"Jarrell" Reprise

A Fresh Look at the Unusual Central Texas Tornado Outbreak of May 27, 1997

Lon Curtis


Waco, Texas


Alan R. Moller


Fort Worth, Texas

May, 1997 Southern Plains Overview

    >Only 48 tornadoes in the first 24 days of May… then … 98 tornadoes in three days, including:

    >25th: 48 tornadoes, mainly in Kansas, southern Oklahoma, TX Red River valley…

    >26th: 26 tornadoes, mostly in eastern Oklahoma…

    >27th: 24 tornadoes, central Texas…

1200UTC (7am CDT) Synoptic Situation

>250mb low centered near Scottbluff, NE

>Upper tropospheric jet southwest to northeast over northern Oklahoma; Central Texas in right entrance region …

>500mb low centered in central Nebraska with trough axis southward to western Texas …

>850mb low over southeast Kansas and 850mb cold front along TUL-ADM-ABI line

>Significant cold advection at 850mb occurring over Texas Panhandle and western Oklahoma north of the front …

Synoptic situation ... an atypical pattern

The NOAA Service Assessment for this event called the synoptic situation "atypical" because (1) tt occurred well south of stronger mid-level flow, (2) a strong low-level jet (usually a feature in violent tornado events) was absent, and (3) there was 'apparently' weak low- and mid-level winds and weak vertical wind shears (an uncommon scenario for violent tornadoes).
Service Assessment
The Central Texas Tornadoes of May 27, 1997
 N.O.A.A. 1998

... but this "atypical pattern" produced:

34 reports of severe weather

12 reports of large hail

24 reports of tornadoes, including several significant tornadoes:

One F5, one F4 & three F3 tornadoes

Why a "Jarrell" Reprise?

    Five years have passed since this unusual event. In the interim, there have been many other strong/violent tornadoes and, unfortunately, many lives have been lost as a result. Although there was an initial burst of interest in examining this event, the insuing events (in particular, the Oklahoma outbreak on May 3, 1999) resulted in redirection of interest away from this event, and a tendency to view this case as an unusual, perhaps inexplicable outlier. We think the case may not be such an outlier (i.e. that there may be other cases where significant tornadoes have occurred given similar atmospheric parameters). In a nutshell, our reasons for re-examining the case are these:

>Unusually strong tornado for apparently weak low- and mid-level upper flow

>Unusually deviant storm and tornado motion

>Question: What was the role of gravity wave in storm evolution and motion?

>Question: How much of an outlier was this event?

The Gravity Wave:

Was it a necessary component?

    Strong to severe thunderstorms developed over eastern Oklahoma on the evening of May 26th. Those storms evolved into a large convective complex that moved into western Arkansas during the overnight hours. Before sunrise, the storm complex began to collapse, generating one or more gravity waves. One wave propagated southwestward across northeastern Texas, reaching central Texas before midday.  More than one meteorologist investigating this event proposed that the gravity wave was involved in either storm initiation or storm motion. Among other issues, it was suggested that convective initiation occurred very early in the day considering the capping inversion found in the morning soundings. Corfidi (Preprints, 19th SLS, 1998) suggested that the motion of the low-level pressure perturbation field associated with the gravity wave favored new updraft development on the southwest side of the cell … permitting the original updraft to replicate itself southwestward in nearly continuous fashion. On the other hand, Purdom & Motta (Appendix B, NOAA Service Assessment, 1997) examined satellite imagery and concluded that the gravity wave may have played a role in the initiation of the storm complex … (but) that gravity wave moved south of the storm complex before the beginning of tornado activity. We will return to this issue later.

    It may be useful to briefly review some of the previous research by others with respect to this event. One of the troubling issues arising from the events of May 27th, 1997 was the origin of low- and mid-level vorticity and subsequent rotation of the storm updrafts in what was perceived to be an environment characterized by weak low-level flow and meager mid-level flow.

