Corey Potvin, CIMMS/NSSL
Impacts of missing initial condition scales on supercell forecasts: Implications for convective-scale data assimilation
CoauthorsElisa Murillo, Montgomery Flora, and Dustan Wheatley
Abstract: Previous investigations of convective storm predictability at O(1 h) lead times have focused on the sensitivity of simulations to model grid spacing, small-scale errors in the initial condition, and uncertainty in the analysis of the storm environment. Comparatively little has been done to examine the sensitivity of storm forecasts to initial condition resolution. Impacts of initial condition resolution errors are of critical concern to the envisioned Warn-on-Forecast (WoF) paradigm, in which kilometer-scale ensemble forecasts will be initialized from analyses of storms. If initial condition resolution errors substantially impact subsequent storm evolution, then the accuracy of convective-scale forecasts will be limited until sophisticated data assimilation techniques can be run on fine grids in real-time. On the other hand, if storm evolution is relatively insensitive to the initial condition resolution, then sacrificing analysis grid resolution for forecast grid resolution could yield more accurate forecasts given available computational resources. The simplest way to accomplish this would be to perform data assimilation on a relatively coarse grid, then interpolate to a finer grid on which the model is subsequently integrated over the desired forecast period.
This study examines the evolution of initial condition resolution errors using both idealized and full-physics simulations of supercells. Initially missing scales regenerate within 10-20 min of model integration, but produce forecast errors that increase with cutoff wavelength. The errors, however, do not fundamentally alter the storm evolution, even for cutoff wavelengths as large as 16 km, suggesting that using a much coarser grid for data assimilation (e.g., 3 km) than for subsequent downscaled forecasts (e.g., 250 m) may not strongly limit forecast accuracy over the lead times examined (0-2 h). This result motivates the development of dual-resolution data assimilation and prediction systems.