A Comparison of Rapid Update Cycle 2 (RUC-2) Model Soundings with
Observed Soundings in Supercell Environments
Richard L. Thompson and Roger Edwards
Numerous studies have used so-called “proximity” soundings to diagnose various aspects of severe storm environments, including Maddox (1976), Davies and Johns (1993), Kerr and Darkow (1996), Rasmussen and Blanchard (1998), and Bunkers et al. (2000). However, observed soundings in near-storm environments present at least a couple of problems. First, there are questions regarding the most relevant location for a supercell proximity sounding. Supercells can exert an influence on the local vertical wind profiles, as discussed by Weisman et al. (1998), which can complicate the choice of a proximity sounding location. Second, supercells are essentially randomly distributed with respect to the observed soundings, making consistent data collection difficult at best. Based on discussion by Maddox (1976), and the personal experiences of the authors, an exceptionally long period of time must pass before a sufficient number of “representative” soundings can be collected. Additional discussion of problems associated with proximity soundings can be found in Brooks et al. (1994).
Thompson (1998) attempted to circumvent storm induced sounding effects and small sample sizes of observed proximity soundings by using Eta model analysis grids. These grids do not incorporate sub grid scale processes associated with individual supercells; however the grid point data can be “contaminated” by model generated precipitation during the model assimilation time period, or by direct observations within precipitation areas at the 0-hour (hereafter 0-h) model start time. A primary advantage to use of model analysis grids is the collection of a much larger sample of storm cases in a shorter period of time, since fortuitous positioning of an observed sounding is not necessary for a particular case. However, model analysis grids must be consistent with observed data if they are to serve as a diagnostic tool, and the analysis grids must be available frequently in time so that the majority of supercell cases can be considered. An ideal model analysis frequency of once per hour allows model initialized “proximity” soundings for almost every supercell observable in the model domain, conforming to the time and space guidelines of almost any previous study.
For these reasons, the second generation Rapid Update Cycle model (RUC-2, Benjamin et al. (1998)) was chosen for this study. RUC-2 analysis grids are available at the Storm Prediction Center on an hourly basis, along with nationwide WSR-88D reflectivity and velocity imagery (0.5o and 1.5o elevation angles) to identify supercells. The purpose of this study is to compare RUC-2 analysis soundings with observed soundings within the same general thermodynamic environment as observed supercells, and to make recommendations for additional proximity sounding studies based on the RUC-2 analyses.
Soundings were generated at the SPC from RUC-2 model analysis grids for supercell cases during the time period from 3 March 1999 to 18 May 2000. To be included in this study, the closest rawinsonde sounding to each supercell case must not have been obviously contaminated by precipitation (e.g., deep saturated profiles), or convective outflow (e.g., low-levels substantially cooler than nearby observations away from the supercell). If the closest observed sounding was contaminated by these effects, or if the observed sounding was truncated below the equilibrium level, then no RUC-2 comparison soundings were collected for that particular supercell case. Also, all observed comparison soundings displayed surface-based CAPE (e.g., so-called “elevated” supercells were not included in this study). Based on these criteria, a total of 51 observed and 0-h RUC-2 soundings were collected and compared. The majority of the comparison soundings (48) were from 00 UTC, while only three were from 12 UTC. In addition, a smaller sample (24) of 1-h RUC-2 forecast soundings valid at 00 or 12 UTC were collected to evaluate the accuracy of short-term RUC-2 forecast soundings from model runs that did not include rawinsonde data.
3. COMPARISON RESULTS
The basic findings of this study are presented in Table 1. The RUC-2 analysis soundings where characteristically 1-2o C too dry at the surface and 850 mb, as well as 1-2o C too cool at the surface and too warm at 850 mb, when compared to the observed soundings at the same time and location. This minor cool and dry bias of the RUC-2 soundings at the surface, combined with the warm bias at 850 mb, contributed to a tendency for convective inhibition to be somewhat overestimated in the mean, and surface-based CAPE to be underestimated by 500-1000 J kg-1 . Temperature errors in the middle and upper troposphere were substantially smaller than in the lower troposphere (as shown in Table 1a, and inferred by the small 500 mb lifted index errors in Table 1c), which indicated that most of the CAPE errors were the result of a surface cool and dry bias of the RUC-2 analyses. There is the potential for some of the surface errors to be the result of differences between the RUC-2 surface elevations/pressures and those of the co-located observations, as well as due to interpolation differences in the RUC-2 soundings constructed from grids of 25 mb “vertical” resolution. We attempted to minimize such errors by choosing the “surface” as the observed significant level observation closest to the RUC-2 surface pressure, with typical pressure differences of 5-10 mb.
