Supercell and Tornado Parameters from a Large Dataset of Simple Forecast Soundings
Jonathan M. Davies      Wichita, Kansas        jdavies1@cox.net

Important Note: This paper has not been peer-reviewed, so use the information with caution.

ABSTRACT (submitted to Central Iowa NWA Severe Storms Conference 2001)

Rasmussen and Blanchard (1998, hereafter RB1998)) and Edwards and Thompson (2000, hereafter ET2000) have looked at a variety of parameters associated with supercell and tornado environments. The former study dealt with actual upper air sounding observations, while the later study used  model soundings.   Because model analysis and forecast soundings are widely available to most operational forecasters, a large dataset of simple forecast soundings updated by current surface data is examined in this study. A total of 1371 forecast profiles from FD data (Davies 1993, 1998a) associated with both supercell and non-supercell events was accumulated by the author during the period 1992-2000.  While these soundings are inferior in resolution and quality when compared to the RUC-2 analysis soundings used in ET2000, the large size of the dataset makes for a potentially useful examination.  It is suggested that if significant signals and patterns can be found using these simple soundings, results may be extrapolated and refined using better quality model data.   This study will use 987 supercell cases from the author's FD dataset to look at parameters examined by ET2000 and RB1998, including  height of lifting condensation level (LCL), vertical shear, and combinations of CAPE and vertical shear.  Results will be compared to ET2000 and RB1998, as well as a subset of overlapping cases from the RUC-2 dataset used in ET2000.  Similarities and differences will be noted.   Some additional parameters that focus on the lower portion of the thermodynamic profile, such as height of the level of free convection (LFC), and CAPE in the lowest 3 km of the profile, will also be examined for potential application.

1.  Introduction and dataset background

The purpose of this study is to document parameter tendencies and results  from a large dataset of simple forecast soundings associated with a variety of severe weather events.  Many of the same parameters associated with supercell and tornado forecasting  in studies by Rasmussen and Blanchard (1998, hereafter RB1998), and Edwards and Thompson (2000., hereafter ET2000) will be examined.   It should be useful to see if some of the results from these studies are duplicated and reproduced with a large dataset of simplified forecast soundings, and to see if some of the results can be extrapolated to more detailed forecast soundings, such as the RUC-2 profiles. 

The author started saving selected FD (Forecast Diagnostic) profiles in 1992 (Davies, 1993) with the goal of building a large set of forecast soundings associated with various types of severe weather.  The idea was not to try to replace actual sounding observations, but to see if any useful parameter patterns could be discerned from simple forecast soundings when actual sounding observations were not available.  It is also true that a larger dataset can be obtained more quickly with model soundings (see ET 2000) rather than observed soundings, which are limited by availability and the coarseness of the upper air observation network.   FD profiles (e.g., Woodall 1989) were used to build the dataset because, unlike other point forecast soundings in the early and mid 1990s, these were readily available through weather information services and eventually the Internet.  The small file sizes made these profiles easy to archive, and provided literally hundreds of cases from which point soundings or mapping fields of specific parameters could be generated.  In recent years more detailed soundings, such as the RUC and RUC-2 profiles,  have become widely available on the Internet, rendering the coarser FD profiles largely obsolete. Nevertheless, because of the large sample size, this dataset of FD profiles should yield some useful information. 

From 1992 through 2000,  the author accumulated 1371 severe weather cases, each with an associated FD profile in or near the warm sector within 60 miles and 2 hours of a severe weather occurrence to the east, southeast, or south of the case location.  Severe weather occurrences were generally separated by at least 2 hours or 100 miles in distance.  Therefore, for larger severe weather events occurring over a longer time period and larger area, multiple cases could be selected.   For most events, only 1 or 2 cases were selected, but there were a few events where as many as 4 or 5 cases could be selected from the same event according to the time/distance criteria.  Virtually all cases were in the eastern 2/3 of the United States. 

Surface observations uncontaminated by outflow were used to provide real boundary layer data for the profiles.   Because there is no moisture profile with the FD data, one is limited to using surface parcels as lifted parcels, and there is no information regarding moisture patterns and layers above the surface.  Other weaknesses are well known.   These include sparseness of detail, valid forecast times that are at best 6 hours apart, and the fact that these are simplified forecast soundings (from the NGM model, Davies 1993) that are subject to the problems and inconsistencies of model solutions.  To provide some degree of consistency, only 6 and 12 hr forecasts were used.

