Extended Core Test


The Weather Research and Forecasting (WRF) model contains two dynamic cores: the Non-hydrostatic Mesoscale Model (NMM) core, developed at the National Centers for Environmental Prediction (NCEP), and the Advanced Research WRF (ARW) core, developed at the National Center for Atmospheric Research (NCAR). Each dynamic core corresponds to a set of dynamic solvers that operates on a particular grid projection, grid staggering, and vertical coordinate. The WRF model also contains a multitude of physical parameterizations, many of which can be used with both dynamic cores.

Several comparisons of forecasts by the ARW and NMM dynamic cores have been conducted in the past, including the WRF Test Plan (Seaman et al. 2004), the NSSL Spring Program (Kain et al. 2005), the Developmental Testbed Center (DTC) Winter Forecasting Experiment (DWFE, Bernardet et al. 2008) and the Rapid Refresh Core Test (RRCT, Brown et al. 2007). In all of these studies except for the Rapid Refresh Core Test, differences between the configurations using ARW and NMM went beyond the dynamic core to include variations in initial conditions, physics packages etc. Therefore, a formal assessment of the differences between the forecasts by the two cores was not possible until the RRCT. The results of the RRCT indicated that, out to a 24-h lead time, the differences between ARW and NMM forecasts for the configuration employed were quite small. The forecast from each core had strengths and weaknesses, as described in the DTC's RRCT report, with the ARW fairing an overall mild advantage.

The similarity in objective verification scores between ARW and NMM obtained during the RRCT led to the idea that research results obtained with one core could be transferred to the other. However, to ascertain this possibility, it was deemed necessary to test whether the similarity in results was limited to the first 24 h or could be carried onto a 60 h forecast. With this motivation, the Extended Core Test was designed with two goals: 1) to quantify the errors in the forecasts produced by the ARW and NMM dynamic cores of the WRF model in a given configuration, and 2) to quantify the differences between forecasts produced by the two dynamic cores in a given configuration. Both analyses were based on objective forecast verification statistics of the model.