Performance of the New NCEP/EMC High-Resolution Window and NAM CONUS Nest Models for Operational Severe Weather Forecasting Applications at the Storm Prediction Center
The HiResW features two CAMs run with different dynamic cores: ARW and NMMB. Among the changes included in the June 11, 2014, operational implementation are: 1) expansion to a full CONUS domain, 2) increased horizontal and vertical resolution, 3) replacement of the WRF-NMM dynamic core with the NMMB, 4) WSM6 microphysics in the ARW core, and 5) use of initial conditions (ICs) from the Rapid Refresh (RAP) model. Since it has been shown previously that CAM forecasts of convective storms can exhibit sensitivity to ICs, this change will likely introduce additional spread in the HiResW convective forecasts compared to CAMs initialized with NAM ICs, and also impact statistical properties of CAM ensembles that have HiResW members.
Changes to the NAM Nest, scheduled to be implemented operationally in August 2014, include: 1) ICs derived from the hybrid variational ensemble analysis with global EnKF, 2) use of the RRTM radiation scheme, 3) removal of parameterized convection, and 4) the latest version of the Ferrier-Aligo microphysics.
The SPC has had access to parallel gridded output from these new models starting in summer 2013 (NAM Nest) and early winter 2013 (HiResW) to facilitate longer periods of comparative evaluation and feedback to model developers, focusing on days with severe convective storms. Subjective evaluation of the simulated reflectivity forecasts from the new NAM Nest and HiResW, including during the 2014 Hazardous Weather Testbed Spring Forecasting Experiment, found that convective storm structure and intensity are more realistic and compare more favorably to radar observations of severe thunderstorms than their operational counterparts. In addition, objective verification of hourly forecasts of simulated reflectivity using a spatial neighborhood method indicates that the new model versions are more skillful compared to the previous operational versions, with somewhat improved statistical measures of POD, FAR, CSI, and bias.