3.  Data categorization

Each case was associated with a "type" of severe weather, and categorized as supercell or non-supercell.  Because specific radar data was not archived or available for many of the cases, particularly before the Internet came into wide use, supercells were distinguished in several ways.  First, if radar data was available and significant rotation was indicated for the storm in question, either explicitly from velocity data or suggested from the reflectivity pattern, the case was categorized as a supercell.  Second, if radar data was not available, any of the following was used to categorize a storm as a supercell:  a) a tornado or severe warning mentioning a radar "indication" of rotation,  b) a reliable storm chaser visual report of a supercell, c) a hail report of 2.00" or greater as in RB1998.   Storms not falling into these observations were categorized as non-supercell.  This methodology is far from perfect, but should result in the majority of storms being categorized in a workable and reasonably representative manner.

Local storm reports and Storm Data were used to categorize storms within supercell and non-supercell groups as follows:

The most common error in categorization is likely to be the misclassification of a supercell as a non-supercell due to lack of information.   Conversely, the conservative methodology used here should be reasonably accurate regarding whether a storm classified as a supercell is actually a supercell.  This study will deal exclusively with the supercell portion of the FD dataset (987 cases) in order to make comparisons with the afforementioned studies by RB1998 and ET2000.

4.  FD profile compatibility with RUC-2 profiles and observed soundings

To attempt to address the question of how "accurate" or "compatable" the coarse FD soundings are in general when compared to more detailed RUC-2 analysis soundings and actual sounding observations, the RUC-2 supercell dataset covering 1999 through 2000 was generously provided by the authors of ET2000.  Their RUC-2 dataset was examined for overlapping cases with the author's  FD dataset.  If an FD case profile was within 50 miles and 2 hours of a RUC-2 case profile, the RUC-2 case was selected for comparison using the same surface data and elevation as the corresponding FD case.  This overlap between the two datasets resulted in 146 profiles for comparison regarding computation of parameters.

Computation differences from the methodology of ET2000 involved the use of estimated (rather than obseved) supercell motions based on environmental winds (Davies 1998b) using a method similar in results to Rasmussen and Straka (1998).  In order to be compatible with the FD soundings, only surface-based lifted parcels were used in the RUC-2 parameter computations.  Regarding CAPE computations, the virtual temperature correction (Doswell and Rasmussen 1994) could not be used directly for the FD dataset because of the lack of moisture data in the FD profiles.  A workable solution was found by using the scatterdiagram in Fig. 3 from Doswell and Rasmussen to estimate an average correction based on amount of uncorrected CAPE.  The virtual temperature correction to CAPE was computed directly for the RUC-2 profiles.

It should be noted that if the selected RUC-2 case analysis surface temperature was less than the actual surface temperature (ET2000 noted a slight cool dry bias in many RUC-2 soundings), the bottom 150 mb of the profile was automatically "warmed" and interpolated to smooth the actual surface temperature into the profile.  This was done to provide a realistic temperature profile in the lowest part of the sounding, avoiding unrealistic and unrepresentative situations where the computed LFC would otherwise be at ground level. 

Next, these 146 cases were examined for proxmity in time and space to actual radiosonde soundings.  The 50 mile/2 hour rule used previously yielded a subset of 22 observed sounding cases within the 146 cases.  While this subgroup is small, it nevertheless serves as a direct first  point of reference and comparison between the two datasets.

Below is a comparison of both FD and RUC-2 mandatory level temperatures at 850, 700, 500 and 300 mb for the 22 observed soundings.

<--FD/RUC-2 comparison with observed sounding temperatures

As would be expected,  the more timely and detailed RUC-2 analysis profiles are more accurate.   But the FD forecast profiles are surprisingly good in comparison, and actually "outperform" the RUC-2 profiles at 300 mb in this small subset.

Next  is a comparison of common thermodynamic parameters (sfc-CAPE, sfc-CIN, sfc-SLI as in Thompson and Edwards 2000):

<--FD/RUC-2 comparison with observed sounding thermodynamic parameters

Again the RUC-2 profiles are better in accuracy, but regarding CAPE and CIN, the FD profiles are reasonably close.

