Mesoscale models are important, useful tools for analyzing and forecasting snail-scale atmospheric phenomena. Ideally, finer grid-point resolution should make a model more likely to capture realistic small-scale structure. Because these models work to resolve phenomena that exist on very fine spatial- and time-scales, they are subject to high variability. Accurate initialization of mesoscale models is crucial to skillful short-term forecasting. This study exercises four different initialization and model physics experiments of four nested MM5 forecast domains and examines their respective short-term (f03, f06, f09, f12) forecasts. The exceptionally rich meteorological data set taken from the Southern Coast Ozone Study of 1997 (SCOS97) provides the basis for our model verification. We show that 3km and 9k resolutions produce better forecasts than the 27km resolution model, however, differences between the 3km and 9km resolution forecasts are essentially insignificant. We also show that different model initializations and physics schemes have an insignificant impact on improving the absolute accuracy of the numerical forecasts produced by a non-hydrostatic mesoscale model.