Previous Research:

Sources of Vorticity

    Magsig et al (16th W.A.F., 1998 and 19th S.L.S. 1998) examined the various tornadoes generated by the storm complex and found that all significant tornado damage occurred with three consecutive storms ... Lake Belton, Jarrell and Pedernales Valley. (This nomenclature is based on the location of the most significant tornado damage; there were other tornadoes.) Their examination of the data resulted in a judgment that the origins of mesocyclone and tornado rotation associated with the six well-sampled tornadoes did not come directly from stretching of vertical velocity on the pre-existing windshift … but from storm generated sources (tilting of baroclinically generated horizontal vorticity and/or stretching of vertical vorticity on storm generated gust fronts). (The pre-existing windshift boundary was a north-northeast - south-southwest boundary that bisected the area from near Waco to between Temple and Killeen, to west of Austin. It was readily identifiable on high-resolution visible satellite imagery, in WSR-88D base reflectivity data, and in surface observations.)

    Not directly related to this event but helpful in understanding the question of how the low-level vorticity may have been generated is research by Lee et al (19th S.L.S., 1998) into what they referred to as Non-Supercell Tornadoes*. Their conclusions were:

>Strong relationship exists between environmental CAPE and misocyclone/NST evolution and intensity.

>Boundary layer vertical shear was shown to be a critical factor in NST evolution …

>Simulations indicate that ambient vertical shear within a range of ~80 to ~120 percent of the optimal boundary layer vertical shear (of opposite sign to the cold pool circulation) produced well defined families of tornadic strength vortices ...


    *[We prefer the term non-mesocyclone tornado. The object of drawing the distinction in the first place is to ascertain whether the tornado was directly related to the mesocyclonic process or resulted from other processes, including those described by Lee et al. Our “quarrel” is with the semantics, not with the science presented by Lee et al.]

Previous Research:

Model Simulations

    Wicker et al (e-published, 1998 TAMU Dept. of Meteorology & CIAMS), utilizing the COMMAS model, simulated this event by running two simulations using the observed surface conditions at Temple and a modified composite sounding based on an M-CLASS sounding that was launched near Calvert, about 55 miles east-southeast of Temple (TPL). In one simulation, the conditions did not include the NNE-SSW boundary. That simulation produced a single, intense updraft that evolved into a storm that died within a hour as the cold pool generated by precipitation choked the updraft. A second simulation with the NNE-SSW boundary produced a persistent supercell-like "storm"  that developed intense low-level cyclonic circulations & repeatedly developed new updrafts on the southwest side of the main complex with new updrafts being drawn into and merging with the main complex.

    [Later we will briefly discuss renewed efforts to simulate this event using I-COMMAS, the successor program to COMMAS.]

Objective Analysis-1200UTC (7am CDT)

    To begin our reprise of this event, we used GEMPAK to objectively examine a number of atmospheric fields beginning 36 hours prior to the event (i.e. at 1200UTC on May 26th). The images below represent a small fraction of those we have examined.
(Click individual image for full-size)

     The top left panel is a plot of temperature advection at 850mb, which shows significant cold advection over Oklahoma and northwest Texas. This pattern was supporting a surface cold front that was along a Fort Worth-Abilene-Big Springs line at 7am CDT. The top right panel is a plot of 700mb omega. It reflects an area of upward vertical motion over western and central Texas. The center left panel is a plot of 500mb heights and isotachs. A band of 35 to 40-knot flow is depicted across the Texas Panhandle and much of Oklahoma to the south of a vort center in Nebraska. The right center panel is a plot of 300-700mb Q-vectors, and the lower left panel is a plot of Q-vector convergence for the same tropospheric section. Upward vertical motion is indicated by the area of Q-vector convergence that stretches from southwestern Oklahoma into central Texas. Not shown here, the Q-vector convergence analysis for 700-850mb also depicted an area of upward motion over central Texas.1

    The right lower panel is a plot of 250mb heights and isotachs, which depicts an upper tropospheric jet across the eastern Texas Panhandle, Oklahoma, and Kansas, eastward into Illinois. The location of this jet feature placed central Texas in the right entrance region of the jet max. All of these analyses taken together suggest that central Texas was in an area of persistent upward motion as early as 7am CDT on May 27th.
1 What's a Q-vector? A Q-vector contains the effects of geostrophic deformation on the horizontal temperature gradient vector in both magnitude and direction. If geostrophic deformation and temperature gradients exist (i.e., a disturbance exists), then thermal wind balance is being destroyed. The Q-vector describes this geostrophic "disturbance", which is forcing the atmosphere away from hydrostatic and geostrophic balance. For more information on Q-vectors, see http://meted.ucar.edu/awips/validate/qvector.htm .