This third group is a comparison of common wind-related parameters (deep layer shear, SRH, and s-r flow as in Thompson and Edwards 2000):

<--FD/RUC-2 comparison with observed sounding wind parameters

The RUC-2 outperforms the FD profiles here except in the case of BRN shear.  Yet again, the FD profiles are surprisingly close to the RUC-2 profiles in accuracy.  The poor performance of both profile types regarding s-r winds at 300 mb probably suggests that this parameter should be averaged through a broad layer rather than measured at a fixed-point level, as done with the computations here.

Below is a comparison of errors (mean and mean absolute) similar to Thompson and Edwards (2000) regarding the 22 observed soundings in common between the two datasets:

<--FD/RUC-2 errors (comparison with observed sounding temperatures & parameters)

As expected, the RUC-2 data and parameters are more accurate in the mean, but the FD errors with most parameters are actually reasonably close to the RUC-2 errors.  For example, the FD temperature absolute errors are generally within .5 deg C of the RUC-2 absolute errors, and apart from the 0-6 km shear, the wind and thermodynamic absolute errors are quite close as well.

Finally, a comparison of mean differences and mean absolute differences between FD and RUC-2 parameters for the 146 cases in common between the two datasets is shown below.  This is less useful because we don't have actual soundings to compare to, but the larger sample size makes this worth examining briefly.

<--FD/RUC-2 differences in temperatures & parameters

The mean absolute differences between the two datasets are relatively large with some parameters (e.g., CAPE and SRH), but the mean differences are considerably smaller for each parameter.  This could suggest that as a dataset of FD profiles becomes larger, many differences cancel out and resulting parameter values become more representative of those from a corresponding group of RUC-2 profiles.  As we will see in later parameter comparisons between datasets, this appears to be true.

5.  Parameter examination

Most of the parameters examined by RB1998 and ET2000 are examined here using the 987 supercell cases from the author's FD dataset, as well as the 146 supercell cases from ET2000 that overlap the FD cases.  Box and whisker plots based on rank and percentile are used to display results, as in RB1998 and ET2000.    Hatched boxes are bounded and labeled by the 25th and 75th percentiles, with labeled whiskers showing the full range of values in each category.  The median parameter value is marked and labeled within each box.

In general,  parameters examined here fall into one of three groups:  1)  parameters assessing thermodynamic aspects of the atmosphere;  2) parameters assessing wind-related factors; and 3) parameters combining both wind and thermodynamic factors.

a.  Parameters related to thermodynamic factors

Results from the FD dataset for non-tornadic, weak tornadic (F0-F1), and significant tornadic (F2-F5) supercell cases are shown below for two key thermodynamic parameters that were examined in ET2000, total CAPE and LCL height AGL.   LCL height in particular was found by both RB1998 and ET2000 to be relevant in tornadic and non-tornadic discrimination.   Added to these parameters are CAPE below 3 km AGLand LFC height AGL, which relate to unpublished findings from Rasmussen (1998) suggesting that low-level CAPE may be a useful discriminator between tornadic and non-tornadic environments.

<CAPE    <CAPE below 3 km    <LCL    <LFC    (FD cases)

The results for CAPE and LCL height are quite similar to those from RB1998 & ET2000.  Tornadic storms tend to have somewhat larger amounts of CAPE and lower LCL heights, although there is a good deal of overlap between categories.  Low-level CAPE (below 3 km) and LFC height were not discussed in RB1998 or ET2000, but the results here show significant separations in the distributions between categories for both parameters.  The boxes and medians in each category strongly suggest that tornadic storms have more low-level CAPE and lower LFC heights in their environment.

To more directly illustrate discrimination between non-tornadic and significant tornadic cases, the following plots of the same parameters drop out the weak tornado category, similar to some plots in ET2000:

<CAPE    <CAPE below 3 km    <LCL     <LFC   

In particular, the ability of low-level CAPE and LFC height to discriminate between these categories is suggested by showing little overlap of the boxes representing the middle 50% of data.  LCL height appears as a moderate discriminator, with some overlap between categories.

