THE REDBANK CREEK FLASH FLOOD OF 19 JULY 1996
On 19 July 1996 widespread severe flash flooding struck the western Pennsylvania counties of Venango, Clarion and Jefferson. The flooding produced $225 million in damage and caused one fatality. The worst damage ($100 million) was concentrated in Jefferson County, primarily in the towns of Punxsutawney, Summerville, and Brookville. Record river flooding resulted from the widespread flash flooding.
The Areal Mean Basin Estimated Rainfall (AMBER) program (Davis and Jendrowski 1996) will be used to examine the Weather Surveillance Radar, 1988 Doppler (WSR-88D) rainfall estimates observed in several of the watersheds. AMBER computes Average Basin Rainfall (ABR) for each watershed segment and radar rainfall estimates for each rain gage. The AMBER rain gage estimates can be compared with the rain gage measurements to validate the WSR-88D rainfall estimates of ABR.
The WSR-88D can use one of two reflectivity-rainfall (Z/R) relationships for convective rainfall. The standard convective Z/R (Z = 300 R 1.4), or the tropical convective Z/R (Z = 250 R 1.2). The tropical Z/R should be used when warm rain processes dominate in a convective cell. Using the tropical Z/R nearly doubles the WSR-88D estimated rainfall. The standard convective Z/R was used for all ABR computations in this paper.
2. METEOROLOGICAL SUMMARY
The Pittsburgh sounding on 19 July 1996 at 0000 UTC was very moist with high relative humidity to 9.2 km, a 14 oC 850 mb dewpoint, and precipitable water of 50 mm. The warm coalescence layer was 3.7 km deep with a narrow vertical distribution of CAPE. All of these factors pointed to the efficient production of rainfall, and the possible occurrence of tropical rainfall rates (Chappell 1993). The only factor working against the occurrence of heavy rainfall at the ground was the strong winds aloft indicating cell speeds of 20 m s-1.
Thunderstorms repeatedly formed over eastern Crawford County and moved southeast to western Clearfield County at speeds of 20-25 m s-1. A warm frontal boundary across southwest Pennsylvania acted to focus and anchor the storms.
*Corresponding author address: Robert S. Davis, Pittsburgh National Weather Service, 192 Shafer Rd., Moon Twp, PA 15108;
1996). The observed cell movement on radar was about 310 o at 20-25 m s-1. Flash flooding seldom occurs with such high cell training speeds. However, when a large number of cells train over the same area, even at high cell speeds, flash flood producing rainfall can be produced over a large area.
Fig. 1. Rain gage observations for all gages reporting more than 50 mm of rainfall. Shaded gray area is 100 mm or more of rainfall.
3. VENANGO AND CLARION FLOODING
The soil moisture of the watersheds in Clarion, Jefferson, and Venango Counties was relatively high as indicated by the Flash Flood Guidance (FFG) of 28 mm for 1 h and 48 mm for 3 h. FFG is defined as the ABR needed in a specific period of time to initiate flooding on stream or creek. ABR is the only valid rainfall comparison with FFG. ABR for small watersheds can only be calculated in near real time using WSR-88D rainfall estimates.
A flash flood warning was issued at 0804 UTC for Venango, Clarion and Jefferson Counties. Most of the small streams in Venango and Clarion Counties experienced flash flooding. Flood damage was estimated at $50 million in Venango County and $75 million in Clarion County.
Two automobiles stalled in high water on a bridge in the town of Van in Venango County (Fig. 2). The drivers fled before the rising water of East Sandy Creek swept the cars downstream. The AMBER estimate ABR was 126 mm in the basin area upstream of Van (150 km2).
The only fatality during this entire flash flood event occurred in the Deer Creek watershed in a trailer park near the city of Shippenville in Clarion County (Fig. 3). Around 0900 UTC an elderly woman was killed and her husband injured when their trailer was swept away by the rising water.
Fig. 2. The East Sandy Creek watershed.
Fig. 3. The Deer Creek Watershed.
The impact of high cell speed on observed rainfall on the ground is to spatially smooth amounts in the direction of cell movement. The heaviest rainfall in these two counties fell from 0700-1000 UTC with a relatively uniform swath of rain from 100-120 mm. Figure 4 shows the distribution of rainfall rates in the headwaters of Deer Creek. This rainfall distribution is typical of the ABR observed across basins in Venango and Clarion Counties. Examination of the smaller watersheds in Figs. 2 and 3 shows ABR values in the same range as the rainfall distribution in the gray shaded area across Venango and Clarion Counties. With slow cell speeds, heavy rainfall tends to be isolated in small watersheds like Deer Creek, with large gradients of rainfall, even in the direction of cell movement (Davis 2000).
The headwaters of Deer Creek received about 25 mm of ABR from 0530-0700 UTC, completely saturating the ground. Most of the 70 mm of ABR that fell from 0700-0900 UTC (Fig. 4) became runoff and produced the flood wave at the trailer park near Shippenville.
Fig. 4. ABR (mm) and ABR rate (mm h-1) for the headwaters of Deer Creek. Time in UTC.
