Published in the 20th Conference on Severe Local Storms, Orlando, Florida, American Meteorological Society;   11-15 September 2000.

 

  SOME REMARKABLE SUPERCELL SIMULATIONS FROM A QUASI-OPERATIONAL LOCAL-SCALE MODEL: SKILL OR "SHEAR" LUCK?

                                                             Ed Szoke1,2   and Adrian Marroquin
                                                             NOAA Forecast Systems Laboratory
                                                             Boulder, Colorado   80303
 

1Corresponding Author address:
       Ed Szoke, 325 Broadway, R/E/FS1, Boulder, CO 80303.
       Email: szoke@fsl.noaa.gov

 
2Joint collaboration with the Cooperative Institute for Research in the
      Atmosphere, Colorado State University, Fort Collins, CO.
 
 

1. INTRODUCTION

For several years the NOAA Forecast Systems Laboratory (FSL) has been evaluating the concept of running a local numerical model onsite at a National Weather Service Forecast Office (WFO). The advantages to locally running a small-scale model include the ability to incorporate local datasets that may not be available at a national center. These data might range from mesonet surface reports from stations operated by state transportation departments to mesonet observations made available through cooperative efforts with local television stations. Of course it is imperative that such observations be quality controlled before their inclusion in a local analysis being used to initialize a local-scale model. For tests at FSL, the Local Analysis and Prediction System (LAPS, McGinley et al., 1991; homepage at http://laps.fsl.noaa.gov) was used to assimilate a wide range of data into a mesoscale analysis at 10-km grid spacing. The LAPS analysis has been used to test several different models at 10-km grid spacing, including the Eta, MM5, and two versions of the RAMS model, with displays on the FSL Web site.

About a year ago one version of the RAMS model, a parallelized version known locally as the Scalable Forecast Model (SFM, Hart et al., 1996), was made available to the local (Boulder, BOU) WFO via the Advanced Weather Interactive Processing System (AWIPS, Wakefield, 1998) workstation, enhancing the usability of the model to the forecasters. The SFM is run at 10-km grid spacing, initialized with LAPS, with boundary conditions provided by the Eta model, and run out to 18 h four times per day on a computer at the BOU WFO. The model uses the Schultz (1995) explicit microphysics scheme.

An explicit simulation is clearly an issue of some concern for convective simulations at a grid spacing as large as 10 km, and with this in mind, we believe that the best potential for adding value to a forecast in convective scenarios would include the early stages of events forced by topography (diurnal and/or organized upslope) or some other organized feature. Thus, the forecaster might be able to successfully use the SFM model to predict when and where thunderstorm development might occur, for example, on many days where terrain heating might drive a diurnal upslope flow. As expected, on many days convection in the SFM tended to organize more than would occur in actuality given the available vertical wind shear, with larger scale, more organized and persistent outflow boundaries produced by the model. There were, however, a few remarkably accurate forecasts on a limited set of days when there was sufficient vertical wind shear and instability to support supercell storms. The four days of successful simulations, out to 10 h and beyond in some cases, occurred during the 1999 convective season in the BOU WFO forecast area. Here, we mainly show one of these cases, and try to discern whether such successful simulations may represent some real forecast skill for the model in a type of event where the actual scale of convection (a persistent, relatively large updraft that occurs with a supercell storm) most closely matches the model's natural scaling when using 10-km grid spacing, or whether these solutions were just a matter of "shear" luck. We recognize the limitations present with our rather small sample, and indeed had hoped to collect more supercell cases before the deadline for this paper, but thus far it has been a rather pathetic severe weather season in the year 2000 in northeastern Colorado.

 

2. BRIEF DESCRIPTION OF THE SFM

As noted, the model runs in real time on a multiprocessor computer located at the BOU WFO and connected within the firewall to their AWIPS. There are normally four runs per day, of 18 h duration, starting at 0300, 0900, 1500, and 2100 UTC. As soon as a forecast hour is complete it can be viewed on AWIPS, overlaid with other data and/or model output from some other model as desired. The domain of the model is shown in Fig. 1, and is considerably larger than the forecast area of responsibility of the BOU WFO. As noted earlier, the model employs an explicit microphysics scheme developed by Schultz that carries various precipitation species. Using a conversion scheme, one of the model displays available on AWIPS is of equivalent radar reflectivity, so that one can overlay actual and model predicted reflectivity together, as will be done in the examples that follow. The model does not currently employ a hot start, and the delay until storms begin will depend on the forcing (low-level convergence, for example) that is contained in the LAPS analyses.

 
 Figure 1. Model domain and county outlines on a topography image (m), and METAR sites.

 

3. AN EXAMPLE OF SFM CONVECTION

Somewhat typical behavior of how the model "scales" convection under conditions of nominal, but below supercell criteria (which we consider to be when the hodograph length in the lowest 6 km AGL is at least 25 m/s) is shown in Fig. 2. Two different forecasts (one from the 1500 UTC run on 23 June 99, and the other from the next run initialized at 2100 UTC) of model low-level reflectivity and surface winds are presented together, overlaid with observed low-level reflectivity (from a NOWRAD display) for a case where the vertical shear through 6 km AGL was about 15 m/s. Convection began near 0000 UTC, and in both model runs about an hour later but in approximately the general area of higher terrain along the eastern slopes of the Front Range in southeastern Colorado. While in the real world a number of individual cells developed, with varying lifetimes, in the model solution each run produced one main storm (in Fig. 2a the northern model storm is from the 2100 UTC run, and the southern storm from the 1500 UTC run), which grew in size and organization over time. In Fig. 2b the forecast storm has actually dissipated for the 1500 UTC run, while the storm from the 2100 UTC run has taken on a bow shape and accelerated faster than the actual, much less organized convection on this day.

 

 

 
 Figure 2. SFM forecasts of surface wind and reflectivity (contours at 10 dBZ intervals, starting at 10 dBZ) from two runs, overlaid with observed low-level reflectivity (image, starting at 5 dBZ, in 5 dBZ gradations). In Fig. 2a is a 5 h forecast from the 2100 UTC run with an 11 h forecast from the 1500 UTC run, both valid at 0200 UTC. In Fig. 2b a 9 h forecast from the 2100 UTC run is presented with a 15 h forecast from the 1500 UTC run, valid at 0600 UTC, 24 Jun 99.

 

It is certainly not surprising to have fewer model predicted storms than observed, since the actual scale of updrafts with individual cells on lower shear days will be far less than what could be captured with a model employing a 10-km grid spacing. The amount of organization in the model appears to increase some with vertical wind shear, as would occur in the real world, and as shown in this case results in a rather poor solution after a few hours. When this model was introduced to forecasters at the BOU WFO, this type of behavior was explained, and it was emphasized that the best manner to use the model for forecasting summer convection might be: a) in the general location and timing of initial convection, b) where the model ends up producing and not producing storms (generally as occurred for this case), and c) realize that once storms form processes such as gust front interactions, etc., will be on scales that the model cannot forecast in its current grid spacing.

4. AN SFM SUPERCELL STORM FORECAST

The convective season of 1999 was the first full season with the SFM running in an "operational" mode at the BOU WFO, and there were four occasions to save runs when supercell storms formed. Here we will examine the case of 10 June 99, with a good estimate of the vertical wind shear for this day given by the hodograph in Fig. 3.

 
 Figure 3. Hodograph (scale in m/s) for Denver for 0000 UTC on 11 June 99. Surface to 6 km AGL portion of the hodograph is highlighted. Tick marks are at 1, 2, and 3 km AGL.

Output from the 1500 UTC 10 June 99 run of the SFM will be examined for this case. The model did a good job of forecasting the location and approximate timing of the initial cells that developed along the Front Range around noon local time (1800 UTC), as seen in the 4 h forecast in Fig. 4. The main, long-lived storm on this day evolved from the cell west-southwest of Denver (DEN) in Fig. 4, and over the next several hours moved east, first right across Denver International Airport (DIA), producing several cm of hailfall, and more severe hail reports as well as a brief tornado on its way to near the Kansas border by 0300 UTC 11 June.

 
 Figure 4. SFM 4 h surface reflectivity and wind forecast valid 1900 UTC on 10 June 99, overlaid with an image of observed 0.5 degree reflectivity, and a METAR plot. The reflectivity image begins at 5 dBZ, and is in 5 dBZ gradations. The SFM reflectivity contours are every 10 dBZ, starting at 10 dBZ.

A sequence of SFM prediction beginning at hour 6 and extending through hour 12 are presented in Fig. 5. Although the details of the observed reflectivity are impossible to discern from Fig. 5, it is apparent that the SFM did a remarkably excellent job with the location and timing of the main storm as it trekked across eastern Colorado. At 6 h (Fig. 5a) and 9 h (Fig. 5b) the model forecast is almost exactly on top of the observed storm, and by 12 h is only slightly behind the storm position. In terms of reflectivity, the SFM consistently produced a core of greater than 50 dBZ echo, while the observed core was stronger, generally exceeding 60 to 65 dBZ, but this is not necessarily unexpected with the 10-km SFM grid spacing. It is interesting that at 12 h the SFM forecast storm has not accelerated past the observed storm (as in the example in Fig. 2), but is actually slightly behind the actual storm. Given that the storm motion is also predicted correctly, which is off the hodograph in Fig. 3, the implication may be that the SFM is indeed predicting a supercell-type storm in this case, so this excellent prediction may in fact be more than just "shear luck."

