About the NSSL Realtime WRF Forecasts
The National Severe Storms Laboratory (NSSL) and Storm Prediction Center (SPC) first began exploring the use of 4-km grid-spacing convection-allowing Weather Research and Forecasting (WRF) model simulations as potential forecasting tools during the 2004 and 2005 NOAA/Hazardous Weather Testbed Spring Forecasting Experiments (Kain et al. 2006). The results from these tests were extremely positive with forecasters particularly impressed by the ability of the WRF simulations to depict realistic convective-scale storm structures associated with phenomenon like mesoscale convective systems and discrete supercells. These successful tests along with other experiments using convection-allowing WRF simulations conducted by NCAR (Done et al. 2004) motivated NSSL scientists to establish a more permanent experimental modeling framework to provide storm-scale guidance to SPC forecasters and serve as a testing ground for the development of storm-scale model diagnostics (e.g., Kain et al. 2010). Collaborations with scientists at NASA-SPoRT provided the necessary computing resources for daily forecasts beginning in 2006 and this real-time modeling framework became known as the NSSL-WRF.
The NSSL-WRF uses 4-km grid-spacing and is run twice daily at 0000 and 1200 UTC with forecasts to 36 h over a full CONUS domain integrated using NOAA’s high performance computing resources. The configuration has remained relatively constant with only two sets of updates as of this writing. In June 2009, the model domain was expanded and the WRF model version was updated from 2.2 to 3.1.1, and in April 2013 the WRF model version was updated to 3.4.1. Physics parameterizations include the MYJ boundary layer scheme and WRF single-moment six-class microphysics scheme. Despite recent works that illustrate advantages to using double-moment microphysics in high-resolution supercell analyses (e.g., Jung et al. 2012) and idealized simulations of squall lines (Bryan and Morrison 2012), research analyzing various WRF model configurations using single and double moment microphysics scheme as part of recent NOAA/Hazardous Weather Testbed Spring Forecasting Experiments has yet to find a quantifiable objective or subjective improvement using double moment schemes over the 15 to 30 h forecast period in which the NSSL-WRF forecasts are most heavily utilized (Clark et al. 2012, 2014). Thus, because of its relative simplicity, computational efficiency, and familiarity with SPC forecasters and other users, NSSL-WRF continues to use WSM6.
Deterministic NSSL-WRF Configuration
Daily, real-time runs of the Weather Research and Forecasting (WRF) model are generated using 256 processors on the Jet HPC cluster (Raytheon/Aspen Systems) in Boulder, CO. The current configuration includes:
- WRF version 3.4.1
- MYJ BL/turbulence parameterization
- WSM6 microphysics
- RRTM longwave radiation
- Dudhia shortwave radiation
- Noah land-surface model
- Positive definite advection of moisture
- 4 km grid length (1200x800)
- 35 vertical levels
- Time step 24s
Initial and boundary conditions are obtained from interpolation of the routinely available 40km NAM Model fields obtained from EMC/NCEP, using the WRF Preprocessing System (WPS). Initialization time is 00 UTC and 12 UTC and forecast length is 36 h.
Ensemble NSSL-WRF Configuration
In early 2014, NSSL requested and was granted an increase in computing allocation on the Jet HPC cluster (Raytheon/Aspen Systems) in Boulder, CO. The increased allocation is being used to run a nine-member NSSL-WRF Ensemble. Currently, the 9 members are comprised of the regular NSSL-WRF, which uses the 0000 UTC initialized NAM for ICs and LBCs, one member that uses the 0000 UTC initialized GFS for ICs and LBCs, and 7 members that use different members of NCEP's 2100 UTC initialized SREF system for ICs and LBCs. The SREF-initialized members are initialized at 0000 UTC using 3-h SREF forecasts (valid at 0000 UTC), and LBCs for the SREF-initialized members come from the corresponding SREF member used for the ICs. The SREF system ICs/LBCs include 3 WRF-ARW members (the control member and two perturbed members), 2 NMM members (the control and one perturbed member), and 3 NMMB members (the control and two perturbed members). The domain and physics parameterizations for each NSSL-WRF ensemble member are identical to the regular NSSL-WRF. Although the unvaried physics will have lower spread than a varied physics ensemble, forecasters are much more familiar with the behavior of the NSSL-WRF physics, and this will allow a clear diagnosis of the contribution of ensemble variance from the ICs/LBCs. Note, the current ensemble configuration is preliminary and will be changed based on subsequent objective and subjective evaluations and the development of new ensemble data assimilation approaches at NSSL. However, the configuration of the regular NSSL-WRF run will not change.
For more info on these experimental high resolution model forecasts, email Adam Clark at firstname.lastname@example.org