Comparison of impacts of WRF dynamic core, physics packages, and initial conditions on warm season rainfall forecasts

A series of tests were performed for 15 warm season convective system cases occurring during August 2002 to compare the impacts on rainfall forecasts of the choice of dynamic core, physical schemes, and initial conditions in the WRF model. The sample of cases was chosen from a subset of four months of retrospective simulations performed by WRF Developmental TestBed scientists. The 48 hour simulations covered a roughly 2500 x 2500 km domain centered over the central United States. In most tests, 8 km grid spacing was used and the model used Eta output for initial and boundary condition data. For all 15 cases, WRF was run with two dynamic cores, the Eulerian mass (EM) core and the nonhydrostatic mesoscale model (NMM) core. Two physics packages were used: the NCAR package consisting of Ferrier microphysics, Kain-Fritsch convection, Dudhia/RRTM radiation, NOAH land surface model, and the YSU planetary boundary layer scheme, and the NCEP package consisting of Ferrier microphysics, Betts-Miller-Janjic convection, GFDL radiation, NOAH land surface model, and the Mellor-Yamada-Janjic planetary boundary layer scheme. In addition to these tests, additional runs were available for comparison that used the EM dynamic core and NCAR physics but with 10 km grid spacing and RUC initial conditions instead of Eta.

It was found that initial conditions play a very limited role except in the first 6-12 hours of the forecast. The dynamic core choice has its biggest impact after 12 hours, and physical scheme choice has a strong influence throughout the 48 hour simulations. Subjective analysis of the events suggests that the physical scheme exerts the most noticeable impact on the forecasts by altering the fine-scale character of the precipitation (most likely a result of the differing convective schemes used), but the dynamic core may have had a larger impact on the location of the general regions of rainfall. Preliminary results suggest that ensembles used for very short-term guidance may benefit most from the use of mixed physics, while ensembles used for 1-2 day or longer forecasts will benefit from the use of both mixed physics and mixed dynamics.