Rapid-Refresh Core Test: aspects of WRF-NMM and WRF-ARW forecast performance relevant to the Rapid-Refresh application

The present Rapid Update Cycle (RUC), the NCEP situational awareness and very short range forecast model, operational since 1994, is scheduled to be replaced by the Rapid Refresh (RR) in 2009. The Rapid Refresh will, like the RUC, be an intermittent (hourly) assimilation-forecast system, but the domain will be enlarged to cover nearly all of North America, and the current hydrostatic, hybrid-sigma-isentropic RUC forecast model will be replaced by WRF.

Since there are 2 cores available in WRF, it is necessary to arrive at a decision on which should be used for the RR. It was agreed that the Global Systems Division (GSD) within the Earth System Research Laboratory of NOAA and the WRF Developmental Testbed Center (DTC) would work together in partnership to conduct a comparative evaluation of the two available dynamical cores within the WRF software framework. These cores are the Advanced Research WRF (ARW) and the Nonhydrostatic Mesoscale Model (NMM). The purpose of this evaluation would be to provide a scientific basis for a recommendation to NCEP on which of these cores should be used for the initial RR implementation.

There were four aspects to this task: 1) deciding on evaluation criteria and design of the experiment necessary for the evaluation, 2) code modification and testing necessary to ensure that initial and boundary conditions, as well as physical parameterizations and the feedbacks of these with each dynamical core, were identical, or as similar as practicable, 3) running the forecasts and generating appropriate output, and 4) evaluation of the results, both objectively (using the NCEP verification system) and subjectively.

We decided to conduct the comparison using domains and grid configurations as similar as possible to the present operational RUC. This allowed use of RUC initial conditions (including hydrometeor species predicted by the NCAR microphysics used in RUC), but did require modification of the WRF Standard Initialization to accomodate these files. Analyses from the developmental 13km RUC cycle at GSD were used for these initial conditions. In order to be consistent with the procedures used for the operational RUC, boundary conditions were derived from the NAM (North American Mesoscale model, at this time, the Eta) forecasts initialized 6-h previous to the initial time of the core-test runs. For example, the run initialized at 12Z used boundary conditions from the NAM forecast from 06Z. Each model was run for 24h from 0000 and 1200 UTC of each day (for which initial and boundary conditions were available) for one month from each season (15 July - 15 August 05, 1-30 November 05, 15 January - 15 February 06, and 25 March - 25 April 06). The forecast domains and grid configurations covered nearly identical CONUS domains, but fit snugly inside the present RUC domain. It was not possible to make these identical because the rotated latitude-longitude projection used exclusively by the NMM is not available in the ARW.

Each of the four tasks listed above were carried out collaboratively by GSD and the DTC. Most demanding was the code preparation and testing to ensure interoperability of the physics schemes in the 2 cores. It was decided early on that the experiment would be divided up into 2 phases. These phases differed only in the physics suites used. Phase 1 consisted of the NAM physics as modified by the Environmental Modeling Center (EMC) of NCEP in February 2006 in preparation for the WRF-NMM replacement of the Eta as the operational NAM model at NCEP on 20 June 2006. Phase 2 used a "RUC look-alike" suite, including the RUC land-surface scheme (instead of the NAM version of the Noah scheme), the latest version of the Grell-Devenyi convective scheme (in place of the Betts-Miller-Janjic used in NAM), and the latest version of the NCAR-Thompson mixed-phase bulk microphysics with 2-moment snow and graupel size distributions (in place of the Ferrier scheme, which advects only one total condensate variable and partitions cloud, rain, snow and graupel diagnostically).

Verification was done both objectively and subjectively. The objective verification used as its foundation the NCEP verification system, with additional testing of "no-difference" null hypotheses by the DTC. The subjective verification was done primarily by experienced meteorologists in GSD. The forecasts from a particular phase were very similar, the most obvious difference being somewhat more horizontal small-scale (on scales less than 10-15 horizontal grid intervals) detail apparent in the ARW forecasts than in those produced by the NMM.

The NOAA/ESRL/GSD conclusions from the evaluation were summarized in a memo to NCEP on 31 August 2006, with the subject "WRF Rapid Refresh Dynamic Core Evaluation and Recommendation." The summary of conclusions in this memo reads "Some significant advantages were evident for one core or the other, dependent on variable or vertical level, with a slight edge for ARW overall, but we judged that there was no strong overall advantage for either." In our presentation we will illustrate various of these advantages for both cores.