The RUC-2 profiles from ET2000 corresponding to FD cases were used to compute the same parameters.  RUC-2 results for the non-tornadic and significant tornadic cases regarding low-level CAPE, LCL height, and LFC height are shown below:

<CAPE below 3 km    <LCL      <LFC                        (RUC-2 cases)

The results for all three parameters are strikingly similar to those from the larger FD dataset regarding the distribution of the middle 50% of data.  This strongly suggests two things.  First, as suggested in the previous section, the results from the larger FD dataset appear largely representative of and compatible with results computed from RUC-2 profiles.  Second, the results reinforce the idea that low-level CAPE, LCL height, and LFC height parameters are useful tornado/no tornado discriminators. 

b.  Parameters related to wind factors

Results from the FD dataset for the non-tornadic, weak tornadic, and significant tornadic supercell categories are shown below for some common wind-related parameters used in severe weather forecasting.  These involve deep layer shear (0-6 km shear and BRN shear), low-level shear (0-1 km and 0-3 km SRH), and storm-relative flow aloft (0-3 km s-r winds and 300 mb s-r winds):

<0-6 km shear   <BRN shear    <0-1 km SRH    <0-3 km SRH

<3-6 km s-r wind             <300 mb s-r wind

The deep layer shear parameters exhibit little variation or discrimination between categories, the same result as found in RB1998.  The low-level shear parameters show little variation between non-tornadic and weak tornadic categories, but tornadic cases show generally larger values of helicity similar to results from RB1998 and ET2000.   Here, the 0-1 km SRH does not appear to be significantly better than 0-3 km SRH in separating the tornadic cases.  But it is quite possible that the lack of low-level wind detail in FD profiles may mask differences in data separation between the two layers.  For storm-relative flow aloft, as with deep layer shear, little variation is seen between categories, which is similar to results in Thompson and Edwards (2000).

From the two-category plots below, only storm-relative helicity shows some degree of ability among stand-alone wind parameters to discriminate between tornadic and non-tornadic cases in the FD dataset.    Even so, the SRH results exhibit a notable degree of overlap in the middle 50% boxes.

<0-6 km shear  <0-1 km SRH  <0-3 km SRH   <3-6 km s-r wind

Results from the corresponding RUC-2 cases for SRH in the same two categories are shown below, suggesting somewhat better discrimination, particularly in the 0-1 km layer:

<0-1 km SRH    <0-3 km SRH                     (RUC-2 cases)

The RUC-2 0-1 km SRH in particular shows more separation than in the FD dataset, which is more in line with results from Markowski et al. (1998).  As suggested earlier, the greater wind detail in low-levels of the RUC-2 profiles probably accounts for this.  However, the 0-3 km SRH from both the FD and RUC-2 profiles also shows moderate discrimination ability, and appears nearly as useful.   

It should be noted that the lack of tornadic vs. non-tornadic discrimination in the datasets regarding parameters such as deep layer shear and storm-relative flow aloft (not shown for RUC-2 cases) does not mean that these parameters have no utility.   For example, the earlier FD data plot for 0-6 km shear suggests that nearly 40 kts of shear or greater is neccessary to support most supercells, which is useful information as a limiting factor.  And in the next section, there is much to suggest that combinations of CAPE and shear together show significant ability to discrimate between tornadic and non-tornadic categories.

c. parameters combining wind and thermodynamic factors

RB1998 showed that combinations of wind and thermodynamic factors such as EHI (Davies, 1993) and VGP (RB1998) can be useful in discriminating between tornadic and non-tornadic environments.    Results from the FD dataset for non-tornadic, weak tornadic, and significant tornadic supercells are shown below for 0-1 km EHI, 0-3 km EHI, 0-3 km VGP, and a CAPE/deep layer shear comparison devised from Davies (1998a):

<0-1 km EHI  <0-3 km EHI  <0-3 km VGP  <CAPE/BRN shear

As in  RB1998 and ET2000, EHI shows notable separation between categories for both the 0-1 km and 0-3 km layers.  Neither layer is favored over the other.   VGP also exhibits notable separation between categories, similar to EHI.  In the CAPE/BRN shear comparison, negative values correspond to a parameter space in Davies (1998a) that appears to favor tornadic storms.  This untested parameter also appears to demonstrate reasonable ability to discriminate between categories in the FD dataset.

Dropping out the weak tornado category in the plots below emphasizes that all four parameters do a reasonable job of discriminating between non-tornadic and significant tornadic cases, with little overlap between the boxes that represent the middle 50% of data in each category:

<0-1 km EHI  <0-3 km EHI  <0-3 km VGP  <CAPE/BRN shear

This reinforces the results from RB1998 and ET2000 that CAPE/shear combinations have some definite utility in tornado forecasting.

