14
The Tornado Outbreak of 24 June 2003 Observed
in a WRF Environment
JAY TROBEC[1]
Sioux Falls, SD
Abstract
The NMM-WRF (Non-hydrostatic Mesoscale Model-Weather Research and Forecasting) framework for numerical weather prediction became operational within the National Center for Environmental Prediction on 20 June 2006. WRF was not in widespread release on 24 June 2003, when a single-day record 67 tornadoes occurred in eastern South Dakota.
We input initialization data from the then-operational EDAS (Eta Data Assimilation System) to take a retrospective approach, to see how the 2003 event would have been portrayed in a high resolution WRF framework.
1. METHODOLOGY
The Weather Research and Forecasting (WRF) Environmental Modeling System (EMS), hereafter referred to by its more common name Workstation WRF, is a full-physics numerical weather prediction package. It includes the same dynamical cores of both the National Center for Atmospheric Research Advance Research WRF (ARW) and National Center for Environmental Prediction Non-hydrostatic Mesoscale Model (NMM-WRF). It is designed for use on Linux workstations at NWS forecast offices and the university community (SOO/STRC, 2006).
In order to recreate the environment at the time of the tornadoes, we initialized Workstation WRF with 1200 UTC 24 June 2003 data from the archived ETA 32 km tiles. The model was run at 3 km resolution over a 375,000 km2 domain in the north central U.S., centered over eastern South Dakota (Fig. 1). The domain included about 40,000 grid points, and the 18 h simulation took about 9 h to run on a dual 1.8Ghz AMD 64 bit processor computer (Daniel Leins, NWS Cleveland, personal communication).
The simulation was run three times to test three different convective schemes. One utilized the Kain-Fritsch (KF; Kain and Fritsch, 1993; Kain, 2004) parameterization, and another run included the Betts-Miller-Janjic (BMJ; Betts and Miller, 1993) convective package. A third run took place with no convective parameterization.
2. CONVECTIVE PARAMETERIZATIONS
During a six-hour period on 24 June 2003, 67 tornadoes were confirmed in southeastern South Dakota (Fig. 2). They took place between 2300 UTC and 0400 UTC (25 June 2003). Although the 1800 meso-Eta was also available, the most recent full Eta run completed in time for forecaster use was at 1200 UTC on the day of the event, which is the run we emulated in our WRF simulation.
The WRF framework includes a routine that mimics the composite reflectivity radar imagery one might see from scanning radars in the NEXRAD network. This product estimates equivalent radar reflectivity based upon the forecast mixing rations, and the Rayleigh scattering abilities of the hydrometeors produced by the model (Koch et al., 2005). It is known to overestimate the reflectivity one might see in ground-based radar mosaic because the model is not subject to radar limitations, such as beam height overshooting low-level precipitation. Features which are invisible to the radar may actually be resolved by the model.
On the morning of 24 June 2003, convective storms were exiting the area from southwest to northeast. The rain shield can be seen on NEXRAD composite reflectivity from1200 UTC-1400 UTC (Fig. 3). By 1400 UTC, most of the rain had moved into Minnesota and North Dakota.
At 1500 UTC, only one storm remained on radar, a thin line of ordinary cells crossing the South Dakota-Minnesota border (Fig. 4a). A comparison of the WRF composite reflectivity products at the same time period shows that the KF and BMJ parameterizations, as well as the run with no parameterization (Fig. 4b-d), pick up on this feature – although they place it slightly south of where it actually occurred. But the WRF runs also depict several areas of spurious 5-10 dBZ returns in central and eastern South Dakota. By 1800 UTC, there is very little actual precipitation (Fig. 5a), with only weak echoes along what appears to be lingering convergence boundaries. The KF parameterization (Fig. 5b) grossly overestimates the areas of persistent light precipitation, possible due to convective feedback. The BMJ and no parameterization schemes (Fig. 5 c-d) depicted the situation more realistically, seeming to pick up boundaries although placing them in an incorrect location.
At 2100 UTC, the surface low was still to the south in Nebraska and there was no convection in South Dakota, but all three WRF configurations go convective along the SE to NW oriented stationary front in south central South Dakota (Fig. 6a-d).
