Examination of the diurnal cycle of rainfall and ensemble prediction strategies in WRF model simulations

Two studies were conducted addressing issues relevant to warm season precipitation forecasts in the United States. First, the representation of the diurnal cycle of rainfall in a 5 km grid-spacing mesoscale model explicitly representing convection was compared to a 22 km grid-spacing model implicitly representing convection. Improvements were expected in the 5 km grid-spacing model diurnal cycle of rainfall because of its lack of a cumulus parameterization scheme (CPS). Previous works show that problems with CPSs adversely affect the timing of precipitation and the ability of the model to represent the mesoscale dynamics that lead to propagation. Results revealed that the 5 km model did have a significantly better representation of the diurnal cycle of rainfall than the 22 km model. The timing, location, and representation of both propagating and non-propagating rainfall areas were superior in the 5 km model. Second, a comparison of forecast skill and spread was made between a mixed-physics (MP) and perturbed initial and lateral boundary conditions (PI) ensemble. Forecast skill was compared using deterministic forecasts derived from each ensemble and using probabilistic forecasts from each ensemble. Results revealed that the MP and PI ensembles had similar skill when the deterministic forecasts were evaluated using equitable threat scores (ETSs). However, when the area under the relative operating characteristic curve (ROC score) was used to evaluate the probabilistic forecasts, the MP ensemble had higher skill at the beginning of the forecast while the PI ensemble had higher skill at the end of the forecast. This behavior was directly related to the spread. In the MP ensemble, because the initial and lateral boundary conditions were the same for each ensemble member, the spread stopped increasing after about 24 hours, and shortly after this time the PI ensemble ROC scores surpassed the MP ensemble ROC scores. However, during the first 24 hours of the forecast, greater spread in the MP ensemble forecasts was accompanied by higher ROC scores than in the PI ensemble. This demonstrates the importance of perturbed lateral boundary conditions in ensembles using limited area models if forecasts are desired beyond 24-36 hours