Looking now at the corresponding RUC-2 cases, EHI and the CAPE/BRN shear comparison are shown below in two categories:

<0-1 km EHI    <0-3 km EHI     <CAPE/BRN shear             (RUC-2 cases)

The RUC-2 results are very similar to those from the FD dataset, with generally good separation between non-tornadic and tornadic cases for all the CAPE/shear parameters.  It may be important to note that the 0-1 km EHI boxes show more overlap than do those for 0-3 km EHI, suggesting that 0-1 km EHI is not necessarily a better discriminator than 0-3 km EHI.  This differs somewhat from the results in ET2000.  But in general terms, it certainly appears that parameters combining CAPE and shear do better than most stand-alone wind or thermodynamic parameters regarding tornado/no tornado discrimination.

6.  Discussion

From the results and data plots above, it appears that very useful information can be gleaned from the large FD forecast sounding dataset examined in this study, despite the fact that these simple forecast soundings are sparse in detail   The surprising similarity between the FD and the RUC-2 parameter data distributions, particularly regarding the middle 50% of data in each supercell category, is very encouraging.  This strongly suggests that many of the examination results from the large FD dataset can indeed be extrapolated to RUC-2 analysis and forecast profiles, which are certainly preferable in operational use.  This dataset also does much to augment and reinforce findings by RB1998 and ET2000.

No attempt was made to rank parameters from this study according to skill, largely because of questions about the detail and representativeness of FD data in individual cases compared to the observed soundings used in RB1998.   But from a subjective examination of the parameter data distribution plots from both the FD and RUC-2 datasets, the following parameters (in no particular order) appear to provide the best potential for discrimination between non-tornadic and tornadic environments:

•             EHI (using either 0-1 km or 0-3 km SRH)

•             Low-level CAPE

•             LFC height

•             LCL height           

•             experimental CAPE/BRN shear comparison

Each of these will be briefly discussed below.           

EHI (see RB1998, Davies 1993) has a physical basis in the tilting and stretching of streamwise horizontal vorticity (related to SRH) by a strong updraft (related to CAPE).  While the result of this process is storm rotation (supercells) and the formation of midlevel mesocyclones, the work of Markowski et al. (1998) and others suggests that strong SRH in the lowest levels also has something to do with increasing potential for tornadoes, possibly via the genesis of low-level mesocyclones.  Although the mechanisms are certainly unclear,  this study reinforces that there appears to be some sort of relationship between EHI in low-levels and tornadoes. 

From the standpoint of using forecast soundings operationally, this study suggests there may not be a great deal of difference between the 0-1 km and 0-3 km layers regarding the effectiveness of EHI (and SRH) as a discriminator between non-tornadic and tornadic environments.  This is somewhat at odds with findings from an unpublished manuscript by Rasmussen (1998) updating RB1998, which found 0-1 km SRH and 0-1 km EHI to be more effective discriminators than through deeper layers, based on observed soundings.   While actual observations of low-level winds (radiosonde data, VAD wind data, profilers) can detect local variations and enhancements of near-storm and storm-influenced SRH in the lowest levels, forecast soundings cannot be expected to predict and detect the same.  This may account for the lack of preference in EHI and SRH layers tested in this study.

0-1 km EHI values above roughly 1.0 or 1.5 appear to be most relevant for suggesting supercell tornado potential.  For 0-3 km EHI, values approaching 2.0 and above appear to be most associated with supercell tornadoes, similar to Davies (1993).  This study also suggests that VGP may be a suitable substitute for EHI.  A potential advantage is that VGP is not subject to estimation or observation of storm motion.  0-3 km VGP values approaching .30 and above appear to be most relevant for suggesting supercell torando potential.

The unpublished follow-up study to RB1998 (Rasmussen, 1998) found that CAPE below 3 km AGLappeared to offer notable discrimination between non-tornadic and tornadic storms.  In the current study using forecast soundings, low-level CAPE likewise appears to discriminate rather well between environments.  This suggests that increased vertical stretching from upward parcel acceleration at lower levels due to the presence of significant CAPE closer to the ground may play a part in tornado generation.

Related to low-level CAPE, LFC height in the current study seems to suggest the same process and relationship, with tornadic storms generally associated with lower LFC heights in both the FD and RUC-2 datasets.  The low-level CAPE and LFC results appear important, and reinforce similar unpublished findings in Rasmussen (1998).    Values of CAPE below 3 km AGL greater than roughly 60 j kg-1 appear most relevant for suggestig tornadic supercells, while values above 90 j kg-1 should be given particular attention based on this study.  LFC heights below 2000 m AGL appear to be most associated with supercell tornadoes, while heights below 1600 m are even more suggestive of tornadoes.  More research about these parameters is needed and suggested.