Mesoscale Analysis

1200UTC (7am CDT)

   At 7am CDT, the mesoscale set-up reflected the following features: (a) a weak sub-synoptic low near Dallas; (b) dryline from the previous day (May 26th) along a Dallas - Junction line; (c) the cold front along a Fort Worth-Abilene-Big Springs line; (d) pooling of surface dewpoints greater than 74 deg F inland from the Gulf of Mexico to the old dryline; and (e) one or more gravity waves over eastern Texas propagating to the south and southwest. The first image below is the 7am CDT hand-drawn analysis created by Lon Curtis on the morning of May 27th. The set of images below that is a time-series of surface moisture convergence analyses utilizing data from the 0-hour output of the RUC model at 7pm on May 26th, and 1am, 7am and 1pm (all CDT) on May 27th. These plots depict a persistent moisture convergence maximum, probably associated with the sub-synoptic low, as it drifted southeastward during the overnight hours from southwestern Oklahoma to near Dallas, then to near Waco by 1pm CDT (Corfidi, private correspondence).

Surface Analysis-1100UTC (6am CDT)

This hand-analyzed regional map was prepared from one of the co-authors just prior to 1200UTC (7am CDT) on the morning of 27 May 1997.

Surface Moisture Convergence

(Click individual image to enlarge to full screen)


Hi-res Visible Image


    This high-resolution visible image is from 1645UTC (11:45am). Note the rope cloud structure that is oriented generally NE-SW from northeast of Dallas to near Waco to west of Austin to north of Del Rio. This feature had persisted most of the morning and coincides with the location of the day-old dryline. The gravity wave was just passing Waco at this time, as may be seen in a loop of visible imagery (below).

Visible Satellite Loop

KEWX Nexrad Radar Loop

    This is a loop of the base reflectivity images from the nexrad site at New Braunfels, TX (KEWX) covering the entire event. At the first frame (1735UTC), the initial thunderstorm is evolving in McLennan County near Waco. The storm(s) propagate southwestward along the NE-SW boundary while (beginning about 1935UTC) additional storms develop west of San Antonio and north of Del Rio. In addition to the storms, several mesoscale boundaries are detected. One of these (seen moving past and the radar and southward about 1900UTC) may be the gravity wave.


Updated with TPL 1800UTC Surface Observation

    Jon Davies (a private meteorologist in Wichita, KS) has been conducting research into issues relating to severe storms and tornadogenesis. In this thermodynamic diagram (used with permission), Davies used the ETA point sounding for Temple, TX (TPL) at 1800UTC and adjusted the lowest portion of the diagram for the observed temperature and dewpoint at Temple at that hour. As shown, this adjustment produced an extremely large CAPE value, with a significant portion of that CAPE in the 0-3km layer. Also note that the lifted condensation level (LCL) and level of free convection (LFC) were quite low and quite close together. This is consistent with some of Davies' recent work which suggests that tornadogenesis is enhanced by low LCLs and LFCs.

KGRK VAD Wind Profile

1729UTC/12:29pm CDT

    VAD Wind Profiles are generated automatically by the WSR-88d (nexrad) radars. This image is from the Central Texas-Fort Hood nexrad, located near Granger, TX, located about 25 miles northeast of Austin and about 60 miles south of Waco. The image is from 1729UTC, less than an hour before the first tornado formed at Spring Valley, southwest of Waco near Hewitt and Lorena. An interesting aspect of this image is the weak southerly flow in the lowest gates, which turns to the southwest around 3000 feet AGL and to northwesterly between 4000 and 6000 feet, then back to a southerly flow around 8-9 thousand feet, then back to a moderate northwesterly flow in the layer from 11 to 16 thousand feet.

Hourly Surface Analyses


(Click individual image for full-size)

20-minute Surface Theta-e


    This is a composite plot of surface observations (across the bottom) at Temple (TPL), Killeen (ILE) and Georgetown (GTU), along with a graph showing variation in surface Theta-e between 1700 and 2100 UTC. In addition, the time of the most significant tornadoes is indicated. Theta-e is plotted in degrees K and each station is denoted by a different color. Temple and Georgetown were both east of the NNE-SSW surface boundary while Killeen was to the west of the boundary. Surface  Theta-e reached a peak of more than 367K at Temple prior to the development of the nearby Lake Belton tornado at Morgans Point Resort. Note that there was little variation in the Theta-e at Killeen until outflow from the supercell storm to the east reached Killeen around 1955UTC. There also was little variation at Georgetown, but this plot ends just prior to the onset of precipitation there.

