopotential height, and hourly accumulated precipitation, while forecast quality is evaluated for hourly and 6-hourly accumulated precipitation.

Time series indicate that mixed-physics ensemble forecasts generally have greater variance than comparable single-physics forecasts; however, the differences are small, especially at the largest spatial scales and after the ensembles are calibrated for bias. The two sets of ensemble precipitation forecasts also have similar areas under the relative operating characteristics curve, fractions skill scores, and reliability. Finally, the two ensemble forecasts are similar in three “high-precipitation” cases. These results suggest that model developers may prefer to implement single- as opposed to mixed-physics convection-allowing ensembles in future operational systems, while accounting for model error using stochastic methods.