LCL height was found to be relevant as a discriminator in both RB1998 and ET2000.  This study reinforces that relevance, with the tornadic cases tending to have lower LCL heights.  This shows up best in the RUC-2 profiles, although the FD profiles have a similar signal.  The importance of LCL height is thought to relate to sub-cloud evaporation and the potential for outflow dominance.  Low LCL heights imply less evaporational cooling from precipitation and therefore less potential for strong outflow that would likely inhibit low-level mesocyclone development.   Supercell tornadoes are limited mostly to LCL heights below 1500 m AGL, as noted in ET2000.  But it appears that the strongest tornadoes are associated with LCL heights below 1000 m AGL.

Finally, the CAPE/BRN shear comparison in this study also shows effectiveness as a tornadic/non- tornado discrimator, even though deep layer shear by itself does not.  This experimental comparison relates to the interaction of deep layer shear with updraft that has been shown to significantly enhance upward parcel velocities beyond the vertical pressure gradient accelerations expected from buoyancy alone (Rotunno and Klemp 1982, and McCaul 1991).  Parameters combining CAPE and shear have focussed mainly on low-level shear and the tilting of low-level vorticity, rather than this deep layer interaction that directly strengthens updraft and the related potential for strong vertical stretching.  Therefore, a parameter combining CAPE and BRN shear (in a different way than BRN) may contribute useful information regarding tornado potential additional to that provided by low-level parameters such as EHI. 

Examples in support of this idea come from the FD and RUC-2 datasets in this study.  The box graphs below show percent of non-tornadic, weak tornadic, and significant tornadic supercells that fall outside a "favorable" parameter space for tornadoes suggested in the study by Davies (1998a):

<CAPE/BRN shear (FD cases)        <CAPE/BRN shear (RUC-2 cases)

More than 1/2 the non-tornadic supercells in the FD dataset fall outside the parameter space, while less than 15% of the significant tornadic supercells fall outside the same parameter space.  In the smaller RUC-2 dataset, the results are similar but a little less pronounced.  This discrimination suggests a need for a CAPE/deep layer shear parameter relevant to supercell tornadoes.  Indeed, within the non-tornadic FD supercell category, there are 94 cases (not shown) that have favorable 0-3 km EHI values (> 2.0) yet appear limited by "unfavorable" CAPE /BRN shear values (falling outside the afforementioned parameter space).  More research is needed in this area.   

A word is in order about storm-relative flow aloft in midlevels (Thompson, 1998).  Although this study suggests no significant ability of this parameter by itself to discriminate between tornadic and non-tornadic cases (similar to Thompson and Edwards 2000),  the consistency of the s-r midlevel flow values for all supercells in this study appears important.   Because the data distributions suggest that s-r midlevel flow approaching 15 kts or greater is associated with the majority of  supercells, this in itself may be a useful limiting factor when considering potential for strong tornadoes.  Also, Thompson and Edwards suggest that better discrimination results may be acheived through combination of s-r winds with thermodynamic-related parameters, such as LCL height.

It is strongly suggested that use of the above parameters together in assessing and estimating pre-storm and near-storm environments can add significantly to information from radar and other sources in making warning decisions.  This is an area that is ripe for future research and documentation within an operational setting.

Future research with the FD dataset assembled here will focus on non-supercell as well as supercell events.

Acknowledgments.  Rich Thompson and Roger Edwards at the Storm Prediction Center are gratefully acknowledged for their willingness to share their supercell dataset of RUC-2 analysis soundings.

REFERENCES

Davies, J. M., 1993: Hourly helcity, instability, and EHI in forecasting supercell tornadoes. Preprints, 17th Conf. Severe Local Storms, St. Louis,                 Amer. Meteor. Soc., 107-111.
Edwards, R., and R. L. Thompson, 2000: RUC-2 supercell proximity soundings, part II: An independent assessment of supercell forecast                 parameters. Preprints, 20th Conf. Severe Local Storms, Orlando, Amer. Meteor. Soc., 435-438.
Rasmussen, E. N., and D. O. Blanchard, 1998: A baseline climatology of sounding-derived supercell and tornado forecast parameters. Wea.                 Forecasting, 13, 1148-1164.

additional references will be forthcoming