SR/SSD 97-28
6-15-97

Technical Attachment

USING GRIDDED DATA TO FORECAST A
LATE SUMMER HEAVY RAINFALL EVENT
OVER SOUTHWEST TEXAS

Monte C. Oaks

NWSO San Angelo, TX

1. Introduction

The mesoscale convective system (MCS) has drawn much attention over the past couple of decades as a leading producer of flash floods across the United States. Numerous studies of these systems have resulted in a better understanding of model biases, which has translated into drastic improvements in skill scores by forecasters. This is especially true in the southern United States where numerical modeling often fails to accurately depict crucial mesoscale features such as ridges of equivalent potential temperature (e ).

As one moves toward the equator, synoptic forcing mechanisms become weaker, and mesoscale features play a much larger role. Models have particular difficulty with mesoscale features in southwest Texas since data for model initialization over the Mexican Plateau and the Gulf of Mexico are more sparse. Nevertheless, with the use of gridded data on personal computers (PCGRIDDS), we are able to maximize the potential of each model. With the use of gridded data, it is possible to find key features from synoptic scale models which can result in dramatic improvements in quantitative precipitation forecasting for late summer MCS events in southwest Texas. This study describes one such case.

2. Model data

Studies from Mesinger (1996) have shown the Early Eta model to be the superior model, compared with the Nested Grid Model (NGM) and the Aviation (AVN) model, when forecasting deep moist convection. In fact, his studies found the quantitative precipitation forecasts from the Eta model outperformed those of the NGM by a factor of two for rainfall events of two inches or more. During the period in which his studies were made, the Early Eta horizontal grid spacing was decreased from 80 to 48 km (on October 12, 1995), giving the Early Eta the best horizontal resolution of the three operational synoptic scale models.

The Early Eta also has the best low-level vertical resolution of the three models, since the highest concentration of its 38 levels are found in the boundary layer. According to Mesinger (1996), NGM development has undergone a "freeze" since 1991, resulting in no recent changes in model performance. Thus, the NGM will provide a useful reference as we see continued developments with the Early Eta.

3. The West Central Texas flash floods of August 28-29, 1996

On the afternoon of August 28, 1996 an upper-level disturbance interacting with abundant low- and mid-level moisture sparked torrential rains in West Texas which lasted until around 1900 UTC on August 29. Flooding began around 2230 UTC on the 28th over Nolan and Taylor Counties, where up to 13 cm (5 in) of rainfall accumulated within a few hours. This caused numerous automobiles to become stranded and some people were rescued from their vehicles.

This area of thunderstorms gradually spread south during the evening producing flooding rains over Runnels, Irion, and Tom Green Counties through 0500 UTC August 29. Southward movement of the storms began to slow during the early morning hours, and by 1000 UTC, an area of nearly stationary thunderstorms produced flooding over parts of Sutton and Kimble Counties. Up to 15 cm (6 in) of rain fell in less than five hours. Conditions further deteriorated by late morning as thunderstorms redeveloped over Irion and Tom Green Counties, shortly afterward causing serious flooding problems. Highest rainfall totals were found in southern portions of these counties with between 16 and 18 cm (6.5 to 7 in) of rain reported. A few houses were flooded and residents along Spring Creek in the community of Mertzon were briefly evacuated from their homes.

4. Brief analysis

An analysis of various levels from 1200 UTC August 28 revealed a strongly diffluent upper-level pattern across West Texas and a low-level shear axis denoting convergence over northwest Texas prior to the event. Although the 200 mb winds (Fig. 1) were fairly light over west-central Texas with speeds around 15 ms-1 (30 kt), a subtle wind speed maximum was approaching from the northwest with 28 ms-1 (55 kt) measured over central New Mexico. Meanwhile, at 500 mb (Fig. 2) a weak shortwave trough was located from eastern Colorado extending south to far West Texas. The 1200 UTC analysis also revealed that the eastern portion of the subtropical ridge over the southwestern United States was weaker over the Southern Plains, allowing for the shortwave to propagate southeast toward north Texas.

