severe local storm paper-1

Preprint, 18th Conference on Severe Local Storms, Amer. Met. Soc.,
San Francisco, CA, Feburary. 1996

REAL TIME PREDICTION OF STORM-SCALE WEATHER DURING VORTEX-95,
PART II: OPERATION SUMMARY AND EXAMPLE CASES
Ming Xue¹, Keith Brewster^1,2, Kelvin K. Droegemeier^1,2,
Fred Carr^1,2, Vince Wong¹, Yuhe Liu¹, Adwait Sathye¹, G. Bassett^1,2,
Paul Janish³, Jason Levit^1,2 and Phillip Bothwell³
¹Center for Analysis and Prediction of Storms
²School of Meteorology
University of Oklahoma
Norman, OK 73019
³National Severe Storms Laboratory, Norman, OK, 73069

1. INTRODUCTION

The Center for Analysis and Prediction of Storms (CAPS) at the University of Oklahoma has a primary mission of developing tools and methods for numerical weather predictions (NWP) on the storm scale. CAPS envisions its research and development effort culminating in a storm-scale weather prediction system that can be used at regional centers or individual forecasting offices without much retooling.

CAPS started a yearly evaluation and testing of its prediction system, known as the Advanced Regional Prediction System (ARPS, Xue et al., 1995), in 1993. During the first two years, ARPS was run in a relatively simple mode, i.e., by initializing from a single forecast sounding on a relatively small (115x115) 1 km resolution domain. Although this type of experiment cannot predict the exact time and location of storm activity, it is valuable in its ability to determine the general mode of convection (Janish et al, 1995; Brooks et al., 1993). More significant to CAPS is that these tests established a close working relationship between CAPS scientists and operational forecasters, and exposed the scientists to and helped them solving the logistical issues pertaining to real time forecasting. Moreover, they paved the way for more sophisticated experiments for 1995 and future years.

In this paper, we present a brief summary of the prediction experiment of 1995 and three sample cases, then conclude in Section 5 with a discussion. The goals, logistic issues, computational aspects and future plans for the experiment are given in Droegemeier et al. (1996).

2. PREDICTION EXPERIMENT DURING VORTEX-95

During the spring of 1995, ARPS was run for a 7 week period daily to produce 6 h forecasts at two resolutions, namely, 15 km and 3 km. The experiment was carried out as part of the VORTEX-95 (Rasmussen, et al., 1994) field program. The ARPS predictions were much more realistic than those of the previous two years in that they used 3-D initial fields on model grids that included real terrain.

The forecasts started from initial fields at 18Z. The 336x336 km² 3 km grid was one-way interactively nested within the 1200x1200 km² 15 km grid (see Droegemeier et al. 1996 for a grid layout and a data flow chart). The 3 km grid was expected to resolve convective storms directly, thereby explicitly predicting the storm development and propagation. The 15 km grid was to provide a set of dynamically consistent boundary conditions for the 3 km grid while resolving meso-[beta] scale features over a bigger domain. It turned out that the 15 km grid also had considerable skill in predicting the location of convective activity.

Limited by the real time constraint of analysis data, the 15 km grid was initialized using the NMC RUC 6 hour forecast interpolated to the ARPS grid. The RUC forecast was also used to force the model boundary. The 3 km grid was, on the other hand, initialized using analysis fields from the OLAPS system (Brewster et al. 1994). As outlined in Droegemeier (1996), future experiments will use analysis that is performed directly on the ARPS grid, and will use, among others, the WSR-88D wide band radar data. An assimilation cycle has been planned for next year.

3. OPERATION SUMMARY

The forecast fields from the 15 and 3 km grids ware made available on the CAPS WWW page (http://wwwcaps.uoknor.edu/Forecasts) at around 18Z and 20Z each day, respectively. The early available time of the 15 km run was possible because the forecast from RUC (instead of analysis) was used as the initial condition. The effects of this were twofold; on the one hand, any error that is present in the RUC forecasts would be carried into the ARPS forecast, on the other, in the regions where observational data coverage is poor (e.g., western TX and eastern New Mexico), the forecast fields may contain more mesoscale details and be dynamically more balanced than analysis data. In at least one case, posterior prediction using analysis yielded poorer results. The latter suggests the need for a data assimilation procedure that can yield a set of dynamically consistent initial conditions while maximizing the benefit of observations.

The ARPS products were generated based on discussions with the VORTEX forecasters. Six color panels were produced at 3 hour intervals for each grid and placed on the CAPS Web page. These fields include temperature, wind, geopotential height, vertical velocity, horizontal moisture convergence at different levels, sea-level pressure, surface dew-point temperature, CAPE and CIN and accumulated surface rainfall. Three hour difference fields were also available. In addition, VORTEX forecasters had access to the 3-D ARPS gridded data that were converted into GEMPAK format for easy viewing and manipulation in NTRANS. An evaluation form was designed and the assistant duty forecaster for VORTEX was assigned to perform daily evaluation on the ARPS products.

4. SAMPLE CASES

The spring of 1995 was particular rich in storm activities. During the two month period from early April to early June, VORTEX had operations on 18 days. Severe tornadoes were reported within the VORTEX domain on April 17, May 7, May 17, June 2 and June 8. On all except May 7, the VORTEX field operation was on the tornadic storms and collected rich data sets using mobile Doppler radar, mobile mesonet, mobile soundings and airborne Doppler Radars. ARPS became operational on April 26 and operated through June 8, when VORTEX formally ended. Since the main interest of the forecasting experiment during VORTEX-95 was on severe weather, we presented here as examples the results from three of the 5 tornado days, i.e., May 7, June 8 and May 16, 1995. Two violent tornadoes that occurred on June 2 in the far western TX were outside the region of pre-configured 3 km grids. To be complete, other type of events such as passing cold fronts also occurred during this period.

4.1 May 7, 1995 multi-squall line case

On 7 May, 1995, the surface chart at 18Z was characterized by a distinct north-south oriented dryline extending from the mid-TX panhandle to southwestern TX. Associated with it were strong southeasterly winds on the east side and southwesterly winds on the west. This region lay ahead of a deep upper-level low centered west of the Rockies, with weak troughs moving northeastward from the southwest. Deep moist air was transported northward by the flow east of the dry line, which evolved into a 20 to 25 m/s low-level jet by the evening (local time). Two main lines of thunderstorms occurred during the day. In the early morning, storms originated from eastern New Mexico (NM) and moved northeastward across the TX and OK panhandles and joined up with a line of cells ahead of the dryline in western TX. By 15Z, these two clusters had become a connected line in western OK and continued to move eastward into central OK. The southern end of this line broke into an isolated supercell later at the OK-TX border and spawned an intense tornado on the west side of Ardmore, OK between 21Z and 21:30Z.

The second line of storms (squall line) formed between 15Z and 16Z in western TX along the surface dryline and propagated eastward ahead of the dryline. An HP supercell formed in the middle of this line at around 21Z at the western OK and TX border and was targeted by the VORTEX operational team. No tornado formed, however. This line continued to propagate eastwards while evolving into a classic solid squall line with a trailing stratiform precipitation region that is clearly seen in the radar reflectivity field (Figure 2a). This squall line extended more than 1000 km and lasted for more than 10 hours while moving across OK into Missouri (MO) late in the evening. The dryline stayed within the TX panhandle.

Figure 1. Predicted surface dew-point temperature (Cdeg., thin contours, interval 0.5), wind (m/s) and the 700 mb vertical velocity (m/s, thick contours, interval 2.5) on the 15 km grid at 00Z, May 08, 1995. Negative contours are dashed.

ARPS had an encouraging success in this day's prediction in spite of the fast movement speed of the squall lines. In Fig. 1, the surface wind (Vsfc), dew-point temperature (Td) fields and the 700 mb vertical velocity (w) fields at 00Z of May 8 (6 h of prediction) are given for the 15 km grid. The Td field shows that the northern end of the dryline remained within the TX panhandle, which agrees with the analysis. The forward extrusion of the middle portion was, however, not observed. The analysis showed that the southern end of the dry line was centered at x=480 km, while the predicted position was at x=660 km. This shows that error can be introduced by the boundary conditions (RUC forecast) that are used to force the model solution.

Figure 2. Observed (a) and predicted (b, in contours) low-level radar reflectivity fields at 00Z, May 08, 1995. Panel (b) also includes the predicted surface winds and a zero [[theta]]' contour shown in thick lines. The squared box in panel (a) corresponds to the forecast domain in panel (b).

At 6 h, the primary precipitation pattern (not shown) was centered north-south in central-western OK, coinciding with a line of strong surface flow convergence and large w at 700 mb (Fig.1). The 15 km grid was not distinctly resolving the two squall lines, and neither did it accurately predict their movement. This may be partly because the model was initiated after both lines had already formed and were propagating. Some of the mesoscale forcing might be missing in the 60 km resolution RUC forecast fields. Still, the ARPS managed to predict the location of the main convective activity.