 

 

 

 
 Figure 5. As in Fig. 4, but for sequence of SFM predictions from the 1500 UTC 10 June 99 run, for 6 h (a), 9 h (b), and 12 h (c). Arrow points to the observed reflectivity core in each image. The 10 dBZ SFM reflectivity contour is highlighted with a thin line, and the 50 dBZ contour with a thick line.

5. SUMMARY AND CONCLUSIONS

The following day (11 June) was another supercell day, and again the SFM made a very accurate forecast for a long-lived hailstorm that began west of Denver. A forecast for 11 h out from the 1500 UTC run of the SFM is shown in Fig. 6, and the main storm was still well-predicted on the 15 h forecast as it moved across southeastern Colorado. Although the observed reflectivity in Fig. 6 gives the appearance of a considerable underprediction by the SFM, the echoes north of the two main cells (indicated by the arrows in Fig. 6) were mostly less than 35 dBZ, except for a couple of 45 dBZ echoes northeast of DEN. The main, long-lived storm was the 60+ dBZ cell that is within the SFM's predicted 50 dBZ contour even at the 11 h forecast time.

 
 Figure 6. As in Fig. 5, except for the SFM 11 h forecast from the 11 June 99 1500 UTC run, valid at 0200 UTC on 12 June.

Two other consecutive supercell days (25 and 26 June 99) were somewhat different, with storms forming at the northern edge of the domain in southeastern Wyoming and moving east-southeast into extreme northeastern Colorado. For these two days the model forecasts were not as accurate as on 10 June, but still had the general idea of area of storm formation and movement. On these two days the rather small domain of the model likely influenced the accuracy of the forecast, since the echoes developed near the northern edge of the SFM domain. For such a case the prediction will be less influenced by the LAPS analyses and more by the boundary conditions provided by the eta model. These cases from 25-26 June 99, and hopefully some from the 2000 storm season, will be shown at the conference.

 

While we have shown a remarkably accurate prediction with the 10 June 99 case, the SFM had rather mixed reviews by the operational forecasters at the BOU WFO. As noted by Szoke et al. (1998), a significant reason for somewhat limited use of the model is the wide variation in the verification of the SFM forecasts, both in the winter but especially in the summer convective season. This is undoubtably in significant part a result of the fairly gross 10-km grid spacing, which results in the problems noted in our first example in Fig. 2. Although limitations of the model and discussion of when it might be most useful have been presented at forecaster workshops, it takes time for forecasters to discern when a prediction can be believed (as in the 10 June case) and when it is likely to go awry rather quickly. Indeed, one of the authors worked a shift on 11 June 99 and was only tempted to pay close attention to the SFM forecast after reviewing the success on the previous day. We hope to decrease the grid spacing in the future, either with the SFM or whatever local model is run, in concert with a finer resolution LAPS initialization, which should improve the overall utility of the local model.

6. ACKNOWLEDGMENTS

We thank John McGinley of NOAA/FSL for his review and Nita Fullerton of NOAA/FSL for a technical review of this manuscript.

 

7. REFERENCES

Hart, L., C. Baillie, M. Govett, T. Henderson, and B. Rodriguez, 1996: FSL's Scaleable Modeling System: A tool for the parallelization of NWP models. Preprints, 11th Conf. on Numerical Wea. Prediction. Norfolk, VA, Amer. Meteor. Soc., 443-446.

McGinley, J.A., S.C. Albers, and P.A. Stamus, 1991: Validation of a composite convective index as defined by a real-time local analysis system. Weather and Forecasting, 6, 337-356.

Schultz, P., 1995: An explicit cloud physics parameterization for operational numerical weather prediction. Mon. Wea. Rev., 11, 3332-3343.

Szoke, E.J., J.A. McGinley, P. Schultz, and J.S. Snook, 1998: Near operational short-term forecasts from two mesoscale models. Preprints, 12th Conf. on Numerical Weather Prediction. Phoenix, Arizona, American Meteor. Society, 320-323.

Wakefield, J.S., 1998: Operational risk reduction: Easing AWIPS into the field. Preprints, 14th IIPS Conf. Phoenix, Arizona, American Meteorological Society, 389-391.