The Sensitivity of Supercell Simulations to Initial Condition Resolution: Implications for Warn-on-Forecast

The effects of initial condition resolution on idealized supercell simulations are analyzed. This study is motivated by the NOAA Warn-on-Forecast (WoF) program, whose goal is to develop a real-time, convention-allowing, ensemble analysis and prediction system. The WoF program envisions a paradigm shift from “warn-on-detection,” or prediction of severe convective storms based primarily on current observations, to a regime where storm-scale data assimilation and prediction systems play a much greater role in the severe weather warning process. The focus of the present investigation is therefore on the prediction of model quantities of greatest significance to severe storm forecasters, including updraft strength, low-level vorticity, surface winds, and rainfall.

Idealized simulations are run using the WRF-ARW model with grid spacing fixed at 333 m. Each control simulation uses a thermal bubble to initialize a supercell. The model fields from each control simulation are then filtered at various stages of storm development using cutoff wavelengths of 2, 4, 8, and 16 km. New simulations are then initialized from the coarsened model states and compared to the control simulations to assess the impact of the reduced initial condition resolution. Isolating the error due to limited initial condition resolution enables straightforward evaluation of the scales that need to be well analyzed to generate reliable model forecasts of various severe storm hazards.

Most of the model variables examined are relatively insensitive to the initial condition resolution, with generally operationally insignificant errors arising in the simulations once the filtered scales have regenerated (within 10-20 min). Large errors, however, occur in the evolution of low-level vorticity, which has important implications for the practical predictability of tornadoes. Results also indicate that the sensitivity to initial condition resolution is itself dependent on the degree of storm maturity. Errors in the simulations initialized early in the storm life cycle do not steadily increase with cutoff wavelength, whereas the simulations initialized once the storm is mature monotonically degrade as filtering is increased. This may be due to the increasing role of smaller scales, and the correspondingly greater errors that result from their neglect, as the storm evolves.