Table 1. Summary of RUC-2 analysis errors for a) mandatory pressure level temperatures, b) mandatory pressure level dewpoints, c) derived thermodynamic parameters, and d) derived vertical shear parameters. All storm-relative calculations were based on the observed supercell motion for each case. The first row displays the mean arithmetic errors, and the second row shows the mean absolute errors. (Error defined as RUC-2 values minus observed values).
As with surface temperatures and low-level dewpoints, the majority of the vertical shear parameters were slightly underestimated in the RUC-2 analysis soundings. Parameters such as 0-3 km storm-relative helicity (SRH) and the Bulk Richardson Number (BRN) shear term (Weisman and Klemp, 1982), which incorporate low-level details from the hodograph, were most sensitive to small variations in the compared wind profiles. The mean absolute analysis errors for the BRN shear term and 0-3 km SRH were substantial (~ 20 m2s-2 and 75 m2s-2, respectively), though small negative mean errors suggested only a slight tendency for the RUC-2 analysis hodographs to consistently underestimate low-level vertical shear. The simple 0-6 km wind vector difference showed even less difference between the observations and RUC-2 analyses, with a mean absolute error of 3.1 ms-1, and a mean error of only -0.6 ms-1. Interestingly, the mean and absolute errors were not particularly large for the storm-relative winds in the mid and upper troposphere (Table 1d.)
Figure 1. Box and whisker plot of surface-based CAPE errors from the sample of 51 0-h RUC-2 analysis soundings, and 24 1-h forecast soundings. The ends of each bar mark the extreme errors, the top and bottom of each shaded box denotes the 25th and 75th percentiles, respectively, and the horizontal line within each box denotes the median error.
It appears that the RUC-2 analysis soundings are a reasonable representation of the observed data, especially in the mean as sample size increased. Mean and absolute errors decreased for all parameters in this evaluation as the sample size increased from a preliminary sample size of 30 to the current sample size of 51 cases. However, the accuracy of the RUC-2 soundings can certainly be questioned in individual cases. For example, surface-based parcel CAPE errors spanned a range of +1488 Jkg-1 to -2348 Jkg-1 (Fig. 1). These extreme CAPE errors were attributed primarily to a surface dewpoint temperature overestimate of 2.4o C, and an underestimate of 4.1o C (Fig. 2), in the respective RUC-2 analyses. The majority of the errors involved underestimates of buoyancy, especially in cases where observed buoyancy was relatively small (e.g., < 1000 J kg-1). In such “weak instability” cases, the RUC-2 analysis soundings occasionally displayed surface-based CAPE values near zero, which suggested that supercells were unlikely. However, the observed soundings did reveal sufficient buoyancy for deep convection. Hence, the RUC-2 soundings must be used with caution in low buoyancy environments, or when buoyancy profiles for individual cases are examined quantitatively. More reliable estimates of surface-based CAPE can be achieved by modifying the RUC-2 soundings for nearby surface observations, as long as RUC-2 surface elevation/pressure is accurate. Operational forecasters may also find it advantageous to use the 100 mb mean parcel CAPE, since errors in this parameter were substantially less than with surface-based CAPE.
Figure 2. Same as Fig. 1, except for surface dewpoint temperatures.
Similar tendencies were noted with the derived vertical shear parameters. The largest errors occurred in the lowest 1 km, with the BRN shear and SRH (Fig. 3) estimates most adversely affected. Virtually all of the mean and absolute errors in 0-3 km SRH occurred in the lowest 1 km, with negligible errors above 1 km. Much of the error in the BRN shear estimates may be associated with differences in the near-ground winds between the observations and RUC-2 analyses. These errors are not surprising in light of the spatial and temporal variability of SRH documented by Markowski et al. (1998) in supercell environments. Unfortunately, the authors are not aware of any consistent means to account for the RUC-2 tendency to underestimate low-level vertical shear, given the coarse spatial resolution of the profiler and WSR-88D velocity-azimuth display wind profiles. Therefore, any attempts to discriminate in the mean between tornadic and nontornadic supercells with RUC-2 SRH estimates should require a large sample size to support any robust conclusions, and may require a probabilistic approach for individual cases. In contrast, mean error for the 0-6 km shear vector magnitude was only -0.6 m s-1, and the mean absolute error was only 3 m s-1. As a result of this analysis, the RUC-2 soundings may be best suited to operational diagnosis of supercell environments based on 0-6 km shear values.
