An Underestimation of WSR-88D Storm Total Precipitation during the Hurricane Erin and Opal Heavy Rain Events over Middle Tennessee

Timothy W. Troutman and Mark A. Rose
NWSO Nashville, TN

1. Introduction

During the 1995 tropical storm season, two tropical weather systems moved onshore from the Gulf of Mexico that eventually brought heavy rain from middle Tennessee. The amount of rainfall that the remnants of Erin and Opal produced greatly exceeded in some locations the storm total precipitation amounts indicated by the NWSO Nashville WSR-88D. The purpose of this paper is to show that in both these cases, the WSR-88D significantly underestimated the amount of rainfall that occurred.

2. Erin

Tropical storm Erin moved onshore over the extreme western panhandle of Florida and the southwest corner of Alabama at 1600 UTC, 3 August. The storm then followed a northwest course through southwest Alabama, Mississippi, and into extreme eastern Arkansas by 0600 UTC, 5 August before turning toward the north. By 1800 UTC, 5 August, the low center was located near Cape Girardeau, MO (CGI). The low then moved northeast to near Evansville, IN (EVV) by 0600 UTC, 6 August (Fig. 1, not shown).

Because of the strength and abundant availability of moisture associated with the storm, quantitative precipitation forecasters at the National Meteorological Center (NMC-now renamed the National Centers for Environmental Prediction (NCEP)) predicted large rainfall amounts for the 24-hour period ending 1200 UTC, 5 August (Fig. 2, not shown). A maximum of greater than 5 inches was forecast in a narrow band extending from just north of Meridian, MS (MEI) to just northwest of Memphis, TN (MEM) during the same period.

By 1200 UTC, 5 August, Erin's circulation center had moved to near Jonesboro, AR (JBR). heavy rain spread into parts of middle Tennessee during the night of 4 August and the early morning hours of 5 August. During the 24-hour period ending 1200 UTC, 5 August, the amount of rainfall over middle Tennessee was generally forecast well by NMC forecasters. The exception, however, was an area of rainfall greater than two inches that occurred from southwest middle Tennessee into extreme northwest Alabama. A maximum of 3.62 inches was reported in southeast Wayne County, TN (Fig. 3, not shown).

Interestingly, WSR-88D storm total precipitation estimates ending at 1234 UTC, 5 August indicated that between 1.5 and 2.0 inches of rain had fallen during this period in the area of the 3.62 inch maximum (approximately one-half of what had been reported), whereas a maximum of between 2.5 and 3.0 inches of rain was indicated by the WSR-88D in Perry County, TN, about 20 miles north-northwest of the area in which the observed maximum rainfall had actually occurred (Fig. 4, not shown). This portion of Wayne County is part of the Shoal Creek basin, a flash flood prone region.

Because the quantitative precipitation forecast (QPF) indicated a potential rainfall amount of about 2.0 inches in southeast Wayne County, the QPF and WSR-88D storm total precipitation estimate were in reasonably close agreement, whereas the actual amount of rainfall greatly exceeded the forecast and estimated amounts for this area by between one and two inches for the 24-hour period ending 1200 UTC, 5 August. It is important to note that the QPF and WSR-88D rainfall estimates would not have alerted forecasters to potential flash flooding had the levels along Shoal Creek in southwest middle Tennessee been near flood stage.

3. Opal

During the morning of 5 October, the remnants of hurricane Opal moved north-northeastward across middle Tennessee (Fig. 5, not shown). The passage of Opal brought heavy rain to a large area of the central third of the state, with two precipitation maxima observed over middle Tennessee. Over a portion of south-central middle Tennessee, an area of 4.0 to 6.0 inches of rainfall was observed, with a maximum of 6.20 inches reported in central Marshall County during the 24-hour period ending at 1200 UTC, 5 October (Fig. 6, not shown). Five to seven inches of rain fell over an area of east-central middle Tennessee, with a maximum of 7.80 inches observed over south-central Cumberland County. Because the eastern edge of the Cumberland Plateau is located just to the southeast of this area, the location of this maximum was obviously enhanced by orographic lifting. (This is particularly true when surface winds are from a southeast direction, as they were in this region of the state as the storm passed through southern middle Tennessee.)

The maximum rainfall that occurred over Marshall County was along a portion of the Duck River in southwest middle Tennessee, which caused the river to approach flood stage at some locations. Meanwhile, the precipitation maximum that fell over Cumberland County occurred near the headwaters of the Sequatchie River, which caused river flooding downstream, primarily at Whitwell in Marion County, Tennessee (Fig. 7, not shown). It is interesting that the entire Sequatchie River basin in middle Tennessee received rainfall totals ranging between approximately 3.5 inches, along the southern extent of the basin, to over 7.0 inches, along the extreme northern extent.

In this case, the WSR-88D storm total precipitation algorithm greatly underestimated rainfall totals across most of middle Tennessee. for example, the radar estimate of 2.0 to 3.0 inches in extreme south-central Cumberland County was approximately one-third of the reported amount (Fig. 8, not shown). Also, the algorithm estimated 1.0 to 2.0 inches to have fallen over central Marshall County, about one-fourth of the observed amount.

The observed precipitation that occurred due to the remnants of Opal again showed that the inaccuracy of the WSR-88D storm total precipitation estimate could have easily deceived forecasters in their estimate of the flood/flash flood potential.

4. Discussion

These two cases show that the WSR-88D does have an algorithm deficiency in estimating storm total precipitation in tropical weather events. It is important here to note the Z-R relationship using the equation taken from Ahnert et al (1983): Z=300R^1.4.

In this equation, "Z" represents the equivalent reflectivity, and "R" represents the rainfall rate (Racy 1995). The Z-R coefficient used frequently is 300. This coefficient can be adjusted accordingly to precipitation type and/or season. It can vary from 140 (for drizzle) to 84000 (for hail).

The WSR-88D averages two adjacent one degree by one kilometer blocks along the same radial out to 230 km in order to determine values for the one degree by two kilometer average precipitation rate scan. This value is subsequently converted to the nearest 0.5 dBR through the following equation: 1 dBR=10 log [R(1mm/hr)].

A time continuity test is then performed on the processed data. Each new precipitation scan is compared to the last valid scan for continuity. The WSR-88D then assessed this data for any unreasonable echo development os dissipation. If any anomalies are found, the scan is thus rejected and a new scan is commenced. However, most WSR-88D radars currently perform under the assumption that each scan is reasonably accurate and therefore do not use the range effect correction, calculated by the following equation: R1(mm/hr)=a[R(mm/hr)^b r^c].

Here, "R1" represents the corrected rainfall rate and "R" is the original rainfall rate. The range is depicted by "r," whereas the coefficients derived for the individual radar are "a," "b," and "c." (The derivations of these coefficients involve extensive research of each radar site.)

Aside from the deficiencies of the WSR-88D precipitation estimate algorithm, another factor that causes discrepancies with rain gage reports is that the radar instantaneously samples a volume of the atmosphere several thousand feet above the surface. The surface projection is over one square mile and measurements made every 5 to 10 minutes. Rain gage data is acquired continuously as precipitation falls, and the area sampled is usually less than one square foot. Thus, such point measurements may not coincide with the radar volume measurements. The WSR-88D comparison problem with rain gages would be effectively compensated by correctly deriving the above three coefficients used in the range effect correction equation. This may aid in the calibration of surrounding WSR-88D radars as well.

5. Conclusion

Two case studies were presented which showed that the WSR-88D underestimates the amount of precipitation that occurs with tropical systems (the second case being greatly underestimated). Radar underestimation of precipitation will most likely continue to occur in these situations until a more accurate Z-R relationship can be applied locally when the weather situation dictates. Until then, forecasters must rely upon other sources of rainfall data (satellite precipitation estimates (AFOS product "NFDSPENES"), local spotter/cooperative station reports, and any rain gages or river rain gage sites which report rainfall), especially during tropical weather events. Otherwise, flooding/flash flooding may occur without the forecaster's knowledge.

Acknowledgements

The authors wish to thank Mike Murphy, Service Hydrologist, NWSO Nashville, TN, for his suggestions and input to this study and Henry Steigerwaldt, Science and Operations Officer, NWSO Nashville, TN, for his review of this paper.

REFERENCES

Ahnert, P.R., M.D. Hudlow, E.R. Johnson, D.R. Greene, and M.P.R. Dias, 1983: Proposed 'On Site' Precipitation Processing System for NEXRAD. Preprints, 21st Conference on Radar Meteorology, Edmonton, Alberta, Canada, AMS (Boston) and Alberta Research Council, Canadian Meteor. and Oceanogr. Soc., 378-85.

Racy, J.P., and M.L. Kopsky, 1995: A Comparison of NEXRAD Precipitation Estimates to Real-Time Data. Central Region Applied Research Papers No. 15, 70-87.