Nexrad Radar & Tornado Photos

(Paired by time)

Spring Valley-Lorena Tornado
Damage rated F2, 2.0mile path, maximum width 75yds

Radar image during Moody Tornado
Damage rated F3, 3.7mile path, maximum width 150yds
(Sorry, no photo available)

Lake Belton-Morgans Point Resort Tornado
Damage rated F3, 1.0mile path, maximum width 275yds

Prairie Dell Tornado
Damage rated F1, 2 mile path, maximum width 100yds

Prairie Dell Tornado at 2026UTC (3:26pm CDT)
Beginning transition that leads to Jarrell Tornado

Jarrell Tornado in multi-vortex organizing phase at 2029UTC (3:29pm CDT)

Jarrell Tornado intensifying at 2032UTC (3:32pm CDT)

Jarrell Tornado intensifying and approaching Double Creek Estates at 2036UTC (3:36pm CDT)

Jarrell Tornado entering Double Creek Estates at 2041UTC (3:41pm CDT)
Damage rated F5, 5 mile path, maximum width 650yds

GRK base velocity at 2039UTC/3:39pm CDT

Cedar Park Tornado
Damage rated F3, 9.2 mile path, maximum width 250yds
(Photo (C) Paul Chamberlain and used with his permission)

Pedernales Falls Tornado
Damage rated F4, ? mile path, maximum width ???yds
(Sorry, no photo available)





    The ETA Data Assimilation System (EDAS) is an intermittent data assimilation system that is generated every three hours by taking the most recent 3-hour forecast from the ETA and updating it with high frequency observations, such as wind profiler, NEXRAD, and aircraft data. The result is a gridded output that often captures features on a smaller-than-synoptic scale. In this study, we looked at the EDAS upper air data at 1800, 2100, and 0000 (all UTC) for indications of sub-synoptic accelerations in the flow that may have had an impact on the evolution of this storm system.  The charts below are from the 2100UTC (4pm CDT) EDAS output.


    These products, reflecting conditions at 2100UTC (4pm CDT), show that the mid-level flow over the area where the Jarrell tornado occurred was considerably more energetic than had been forecast by the morning suite of models, with the 700mb output showing a 25 knot west-southwesterly flow over the area, the 500mb output showing a westerly flow of 40-45 knots impinging from the west, and the 300mb output showing around 30 knots from the west-southwest.

Current Research: Low Shear, High CAPE Tornadoes

    As noted earlier, Jon Davies, has been conducting research into issues relating to severe storms and tornadogenesis for many years.  At the A.M.S. 21st Conference on Severe Local Storms (2002) he detailed some of his work dealing with low shear, high CAPE tornado cases. He found that a number of these cases, including the Jarrell event, invol common factors: a pre-existing wind shift boundary, along which storms develop rapidly; large low-level CAPE (rapid positive buoyancy increase in low-levels), and small near-surface CIN; and storm motion that remains on or close to the boundary, well to the right of the mean environmental wind. A scattergram, used with permission, shows the parameter space (decrease in lifted index per kilometer versus the height (AGL) of the midpoint of maximum lifted index decrease) for several of the low shear, high CAPE cases he has studied.

     Davies writes: "Estimated environments for the "weaker shear" tornado cases from this presentation are indicated by red squares, while environments for several "stronger shear" supercell tornado cases from 1999-2001 are indicated by black open squares.  Notice that, compared to the black "stronger shear" cases, all of the red "weaker shear" cases fall in the upper left half of the diagram. *** "Although these are only a few cases, the diagram hints that this stronger increase in low-level buoyancy may be a notable feature of "weaker shear" environments producing significant tornadoes, and certainly suggests further research in this area.  The diagram also suggests that a somewhat different combination of ingredients comes together for more typical supercell tornadoes in stronger shear environments where low-level buoyancy does not appear to be as crucial.