3. DISCUSSION
The NCEP WRF (WRF-NMM) apparently suffers from convective issues. Crisis changes were made to several components including convective parameterizations in December 2006 in response to what were called “performance issues” such as “convection (deep and shallow) triggering issues” (DiMego, 2006). In this case depicted in Workstation WRF, morning precipitation continued too long and afternoon convection was initiated too early.
Admittedly we looked at a high-end event with tornado production far exceeding climatological probabilities. But it is an intriguing test of how Workstation WRF responds to high-end convective events.
Since the major precipitation features were best represented in the run involving neither the BMJ or KF schemes, it appears there was no need for a convective parameterization scheme (CPS) to resolve convection here. This is similar to the finding of Gilliland and Rowe (2007), who concluded there was no need for a CPS in their modeling of an idealized classic supercell in a 4 km resolution WRF environment. It also extends the recommendation of Galewsky (2006), who suggested it is best to turn off convective schemes at resolutions below ~12 km.
The over-forecasting of precipitation by the CPS’s in WRF can be demonstrated by examining the QPF (quantitative precipitation forecast) output from each of the configurations.
4. CONCLUSIONS
Operational forecasters must be wary when viewing WRF composite reflectivity images, because their realistic depiction of model feature may mislead a forecaster who does not completely interrogate model output. In this case, a 3 km run of Workstation WRF did not provide guidance of much value in convective initiation and propagation, due to over-forecasting regions of convective precipitation.
ACKNOWLEDGEMENT
The author is appreciative of the significant assistance on this project from Daniel Leins at NWS-WFO Cleveland.
REFERENCES
Betts, A. K., and M. J. Miller, 1993: The Betts-Miller scheme. The representation of cumulus convection in numerical models, Meteor. Monogr. No. 24, Amer. Meteor. Soc., 107-121.
DiMego,G., B. Ferrier, Z. Janjic, and E. Rogers, 2006: December 2006 NAM Changes. Mesoscale Modeling Branch, Environmental Modeling Center [online at http://www.emc.ncep.noaa.gov/mmb/namchanges_dec2006/nam_upgrades.nov2006.pdf, accessed 2006].
Galewsky, J., 2006: Resolution dependence of convective parameterizations in WRF, WRF Developmental Testbed Center [online at http://www.dtcenter.org/Visitors.05_06/galewsky_dtc.pdf, accessed 2007].
Gilliland, E.K., and C.M. Rowe, 2007: A comparison of cumulus parameterization schemes in the WRF model. 21st Conf. on Hydrology, Amer. Meteor. Soc., San Antonio, TX, P2.16.
Kain, J. S., 2004: The Kain-Fritsch convective parameterization: an update. J. Appl. Meteor., 43, 170-181.
______, and J. M. Fritsch, 1993: Convective parameterization for mesoscale models: the Kain-Fritsch scheme. The representation of cumulus convection in numerical models, Meteor. Monogr. No. 24, Amer. Meteor. Soc., 165-170.
Koch, S., B. Ferrier, M. Stoelinga, E. Szoke, S. Weiss, and J. Kain: 2005: The use of simulated radar reflectivity fields in the diagnosis of mesoscale phenomena from high-resolution WRF model forecasts. Preprints, 11th Conf. on Mesoscale Processes/32nd Conf. on Radar Meteorology, Amer. Meteor. Soc., Albuquerque, NM, J4J.7.
SOO/STRC, 2006: SOO/STRC WRF Environmental Modeling System, National Weather Service Science Operations Officers [available online at http://strc.comet.ucar.edu/wrf/index.htm, accessed 2006].
Wilson, J.W., and W.E. Schreiber, 1986: Initiation of Convective Storms at Radar-Observed Boundary-Layer Convergence Lines. Mon. Wea. Review, Vol. 114, No. 12, 2516-2536.
Fig. 1. Domain of the 3 km simulation grid. Mapped with Unidata Integrated Data Viewer (documentation at http://www.unidata.ucar.edu/software/idv/).
Fig. 2. Locations/paths of the 67 tornadoes in SE South Dakota on 24 June 2003. From damage survey conducted by NWS Sioux Falls.
Fig. 12. NEXRAD 3h precipitation accumulation from the KFSD WSR-88D, 0300 UTC 25 June 2003. View/scale similar to Fig. 3-Fig. 11.
[1]Corresponding author address: Jay Trobec, KELO-TV, 501 South Phillips Avenue, Sioux Falls, SD 57104. E-mail: