19 JULY 1996 MONONGALIA COUNTY WEST VIRGINIA FLASH FLOOD:
During the early morning of 19 July 1996, Monongalia County in northern West Virginia was hit with flash flooding that damaged 121 homes and 30 private bridges (NOAA 1996). Fortunately there were no casualties. Worst hit were the communities of Cassville and Osage along Scotts Run (Figure 1) where 50 homes were damaged and several cars and a mobile home were swept into bridges (Ryan 1996). The nearby Morgantown area (in particular Westover and Star City) suffered urban flooding.
Fig. 1. Scotts Run basin and Morgantown area in Monongalia County WV.
During 1200-2100 UTC 18 July 1996, Monongalia County's steeply-sloped terrain was drenched by thunderstorm rains of 50-100 mm. Subsequently during 0400-0500 UTC 19 July 1996, an intense thunderstorm delivered 25-30 mm of rain to the Scotts Run watershed and nearby Morgantown area. Flash flooding quickly ensued. Other thunderstorm rains of 20-40 mm during 1000-1500 UTC 19 July 1996 brought minor flooding to areas just south of Morgantown.
Almost 24 hours earlier at 0825 UTC 18 July 1996, the National Weather Service (NWS) had issued a Flash Flood Watch that included Monongalia County. Soil moisture content was high across the upper Ohio Valley due to the 14-16 July 1996 rainfall of 40-80 mm.
Meteorological conditions for 18-19 July 1996 were forecasted to be unusually humid (precipitable water 50 mm) and moderately unstable (maximum 2000 J kg-1) and, therefore, conducive to the development of thunderstorms with torrential rains. The parallel orientation of westnorthwest winds aloft and a warm frontal boundary across the lower Great Lakes raised the potential for thunderstorm rains to repeat over some areas. A follow-up Flash Flood Statement issued at 1955 UTC mentioned how rain amounts exceeding 1 inch (25 mm) in 1 h could cause flash flooding. At 0244 UTC 19 July 1996 a Flash Flood Warning was issued for Monongalia County when radar showed a thunderstorm with intense rainfall bearing down on the county from the northwest. The flash flood watch and warning for Monongalia County had excellent lead time. The focus of this case study is on the specific flash flood that occurred in Scotts Run basin (43 km2). The purpose is to show how modernized hydrometeorological tools may help pinpoint the timing and severity of future flash floods on such small basins. The tools included Weather Surveillance Doppler radar (WSR-88D), the Areal Mean Basin Estimated Rainfall program (AMBER, Davis and Jendrowski 1996), NWS River Forecast Center (RFC) Flash Flood Guidance (FFG, Sweeney 1992), and an algorithm that estimates peak discharge. This algorithm used rainfall-runoff relationships and triangular synthetic unit hydrographs derived by the Soil Conservation Service (McCuen 1998 and Maidment 1993) and similar to those in The Flood Analysis and River Emulator (FLARE) program (Young et al. 1996). A vastly more sophisticated precipitation runoff modeling system (PRMS) has been developed by the Community Hydrometeorology Laboratory (CHL) at the National Center for Atmospheric Research (NCAR, Warner et al. 2000).
Using the Scotts Run flash flood as an example, we will show how WSR-88D reflectivity-based AMBER estimates of Average Basin Rainfall (ABR) can be used to assess the appropriateness of radar reflectivity-rainfall (Z-R) relationships. We will further demonstrate how monitoring ABR rates can give an early alert of ABR totals that can exceed FFG. Lastly, we will explain how ABR can be used to time peak discharge when used with hydrologic programs like FLARE or PRMS.
2. TIMELY Z-R RELATIONSHIP ASSESSMENT
Gaining immediate access to radar-derived ABR and automated rain gage data facilitates real-time comparisons between the two data sets. Warning meteorologists can use such comparisons along with other meteorological data to assess the proper Z-R relationship during an ongoing event. Furthermore, programs like AMBER can give simultaneous displays of ABR derived from different Z-R relationships. This enables the warning meteorologist to visualize the spectrum of possible ABR, use different Z-R relationships for different basins or time periods as the event may require, while not having to make internal changes to the WSR-88D's Z-R relationship algorithm. This last element often requires extensive coordination and so is reserved for overland tropical storm situations.
The 0000 UTC 19 July 1996 rawindsonde observation for Pittsburgh Pennsylvania showed several characteristics favorable for tropical rainfall rate development (Chappell 1993). Precipitable water was 50 mm, CAPE was only 300 J kg-1 despite an equilibrium level of 10 km, and the warm coalescence layer [freezing level minus lifted condensation level (LCL)] was 3.8 km. In such scenarios, the tropical convective Z-R relationship (Z = 250 R1.2) may yield better radar rainfall estimates than the standard convective Z-R relationship (Z = 300 R1.4). The tropical Z-R relationship typically yields double the rainfall estimate of the standard one.
Figure 2 shows the percent difference between AMBERs radar-derived rainfall estimates using the standard convective Z-R relationship and selected automated rain gage observations in Monongalia County for three distinct heavy rain periods during 18-19 July. The sites were 100-104 km away from the Pittsburgh WSR-88D radar. Comparisons were plotted for periods when gage readings exceeded 20 mm.
Fig. 2. Percent difference between AMBER-generated ABR and rain gage observations.
From a cursory perspective, the standard Z-R relationship appeared adequate for the first two heavy rain periods but may have yielded too low an estimate for the third period. Given the samples small size and large variance, the third period appears to be the type of situation warranting closer examination using additional rainfall estimates including tropical Z-R relationship application. Updated rawindsonde, satellite, and lightning observations should also be considered.
3. ABR AND EXCESSIVE RAINFALL RATES
Current practice is to issue countywide flash flood warnings when rain amounts are expected to exceed FFG within a county. The future expectation is flash flood warnings for specific basins when ABR exceeds FFG.
Once each day, RFCs issue county FFGs for rainfall durations of 1, 3, 6, 12, and 24 h. The areal distribution and rate of rainfall is assumed to be uniform during each period. FFG is the ABR needed to create sufficient runoff to initiate flooding. Threshold runoff (TRO) is specified for small, unrated basins. The focus for thunderstorm rains is 1 h and 3 h TRO and FFG. (Longer durations yield nonlinear TRO and FFG increases.) On 18 July 1996, 3 h FFG for Monongalia County and Scotts Run was 53 mm. This means if 3 h ABR in Scotts Run exceeded 53 mm, runoff would exceed the TRO of 13 mm and flooding would begin. 1 h FFG is defined as 60 percent of 3 h FFG, in this case, 32 mm.
The WSR-88D radar delivers reflectivity data and rainfall estimates every 5-6 minutes or 10-12 times per hour. When 5-6 min ABR estimates begin to exceed 8-10% of the 1 hr FFG, a basin's ABR can potentially exceed 1 h FFG within the next 60 min. Consequently, warning meteorologists can use 5-6 min ABR as a flash flood alerting tool for individual basins. For this reason, AMBER displays the latest ABR rate in addition to 1, 2, and 3 h ABR totals (Davis 1998).
At 0420 UTC 19 July 1996, AMBER showed a 6 min ABR of 4.1 mm in the Cassville subbasin of Scotts Run. This 6 min ABR corresponded to an ABR rate of 41 mm h-1 which exceeded the 1 h FFG of 32 mm (Figure 3). So the 0420 UTC 6 min ABR served as a first alert that 1 h FFG could be exceeded within the hour and that flash flood warnings might be needed for Scotts Run. During 0420-0449 UTC additional 6 min ABR rates exceeded 1 h FFG within all subbasins of Scotts Run, reinforcing the idea flash flooding could soon occur. Thirty-five minutes later (at 0455 UTC) 1 h ABR neared 1 h FFG.
Fig. 3. Sequence of ABR rates exceeding 1 h FFG in Scotts Runs three subbasins and nearby Westover for 0400-0500 UTC 19 July 1996.
4. ABR AND PEAK DISCHARGE
Equations (1), (2), and (3) obtained from Maidment 1993 and McCuen 1998 form the basis of the simple algorithm used to transform a small basin's 1 h ABR into peak discharge. Despite the excellent results in this case, the focus is on explaining the relationship between terms. All times are in h with t as the time after 12 UTC when heavy rain began, d as the duration of heavy rain (in this case, d = 1) and tp as the lag time to peak discharge Qp.
Equation (1), a rainfall-runoff relationship, shows Rs, accumulated runoff since 12 UTC, depends on Ps, accumulated ABR since 12 UTC, and Ia, the initial abstract (rainfall amount before any runoff begins). The initial abstract depends on prior soil moisture content. In some hydrologic models Ia is determined using the antecedent index (AI). For Ps < Ia, Rs = 0. The squared numerator in Equation (1) depicts the sudden nonlinear increases in accumulated runoff often observed during heavy rain episodes.
Equation (1) is solved for Rs at times t and t+1 to get the 1 h runoff used in Equation (2) to determine Qp, scaled peak of the triangular synthetic unit hydrograph. Equation (2) shows Qp also depends on basin topography as A denotes area (km2), and tp (Equation (3)) partially depends on L, the basin's length (m) from most distant ridge to point of discharge, and H, the corresponding elevation difference (m). McCuen (1998) includes other forms of Equation (3) that show tp and correspondingly Qp dependent on other basin characteristics, including antecedent soil moisture and types of soil and vegetation.
5. SCOTT BASIN FLASH FLOOD
From 1200 UTC 18 July 1996 to 0400 UTC 19 July 1996, AMBER estimated an ABR of 27 mm for Scotts Run basin using WSR-88D reflectivity data and standard convective Z-R relationship. This 16 h ABR generated 2 mm accumulated runoff as the AI-based initial abstract was 15 mm. At 0244 UTC 19 July 1996, NWS Pittsburgh issued a Flash Flood Warning for Monongalia County until 0545 UTC. A thunderstorm 30 km in diameter with rainfall rates in excess of 40 mm h-1 was moving southeastward toward Monongalia County. The WSR-88D storm movement algorithm projected the intense thunderstorm rains would begin moving across Scotts Run at 0415-0430 UTC.
At 0420 UTC AMBER showed the Cassville subbasin (at the headwaters of Scotts Run) to have a 6 min ABR of 4.1 mm that corresponded to an ABR rate of 41 mm h-1. This high ABR rate was the first alert that 1 h FFG could be exceeded within the hour. During 0420-0449 UTC additional 6 min ABR rates exceeded 1 h FFG. The 1 h ABR for the entire Scotts Run basin of 28 mm neared the 1 h FFG (32 mm) at 0455 UTC. The 1 h ABR for the Cassville and Pursglove subbasins slightly exceeded 1 h FFG. These ABRs were based on the standard convective Z-R relationship whose 1 h rainfall estimate of 26 mm for the nearby Morgantown Lock and Dam site was within 4% of the 1 h GOES gage observation of 25 mm (Fig. 2).
The 1 h ABR ending at 0500 UTC (28 mm) matched the ABR of the previous 16 h, yet this additional rainfall burst generated over 12 mm of runoff in an hour compared to 2 mm of previous runoff. Figure 4 is the resulting triangular synthetic hydrograph based on ABR being uniform in time and areal distribution.
Figure 3 shows uniform time distribution was not a valid assumption. Since the thunderstorm was moving southeastward and downstream through Scotts Run basin, the flood wave would be enhanced since runoff from the headwaters above Cassville could reach the segment between Pursglove and Osage at the same time local runoff would peak in that subbasin.