The 3 km prediction enjoyed a much greater success. On this day, the 3 km domain was centered at the south-southwest OK border. Limited by its size, it did not cover the regions of initiation (further to the west) at earlier time. At 3 hours, one vigorous N-S line of storms (not shown) developed in the model at the north-central part of the domain and moved toward the northeast. By 6 hours (Fig.2b), this first line is near the northeast corner of the domain while a second longer line (shown by the reflectivity contours and surface flow convergence) is seen stretching from the north-central border to the southwest corner of the grid. The second line started at about 4 h near the west boundary and made its way into the central part of the domain. Fig.2b can be compared with the observed reflectivity field inside the square box in Fig.2a that represents the same domain. Overall, the ARPS 3 km grid predicted the location and movement of these two squall lines remarkably well, although storm initiation was delayed by as much as two hours. This may not be surprising considering the less than optimal initiation time and the limited grid size. At the same time, it suggests the necessity for a procedure that assimilates data associated with the existing storms.

4.2 June 8, 1995 TX-OK panhandle tornadic supercell storm case.

On June 8, 1995, there existed an upper-level low over the central and northern Rockies and a strong southwesterly flow over the central states. A deep pool of cold air spread south across the northern and central US behind a cold front. This cold front extended from extreme northeast NM across southern KS and MO to New England. Very moist air persisted to the south of the cold front, with the surface-based CAPE rising in the range of 5000-7000 J/kg in the late afternoon. The vertical wind shear was favorable for the development of updraft rotation.

At the same time, a dryline extended north-south in the TX panhandle west of Amarillo, intersecting the cold front at the far northern TX panhandle. Between 18-19Z, the first storm formed near the front-dryline triple point in the far northern TX panhandle but weakened soon as it moved northwestward into stable air. The main activity occurred a little later in the eastern TX panhandle as several supercells developed along the dryline at around 20-21Z and produced several large and long-lived supercells. These storms became tornadic and remained so until at least sunset as they approached the western OK border.

Figure 3. Predicted surface dew-point temperature (Cdeg., thin contours, interval 0.5), wind (m/s) and the 6 h accumulated rainfall (mm, thick contours, interval 5) on the 15 km grid at 00Z, June 08, 1995. Negative contours are dashed.

The above sequence of events was well predicted by the ARPS. The accumulated precipitation from the 15 km grid at 00Z is given in Fig.3. It shows that the heaviest precipitation occurred at the front-dryline triple point in the OK-TX panhandle. The 3 km grid predicted the initiation of storm cells between 19-21Z along the cold front which then turned between 21Z and 00Z into a north-south orientation roughly parallel to the dryline. These storms moved northeastward across the western TX-OK border into OK and exhibited strong rotation. ARPS finished the 6 h prediction at around 2 pm (20Z) local time, providing a 4 hour forecast lead time. In Fig.4, we compare the ARPS predicted Radar reflectivity field (panel b) against that of the WSR-88D (panel a) at 00Z, June 9. It can be seen that they agree remarkably well.

4.3 May 16 Kansas multiple-tornado supercell case

In the afternoon of May 16, 1995, a nearly stationary cold front stretched NE-SW across KS and a NE-SW oriented dryline was located on the west side of the TX panhandle. Abundant low-level moisture with surface dew-points around 22 Cdeg. was brought south of the front and the east of the dryline. This front-dryline pattern was similar to that of June 8. Late in the afternoon (22-23Z), storms were initiated near the front-dryline intersection just west of Garden City, KS and quickly became severe and developed rotation. A right moving storm produced a number of tornadoes between 23:30Z and 2:30Z as it moved east-northeast to near Hanston, KS. Another storm formed around 23:30Z over the extreme Northern TX panhandle at the confluent region associated with the dryline. This storm produced large hails.

Figure 4. Observed (a) and predicted (b, in contours) low-level radar reflectivity fields at 00Z, June 9, 1995. Panel (b) also includes the predicted surface winds and a zero q' contour shown in thick lines. The squared box in panel (a) corresponds to the forecast domain in panel (b).