Low-level moisture was abundant over much of West Texas with a broad area of 850 mb dewpoints at or above 16 C (Fig. 3) and the 1200 UTC surface analysis (Fig. 4) showed dewpoints ranging from the middle 60s to the middle 70s. The 1200 UTC surface map also indicated a trough from south-central Oklahoma extending southwest to far West Texas, with cloudy skies and intermittent light rain occurring north of the surface boundary. Initialization of this low-level feature would prove to be the most critical factor in model performance.

5. Comparison of NGM and Early Eta Model from the 1200 UTC 28 August 1996 run

The most representative level on PCGRIDDS to compare the model analyses with the 1200 UTC surface map was at 950 mb. The NGM and Early Eta analyses at 1200 UTC both revealed the surface boundary over northwest Texas extending into south-central Oklahoma (Figs. 5 and 6). However, the 950 mb temperatures along the boundary were best correlated with the Eta model, as the NGM failed to indicate the slightly cooler temperatures caused by cloud cover and rain-cooled air over the Texas Panhandle.

Similar results were seen in the 1800 UTC comparison. By this time, the surface analysis (Fig. 7) indicated the boundary had pushed slightly northward and was located along the Red River Valley extending west to the southeast corner of New Mexico. The thermal gradient had strengthened with a temperature difference of 9 C (16 F) between Lubbock and San Angelo. Due to the poor initialization at 1200 UTC, the NGM forecast highest relative humidities over northeast Texas, southwest Arkansas, and northern Louisiana (not shown) resulting in a stronger thermal gradient and strongest surface convergence placed over East Texas (Fig. 8). The Eta Model predicted the surface boundary to be in close proximity to what was seen on the 1800 UTC surface map (Fig. 9), and it kept a weak thermal gradient in the general area. The Eta also accurately predicted that the best surface convergence would be along the Red River Valley. With surface features playing a critical role in where deep moist convection should occur, one would expect the Early Eta to be the model of choice.

Although the models initialized well at the upper levels, convective parameterizations would also work in favor of the Early Eta model. According to Keyser and Johnson (1984), latent heat released by an MCS plays a big role in mesoscale circulations and can be linked to enhancement of upper-level winds to the north of the convective system by superimposing an upper-level anticyclone over the convection. Thus, an increase in 200 mb winds from northeast Oklahoma and southern Missouri, extending southeast across Arkansas to Louisiana and Mississippi by 0000 UTC August 29 (Fig. 10), was likely the result of an MCS which produced heavy rains over north-central Texas the previous evening.

According to Junker (1992), numerical models, especially the NGM, have difficulty simulating the effects of latent heat and often weaken an upper-level jet while an upper-level circulation remains apparent on satellite imagery. This appears to be the case during August 28-29, where the 1200 UTC run of the NGM on August 28 only forecast a 200 mb wind speed maximum moving into northwest Texas and failed to forecast a second upper-level jet (Fig. 11) by 0000 UTC on August 29. The Early Eta managed to forecast this second jet with accuracy (Fig. 12). Furthermore, the Eta model accurately forecast the 200 mb speed maximum to drift south, which resulted in a forecast for enhanced vertical motion from paired jet streaks by 0600 UTC (Fig. 13). The NGM continued to forecast upward vertical motions from only the left front quadrant of the western jet (not shown). The 1200 UTC August 29 analysis at 200 mb (Fig. 14) shows that the wind speed maximum did in fact drift south as forecast by the Early Eta.

With the distinct features indicated over northwest Texas by the 1800 UTC surface analysis (Fig. 7), it would be safe to expect some convergence to be noted at 850 mb. As expected, the Early Eta indicated convergence and held on to this feature through the duration of the flash flood event. Meanwhile, the NGM lost the feature in the first 12-hr (not shown). By 0600 UTC August 29, the Eta (Fig. 15) forecast a e ridge axis over the Mexican Plateau extending north into central New Mexico, with a secondary ridge axis extending northeast from Del Rio to near Waco by 0600 UTC. This secondary ridge was just south of the low-level thermal gradient (not shown) and about 200 km southeast of the 850 mb shear axis located from the Edwards Plateau extending northeast to near Wichita Falls. Southerly inflow of up to 10 ms-1 (20 kt) was forecast to advect this moisture toward the surface feature.