Using Equations (1), (2), and (3) programmed with topographic characteristics of Scotts Run basin (A = 43 km2, L = 9980 m, and H = 213 m), the time of peak discharge at Osage was projected as 0530 UTC or 1.5 h after the first sizeable ABR report within the basin (Figure 4). Ryan (1996) reported that a security guard in Osage reported water first covered the parking lot at 0445 UTC, a car was swept into a fence at 0510 UTC, and a sport utility vehicle and dumpster box were swept downstream at 0520 UTC. By 0630 UTC, Scotts Run was back within its banks. Considering the bankfull-level discharge at 0630 UTC should equal the discharge of the initial report of overflow at 0445 UTC, the timing of these two events plus the accurate timing of the flood crest supports the algorithm's validity in this case.
Fig. 4. Triangular synthetic hydrograph for Scotts Run 0400-0800 UTC 19 July 1996.
Beginning at 0437 UTC, AMBER similarly showed ABR rates exceeding 1 h FFG in the nearby communities of Westover and Star City. Roads and basements became flooded about the time (0500 UTC) the 1 h ABR exceeded 1 h FFG.
AMBER-generated ABR estimates using WSR-88D can facilitate real-time comparisons of radar rainfall estimates and automated rain gage observations. Such comparisons combined with rawindsonde, satellite, and lightning observations can help determine the appropriate Z-R relationship for the episode. AMBER can also display rainfall estimates from several Z-R applications simultaneously. This can be beneficial when the proper Z-R relationship may vary in space and time.
AMBER-generated 5-6 min ABR can be easily converted to 1 h ABR rates. The 5-6 min ABR rate can serve as a first alert that rainfall in the basin is on track to exceed 1 h FFG and that flash flood warnings for the basin may be needed within 60 min.
The next step for improving flash flood warnings for stream basins will be to integrate AMBER-generated ABR with sophisticated hydrologic-assessment tools like PRMS. The Scotts Run example gives insight into the principles used by such tools and promotes the feasibility of more precise warnings at the scale where floods emanate.
Chappell, C. F., 1993: Dissecting the flash flood forecasting problem. Post-Print Volume, 3rd National Heavy Precipitation Workshop, NOAA Tech. Memo. NWS ER-87, 293-297.
Davis, R. S., 1998: Detecting time duration of rainfall: a controlling factor of flash flood intensity. Preprints, Special Symposium on Hydrology, Phoenix AZ, Amer. Meteor. Soc., 258-263.
Davis, R. S., and P. Jendrowski, 1996: The operational Areal Mean Basin Estimated Rainfall (AMBER) module. Preprints, 15th Conf. on Wea. Analysis and Forecasting, Norfolk VA, Amer. Meteor. Soc., 332-335.
Maidment, D. R., ed., 1993: Handbook of Hydrology. McGraw-Hill Inc, 1384 pp.
McCuen, R. H., 1998: Hydrologic Analysis and Design. Prentice Hall, 814 pp.
NOAA, 1996: Storm Data, 38, Number 7, 339-340.
Ryan, E., 1996: Scotts Run rampages over Osage. The Dominion Post. Morgantown WV. 19 July 1996. 1-A.
Sweeney, T. L., 1992: Modernized areal flash flood guidance. NOAA Tech. Memo. NWS HYDRO 44, NWS Office of Hydrology, 32 pp.
Warner, T. T., D. N. Yates, and G. H. Leavesley, 2000: A community hydrometeorology laboratory for fostering collaborative research by the atmosphere and hydrologic sciences. Bull. Amer. Meteor. Soc., 81, 1499-1505.
Young, R. P., G. Garnet, B. Midcap, M. Wooldridge, and A. Rezek, 1996: The Flood Analysis and River Emulator (FLARE) Operator Handbook. NWS/NOAA Charleston WV. 109 pp.