Figure 3. Same as Fig. 1, except for 0-3 km storm-relative helicity.
a. RUC-2 forecast soundings
The hourly RUC-2 analyses allow much flexibility in case selection, since analysis soundings are available near the times and locations of virtually all supercells. However, there are no direct means of evaluating the RUC-2 analysis soundings for the asynoptic times because complete temperature, moisture, and wind profiles are generally not available at any one site outside of the observed 00 and 12 UTC soundings. One means of addressing this problem is to compare the observed soundings to 1-h forecast soundings from the hour prior to the standard rawinsonde times. The 1-h forecast soundings from the 11 and 23 UTC RUC-2 model runs do benefit from various asynoptic wind and temperature data (profiler, aircraft observations, etc.), though these observing platforms do not provide detailed vertical profiles of moisture. Since the 11 and 23 UTC model runs are the farthest removed in time from standard rawinsonde data, it could be argued that these forecast soundings should have thermodynamic errors as large as any during the day. Also, the majority of differences between the 1-h forecasts and the observed soundings should be attributable to differences in the initial model fields instead of model forecasts, given that it is only a 1-h forecast. Therefore, the 1-h forecast soundings from 11 and 23 UTC valid at 12 and 00 UTC, respectively, can be used to assess the accuracy of RUC-2 soundings at asynoptic times.
Forecast errors for the 1-h RUC-2 forecast soundings (not shown) valid at either 12 or 00 UTC were similar to the 0-h errors. Surprisingly, the mean errors for most parameters were slightly smaller than for the larger sample of 0-h soundings in Table 1. However, the mean absolute errors are larger for the 1-h soundings, as inferred by the larger boxes for the 1-h forecasts versus the 0-h analyses in Figs. 1-4. Regardless, the differences between the 0-h soundings (which include rawinsonde data in the analyses) and the 1-h soundings (which do not have rawinsonde data) are not particularly large. This suggests that the asynoptic RUC-2 analysis soundings can be used to compare supercell environments in a quantitative manner, given large sample sizes.
Based on a limited sample size of 51 soundings in supercell environments, RUC-2 analysis soundings appeared to be reasonably representative of observed soundings at the same locations. The RUC-2 soundings had a tendency to underestimate surface temperatures and dewpoints, with resultant surface-based CAPE values consistently underestimated by 500-1000 Jkg-1. Vertical shear parameters that are very sensitive to variations in low-level winds, such as 0-3 km SRH and BRN shear, were likewise substantially underestimated. However, mid-tropospheric temperature profiles and deep layer vertical shear parameters were reliably represented by the RUC-2 analyses. A ramification of these results is that the RUC-2 analysis soundings may be suitable as surrogates for observed “proximity” soundings in severe storm environments, with the caveat that large sample sizes are utilized for quantitative comparisons between, for example, tornadic and nontornadic supercells. Operational diagnosis of supercell environments and discrimation between supercell types, in specific cases, may be best suited to some sort of probabilistic approach.
Benjamin, S. G., J. M. Brown, K. J. Brundage, B. Schwartz, T. Smirnova, T. L. Smith, L. L. Morone, and G. Dimego, 1998: The operational RUC-2. Preprints, 16th Conf. on Wea. Analysis and Forecasting , Pheonix, AZ, 249-252.
Brooks, H. E., C. A. Doswell III, and J. Cooper, 1994: On the environments of tornadic and nontornadic mesocyclones. Wea. Forecasting, 9, 606-618.
Bunkers, M. J., B. A. Klimowski, J. W. Zeitler, R. L. Thompson, and M. L. Weisman, 2000: Predicting supercell motion using a new hodograph technique. Wea. Forecasting, 15, 61-79.
Davies, J. M., and R. H. Johns, 1993: Some wind and instability parameters associated with strong and violent tornadoes. Part I: Wind shear and helicity. The Tornado: Its Structure, Dynamics, Prediction, and Hazards, Geophys. Monogr., No. 79, Amer. Geophys. Union, 573-582.
Kerr, B. W., and G. L. Darkow, 1996: Storm-relative winds and helicity in the tornadic thunderstorm environment. Wea. Forecasting, 11, 489-505.
Maddox, R. A., 1976: An evaluation of tornado proximity wind and stability data. Mon. Wea. Rev., 104, 133-142.
Markowski, P. M., J. M. Straka, E. N. Rasmussen, and D. O. Blanchard, 1998: Variability of storm-relative helicity during VORTEX. Mon. Wea. Rev., 126, 2959-2971.
Rasmussen, E. N., and D. O. Blanchard, 1998: A baseline climatology of sounding-derived supercell and tornado forecast parameters. Wea. Forecasting, 13, 1148-1164.
Thompson, R. L., 1998: Eta model storm-relative winds associated with tornadic and nontornadic supercells. Wea. Forecasting, 13, 125-137.
Weisman, M. L., and J. B. Klemp, 1982: The dependence of numerically simulated convective storms on vertical wind shear and buoyancy. Mon. Wea. Rev., 110, 504-520.
_____, M. S. Gilmore, and L. J. Wicker, 1998: The impact of convective storms on their local environment: What is an appropriate ambient sounding? Preprints, 19th Conf. On Severe Local Storms, Minneapolis, MN, Amer. Meteor. Soc., 238-241.