4. REDBANK CREEK - JEFFERSON COUNTY
Record river flooding and flash flooding occurred across the Redbank Creek (Fig. 5) basin (510-540). The flooding in the Mahoning Creek (600) will be
discussed in the next section. The greatest devastation occurred in the towns of Brookville and Summerville. Over 30 businesses in Brookville were destroyed. The death toll in Summerville alone could have been over 20, but for the heroics
Fig. 5. The Redbank and Mahoning Creek watersheds.
of several citizens. An evacuated trailer was swept away on Water Street about 1045 UTC. At about the same time, over 20 people were stranded on the roofs of their homes along Water Street by the rising water. Private citizens in large motor boats rescued these people from their roofs. Less than one hour later all of the houses were washed downstream. Steep valley walls of the Redbank Creek through Brookville and Summerville concentrated flood damage close to the creek.
The AMBER estimated ABR for each of the four major watershed segments of the Redbank Creek is show in Figs. 6-9. Each major segment is divided into at least three subdivisions. Each segment is assigned a number (Fig. 6). Listed for each segment is the radar range from the Pittsburgh WSR-88D, area of the basin, and the ABR from 0500-1200 UTC. For example, in Figure 6, 515 is assigned to the segment between Summerville and Brookville, the radar range is 109-118 km, the area of the segment is 73 km2, and the ABR is 81 mm.
Only rainfall falling in the stream segments 530 (North Fork, Fig. 8) and 540 (Sandy Lick Creek,Fig. 9) contributed to the flooding in Brookville. ABR
Fig. 6. The Redbank Creek: Brookville to Saint Charles, Pennsylvania.
Fig. 7. The Little Sandy Creek watershed.
in segments 530, 540, and 515 produced the flooding in Summerville. The heaviest AMBER estimated ABR (70-90 mm) was concentrated in the watershed segments surrounding Brookville.
The ABR estimates are considerably lower than the rain gage estimates (120-180 mm) across central Jefferson County. A comparison of the radar rain gage estimates with the rain gages (Fig. 10) shows the Redbank Creek rain gages underestimated the actual rainfall by 40-50%, while gages at similar ranges in Clarion and Venango
Counties were within 10-20% of the radar estimates. The radar underestimation may have occurred due to warm rain processes enhancing rainfall amounts in Jefferson County.
The spatial and temporal distribution of rainfall in a watershed can have a significant impact on the size of the flood wave. If rainfall occurs in the headwaters of a stream earlier in time than on the downstream segments, the flood wave can be greatly enhanced. The hourly rate of rainfall in each
Fig. 8. The North Fork of Redbank Creek.
Fig. 9. The Sandy Lick Creek watershed.
segment of Redbank Creek (Fig. 11) shows the rainfall in the headwaters (530 and 540) peaked around 0700 UTC, while the ABR in the downstream segments (510 and 520) peaked several hours later (about 0900 UTC).
The river gage at Saint Charles, Pennsylvania reached flood stage of 5.2 m at 1100 UTC, and the previous record flood stage (1936) of 5.7 m at 1200 UTC. The record crest of 7.3 m at Saint Charles occurred at 1600 UTC on 19 July 1996.
Fig. 10. Percentage Error in radar estimates as a function of radar range (km). Rain gages in the Mahoning Creek basin (¨) and Redbank Creek basin (±).
Fig. 11. ABR rate (mm h-1) for the Redbank Creek subdivisions. Time in UTC.
5. MAHONING CREEK JEFFERSON COUNTY
Stream flooding along Elk Run and Canoe Creek in the east end of Punxsutawney, Pennsylvania forced evacuations ofv homes and businesses by 1100 UTC. Record river flooding occurred on the Mahoning Creek at Punxsutawney by 1600 UTC (Fig. 12). A flood wall was built through town after the record flood of 1936. Water flowed over the flood wall on 19 July 1996 for the first time since its construction. Punxsutawney is built on the flood plain of Mahoning Creek. When water overflowed the flood wall, much of the town was inundated with water, resulting in considerable flood damage. The two rain gages in the Mahoning Creek basin ( Fig. 10) showed the WSR-88D underestimated the observed rainfall by 30-40%. The ABR values listed in Fig. 12 may have to be almost doubled to approximate the actual rainfall.
Fig. 12. The Mahoning Creek watershed.
AMBER has been shown to be an effective warning tool for small watersheds (Davis 1998, 2000). The summation of the small AMBER basin segments into ABR for larger watersheds, such as Red Bank Creek, can provide excellent guidance for flooding on large streams as well.
Flash flooding can occur with high cell training speeds. Rainfall can be spread over a very large area if many thunderstorms train at high speed over the same geographic area. The impact of the widespread heavy rainfall can result in both flash flooding and significant river flooding.
Determination of the correct Z/R relationship can be crucial in estimating flash flood producing rainfall with radar. AMBER can provide parallel ABR databases for both standard and tropical convective rainfall rates. The forecaster can then select the appropriate Z/R relationship that might apply to a given storm system.
Chappell, C. F., 1993: Dissecting the Flash Flood Forecasting Problem. Post-Print Volume, Third National Heavy Precipitation Workshop, NOAA Technical Memorandum NWS ER-87, 293-297.
Davis, R. S., 2000: Detecting Flash Floods in Small Urban Watersheds. Preprints, 15th Conf. on Hydrology, Amer. Meteor. Soc., Long Beach, CA, 233-236.
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.