Correlating Lightning with Critical Values of Convective Indices in New Mexico

David Hefner 
National Weather Service
El Paso, Texas

Abstract

Thunderstorms (and more specifically, lightning) in and around El Paso, Texas display some common characteristics when relating widely used convective parameters to the occurrence of lightning and thunderstorms. In fact this paper will show that certain critical values of these parameters appear to be valid throughout New Mexico, much of eastern Arizona, and far west Texas.

The thunderstorm season from the beginning of June 2002 to mid October 2002, and approximately 45 separate thunderstorm days in this period, were examined to correlate lightning occurrence with such convective parameters as Total Totals Index, Lifted Index, K-Index, and Vertical Totals Index. The indices were not derived from actual observed data, but rather from graphical output available on AWIPS from the ETA model.

Two times, 00Z and 06Z (6pm and midnight local time) were used in the correlation, along with a few cases at 12Z (6am local time). In addition to using the 00Z run from the ETA model, in order to give this study somewhat of a operational utility, the 12Z run of the ETA model from the morning of the events was also incorporated into the correlation against the indices and the 00Z run, since the daytime forecasters would only have access to this specific run.

Methodology

A total of 47 thunderstorm days were examined from June 6 to October 27 of 2002. All days were chosen basically at random, although availability of data and the amount of lightning activity obviously went into the selection process. Although 47 cases were selected not all data parameters total 47 due to missing model data, no lightning activity, etc. Convective parameters used in this research are Total-Totals, K-Index, Lifted Index, and Vertical Totals. From a plan view some subjectivity was used in determining the critical, or minimum value for each of these parameters. I looked at the minimum value where from 95% or greater of lightning strikes occurred. I was not interested in finding a value which included every last stroke of lightning. In other words I ignored some of the extreme outliers.

The specific lightning data used was the one-hour lightning data within the time period of one hour leading up to 00Z (23-00Z) and either the hour leading up to 06Z (05-06Z), or in a few cases, the hour before that (04-05Z). The ETA model was used in this study and the data was displayed on a plan view via AWIPS. The 12Z run was selected because it is the most operationally significant run, that which is used by forecasters for that afternoon and evening's forecast package. The 00Z run was also included for comparison, the assumption being that the 00Z run may contain more accurate figures due to the short forecast time, while the 12Z's longer forecast time would possibly reflect different values.

I chose model data because observed data is either too sparse spatially (upper air data), or too sporadic temporally (missing GOES sounder data due to clouds, for example). And in an operational sense using model data to forecast future lightning/thunderstorms is a quick and hopefully accurate method.

Results

Upon completion of compiling this data, one of the most obvious observations was that the Total Totals index (TT), and Lifted Index (LI) were the best convective parameters out of the four that I tested. Both from a subjective inspection of lightning overlayed on the respective contoured fields, and the more objective look at the standard deviations to get a grasp at how each parameter varied, it was apparent that TT and LI were the best parameters to correlate to lightning strikes. The K-Index (KI), and Vertical Totals (VT), were somewhat less reliable, displaying significantly higher values of variance.

Here is a summary of the statistics with some more detailed statistics to follow.

Entire season: (47 cases)

An examination of this table shows that indeed TT and LI had a pretty good correlation with respect to the deviation of the critical value being low, while the deviation of the critical value for KI and VT were noticeably higher. The surprising aspect of this table is with almost every case of a different valid time, the old run (12Z) appeared to be more consistent with less variability than the new run (00Z).

I broke this overall picture down further into two other groups; one for "dry" season thunderstorms, and one for "wet" season thunderstorms. I picked the season from July 1 to early September for the wet/monsoon season storms because the wet season ended fairly abruptly a few days into September. The rest of the thunderstorms on either side of the period were included in the dry season thunderstorms. Here are those statistics broken down:

Wet season: (29 cases)

Dry season: (18 cases)

There was not too much difference between the two seasons with respect to the critical values. It does appear that there was more deviation from the critical values during the dry season than during the wet season. As a subjective observation while compiling this data, this fact may be explained by the flat gradient and higher values of the parameters during the wet season over the sharper gradient and relatively lower numbers of the parameters during the dry season. Though this observation may make the dry season a little more variable and somewhat less efficient as far as finding a reliable critical number for the parameters, this apparent fact also makes the dry season a little better for forecasting as far as limiting the convective area spatially.

Summary

The major goal of this study was to not to find predictors of thunderstorms using these various parameters; rather it was to find critical values of convective parameters that will limit thunderstorm activity spatially. In other words, once the forecaster has determined that thunderstorms are possible due to synoptic conditions such as large moisture influxes, etc., and/ or mesoscale influences such as boundaries, surface convergence, etc., then these critical values found in this study can be used to show the limits of the area that these thunderstorms are likely to occur in.

In this sense the pattern during the dry season exhibits a significantly sharper data gradient, therefore a better spatial limiter of thunderstorms. Meanwhile, during the wet season this study showed that although the critical values of the parameters were essentially the same, the data distribution displayed a much flatter gradient. This results in a less less efficient method for showing the limits of the thunderstorm area, at least on a WFO scale. It is still pretty useful when examining a larger area across two or three states.

Finally, this study was not exhaustive, though it has a good sample number in it. As a forecaster, use these numbers and see if you find they can be tweaked or are useful as is. I compared the morning model run to the evening run in order to see if the results were the same. It appears that one can have confidence in the morning run in real time since it is as good or perhaps even better than the evening run. Again, see if this pattern holds up through your daily forecasting.

As a side note, a few contour maps of the typical weather pattern for the dry and wet season have been included from various days

Figure 1 (dry season)
Figure 2 (dry season)
Figure 3 (wet season)

Acknowledgement

Thanks to El Paso National Weather Service Science and Operations Officer, Val MacBlain, for his assistance in preparing this paper.