On this day, only the 15 km ARPS run was performed. Despite the coarse resolution, the model was able to correctly predict the location of storm activity along the dryline and the cold front. Figure 5 shows the surface Td, and wind and the 700 mb w fields at 6 h (00Z) from the 15 km forecast. The front and dryline are indicated by the Td contours and the convective activity is clearly seen in the w field. The model predicted convection near the front-dryline triple point.

Figure 5. Predicted surface dew-point temperature (Cdeg., thin contours, interval 0.5), wind (m/s) and the 700 mb vertical velocity (m/s, thick contours, interval 2.5) on the 15 km grid at 00Z, May 17, 1995. Negative contours are dashed.

5. DISCUSSION

A lot has been learn from this experiment, on both the model performance and related logistics issues. NWP products from storm-resolving models are still very new, and it has been and will continue to be an educational process to both the model developers and the end-users concerning the most effective use and best interpretation of the products. One example would be that the fields from a high-resolution storm-resolving model often exhibit 'noise' that in fact are resolved features of convective storms. Such 'noise' is not commonly seen in mesoscale or large scale NWP products.

For a new model system, time is needed to establish a reliable track record and give confidence in its predictions. When it comes to providing guidance to the operation of field experiments, a forecaster may choose to rely more on their own experience in nowcasting than on the model until clear superiority in the model performance is demonstrated. This is precisely goal of CAPS. Clearly, more detailed validation of the model results is necessary and steps have been taken in this direction (Carr et al., 1996). More sensitivity studies are being conducted on the model physics (Wong et al., 1996) and more detailed studies performed on selected cases (Xue et al., 1996). A 4DDA cycle is being added to the system that will incorporate single Doppler wind retrieval techniques developed at CAPS. The use of such high resolution data will be crucial in improving model predictions. We plan to continue the prediction experiment on a periodic basis and a proposed plan for next year can be found in Droegemeier et al. (1996).

6. ACKNOWLEDGMENTS

This research was supported by the Center for Analysis and Prediction of Storms at the University of Oklahoma, which is supported by NSF grant ATM91-20009, and by a supplemental grant through NSF from FAA. Computer resources were provided by the Pittsburgh Supercomputing Center. Further support was provided by NOAA NSSL and SPC.

7. REFERENCES

Brewster, K., F. Carr, N. Lin, J. Straka and J. Krause, 1994: A local analysis system for initializing real-time convective-scale models. Preprints, Tenth Conference on Numerical Weather Prediction, Portland, Oregon, Amer. Meteor. Soc., 596-598.

Brooks, H.E., C.A. Doswell III and L.J. Wicker, 1993: STORMTIPE: A forecasting experiment using a three-dimensional cloud model. Wea. and Forecasting, 8, 352-262.

Carr, F. H., J. Mewes, K. Brewster and K. K. Droegemeier. 1996: Quantitative verification of non-hydrostatic model forecasts of convective phenomena. This volume.

Droegemeier, K. K., M. Xue, A. Sathye, K. Brewster, G. Bassett, J. Zhang, Y. Lui, M. Zou, A. Crook, V. Wong, R. Carpenter, and C. Mattocks, 1996: Real-time numerical prediction of storm-scale weather during VORTEX-95. Part I: Goals and methodology. This volume.

Janish, P.R., K.K. Droegemeier, M. Xue, K. Brewster and J. Levit, 1995: Evaluation of the Advanced Regional Prediction System (ARPS) for storm-scale operational forecasting. Preprint, 14th Conf. on Wea. Analysis and Forecasting. 15-20 Jan., Amer. Met. Soc., Dallas, TX, 224-229.

Rasmussen, E.N., J.M. Straka, R. Davies-Jones, C.A. Doswell III, F.H. Carr, M.D. Eilts and D.R. MacGorman, 1994: Verification of the Origin of Rotation in Tornadoes Experiment (VORTEX). Bull. Amer. Met. Soc., 75, 995-1005.

Wong , V., M. Xue, K. Droegemeier, Y. Liu, X. Song, J. Zhang and L. Zhao, 1996: Impact of physics on the development of severe storms during VORTEX-95. This Volume.

Xue, M. , K. K. Droegemeier, D. Wang and K. Brewster, 1996: Prediction and simulation of a multiple squall line case during VORTEX-95. This Volume.

Xue, M., K.K. Droegemieer, V. Wong, A. Shapiro, K. Brewster, 1995: ARPS Version 4.0 User's Guide. 380pp. Available from the Center for Analysis and Prediction of Storms, University of Oklahoma, Norman, OK, 73019.