Current Research: Model Simulation of "Jarrell" Storm

    Adam Houston and Robert Wilhelmson are using a supercomputer at the University of Illinois-Urbana/Champaign to simulate the "Jarrell" storm and to research the role of pre-existing low-level boundaries on storm evolution in low-shear environments. Their research uses the I-COMMAS model of localized deep convection, a successor to the COMMAS model developed by Wicker, Wilhelmson and Skamarock. This model permits the researcher to specify (and change) environmental conditions between successive runs of the model. In this instance, the researchers specified the atmospheric conditions that preceded the evolution of the "Jarrell" storm, but have been investigating the role of the pre-existing north-northeast to south-southwest boundary by changing its depth and characteristics as well as by omitting it from the specified conditions.

Image from I-COMMAS model simulating one stage of the "Jarrell" storm

    Their preliminary results (reported at the A.M.S. 21st Conference on Severe Local Storms, 2002) are as follows:(a) cell initiation: location of the pre-existing boundary may have fostered mid-level and low-level mesocyclone formation and intensity by allowing cells to track along the gust front; (b) not all tornadoes were preceded by or even associated with mergers and not all mergers preceded or were associated with tornadoes; and (c)development of mesocyclones was not directly dependent on the pre-existing boundary but, the boundary likely resulted in intensification and protraction of cells and the formation mechanism may have promoted cell mergers.

    The use of very high resolution and high speed computer models to simulate severe deep convective storms may offer some of the most important advances in severe storm research in the future. If the I-COMMAS model can be tuned to accurately simulate severe thunderstorms, then researchers will be able to study storms in the laboratory without the need to intercept them in the field.

Palestine Profiler

    The National Weather Service operates a network of wind profilers that continuously detect the wind speed and direction through a substantial part of the troposphere. (Unfortunately, the latest proposed budget for NOAA/NWS terminates funding for the profiler program.) In May of 1997, the closest profiler to the event was at Palestine, about 80 miles east of Waco. The profiler was probably too far east and north of the storm system to provide data pertaining to the early stages of the system, but it does appear that the profiler may have detected the mid-level acceleration depicted by the EDAS plots shown above, as the higher speed winds reached into the Palestine area around 4pm CDT. On the far left, thin red rectangles indicate the accelerated flow in two segments, 2-3km (6500-9500 feet) above ground and 5-7km (15,000-22,000 feet) above ground.

Palestine profiler plot for 48-hour period ending at 7pm CDT on May 27, 1997 (time runs right to left)

Our Findings

>Mid- and upper-level flow strengthened during the day … as seen in EDAS data, the Palestine profiler, and comparison of 12Z and 0Z UA data. This raises the question: how much of an outlier was this event? Davies has accumulated several other cases of strongly deviant motion in proximity to NNE-SSW boundary; we believe that there may be others (possibly the F4 in Mason County, TX 5-11-99). There is interesting research underway to model this storm and the deviant storm and tornado motion mayyet be explained by the work of Houston & Wilhelmson.

>The gravity wave may have aided in preconditioning the incipient storm environment, but was not a necessary component of initiation of this storm system. Moisture convergence over the area had been persistent and pronounced for more than 24 hours prior to the event as evidenced by the mixing ratio at 925hPa:  ~18g/Kg near FTW at 0Z on the 27th, ~16g/Kg near ACT at 12Z on the 27th, and ~18g/Kg near DRT at 0Z on the 28th.

>The sub-synoptic low, originating near Wichita Falls on the 26th, was near Dallas at 12Z on the 27th, and moved south-southwest along the old dryline to near ACT at 18Z, providing enhancement of surface moisture convergence. The sub-synoptic low became the focus of the initial storm development and propagated to the south-southwest with the storm system through the afternoon.

>Persistent and widespread ascent was present over the area through a deep layer of the troposphere as reflected in Q-vector analysis and div-Q plots from both 0Z and 12Z on the 27th. The Q-vector convergence depicted appears to be the result of the right-entrance region of the 250hPa jet which was located primarily over Oklahoma, eastern Kansas and southwestern Missouri between 0Z on the 27th and 0Z on the 28th.

What's Next?

    We intend to take a closer look at mesoscale and sub-synoptic features using satellite cloud motion and derived satellite winds. We also will examine mesoscale winds using VAD wind profiles from the nexrad sites surrounding the event. And we have begun examining other Texas cases that may involve similar processes (F4 tornado in Mason Co., 1999; F3 tornado near Lake Whitney, 2000). Click here to view some preliminary work on those cases.