The NGM forecast for this same time (not shown) showed a broad 850 mb e ridge axis over northwest Texas extending west to near El Paso (Fig. 16) with no low-level thermal gradient. Q-vector analysis of the Early Eta (Fig. 17) revealed that Q-vectors were crossing the isotherms from cooler to warmer temperatures at 950 mb over west-central Texas, indicating frontogenesis along the weak surface boundary. The Eta model performed well as the surface analysis at 0600 UTC (Fig. 18) confirmed that heavy rains had already occurred over San Angelo. Thus, the Eta model made significant improvements over the NGM in locating the most likely area to experience flash flooding.

The Early Eta forecast of mean 700 to 300 mb flow, which is often the preferred path of these systems, indicated a 10 ms-1 (20 kt) southeastward movement during the evening of August 28, but roughly a 5 ms-1 (10 kt) movement by 1200 UTC August 29 (not shown). It should be noted that at 0600 UTC, the wind direction at 200 mb (Fig. 13) was forecast by the Eta model to be nearly perpendicular to the low-level boundary (Fig. 15). Although the upper flow did not turn out to be parallel to the surface feature (Fig. 18), as suggested for a classic mesohigh or frontal-type MCS (Maddox, et al. 1979), the decrease in forward speed of the shortwave proved sufficient to produce flooding. Flooding over Kimble and Sutton Counties was due to thunderstorms along the surface boundary that became nearly stationary.

Farther to the north, the most severe flooding occurred over Irion and Tom Green Counties, largely due to highly efficient rain-producing thunderstorms behind the boundary. This is consistent with studies by Maddox, et al. (1979) which found that heaviest rainfall during mesohigh and frontal-type events generally occurred on the cool side of the frontal boundary.

6. Summary and conclusions

Though the NGM provided useful information for synoptic scale parameters in this case, the additional levels in the boundary layer and decreased horizontal grid spacing used in the Early Eta model gave it a distinct advantage over the NGM in forecasting deep moist convection. This study illustrates how crucial model initialization of low-level features is for quantitative precipitation forecasting in a highly tropical environment.

The Eta also appeared to excel in convective parameterization which resulted in a more accurate forecast of upper-level winds. In the opinion of the author, more attention should be paid to the effects of latent heat on upper- level circulations during the summer. As forecasters gain a better understanding of the effects of latent heat, and as technology allows us to improve model resolution, we should expect to see continued improvements in not only model performance, but our performance as well.

Acknowledgments. The author wishes to thank Greg Jackson (SOO) and Shirley Matejka (MIC) at NWSO San Angelo for their helpful comments and suggestions in the preparation of this manuscript. Patrick and Amy McCullough and Phillip Baker (forecasters) also deserve thanks for their assistance in using Word Perfect 7.0 and MicroGrafX.

REFERENCES

Junker, N. W., 1992: Heavy Rain Forecasting Manual. National Weather Service Training Center, Kansas City, Mo., 91 pp.

Keyser, D. A. and D. R. Johnson, 1984: Effects of Diabatic Heating on the Ageostrophic Circulation of an Upper Tropospheric Jet. Mon. Wea. Rev., 112, 1709-1724.

Maddox, R. A., C. F. Chappell and L. R. Hoxit, 1979: Synoptic and Meso- Scale Aspects of Flash Flood Events. Bull. Amer. Soc., 60, 115-123.

Mesinger, F., 1996: Improvements in Quantitative Precipitation Forecasts with the Eta Regional Model at the U.S. National Centers for Environmental Prediction. Res. Activ. In Atmos. and Oceanic Modelling, WMO, Geneva, 23, 5.27-5.28.