Professional Documents
Culture Documents
Abstract: Small watersheds in urbanizing and suburbanizing regions are an increasingly important element of the flood hazard problem
in the United States and abroad. This paper presents a methodology to estimate flood response in urban watersheds, with the goal of
assessing the performance of systems of small flood control reservoirs for extreme floods. The setting of this study is the Brandywine
Creek watershed in Chester County, Pa. A number of flood control structures were built within the basin, and the availability of data from
U.S. Geological Survey 共USGS兲 stream gauging stations and rainfall estimates from National Weather Service weather radars provides the
opportunity to study the performance of a small watershed flood control system. In this paper, the flood response to rainfall from
Hurricane Floyd, which produced record flooding in Brandywine Creek on September 16, 1999, is examined. A distributed hydrologic
model was developed and used to estimate flood discharges both with and without the existing flood control structures. The model was
calibrated using observed discharge data from the USGS gauging stations, and model simulations were then performed with the flood
control structures removed to estimate their impact on the watershed response to the storm. Results from the model showed that the flood
control structures mitigated flood hazards to a greater degree than had been expected by the studies completed prior to construction. This
paper shows how the USGS streamflow measurement program can be augmented with hydrologic modeling and WSR-88D weather radars
to provide useful hydrologic information for analysis of extreme floods.
DOI: 10.1061/共ASCE兲1084-0699共2006兲11:5共432兲
CE Database subject headings: Floods; Hurricanes; Pennsylvania; Watershed management.
flood peak values range from 1.9 m3 s−1 km−2 at Honey Brook,
the station with the smallest drainage area, to 0.6 m3 s−1 km−2 at
Chadds Ford, the station with the largest drainage area.
The gauge at Honey Brook is cited by the USGS as having
“fair” records, except for those above 28 m3 s−1, which are “poor”
共Durlin and Schaffstall 1999兲. The largest directly measured dis-
charge for the Honey Brook gauging station is 48.7 m3 s−1. The
Honey Brook gauge has a history of revisions to peak discharge
data, notably a 1999 revision of the 1972 peak from Hurricane
Agnes from 230 to 104 m3 s−1 共Durlin and Schaffstall 1999兲. The
peak discharge from Hurricane Floyd was also revised downward.
Fig. 1. Brandywine basin, showing basin outline, streams, gauging The accuracy of discharge observation plays an important role in
stations, and dams assessing the performance of the Brandywine flood control sys-
tem.
Prior to construction of the flood control structures in the West
ince and is characterized by pronounced heterogeneity in land Branch of Brandywine Creek, the USGS conducted a series of
surface properties 共Wolman 1955兲. A band of limestone traversing hydrologic modeling studies on the effects of the West Branch
the watershed in an east-west orientation between Coatesville and structures on flooding in the basin. Once the model was devel-
Downingtown, Pa., is important in determining the drainage net- oped, estimates of the effects of future population growth and
work structure, settlement pattern, and transportation network proposed flood control structures on the flood response of the
within the watershed 共Brandywine Valley Association 1998兲. Al- basin were completed. The model used in the study was devel-
though the basin contains 19 different soil associations, 75% of oped by Dawdy et al. 共1978兲 and included components for soil
the basin consists of a moderate to well-drained loam 共Greig et al. moisture accounting, infiltration, and routing 共Sloto 1982兲.
1998兲. A comparison of the soils map with a bedrock geology The USGS model was implemented for the three subbasins of
map of the region shows that the soils are closely associated with the West Branch of Brandywine Creek: Honey Brook, Coates-
the underlying bedrock geology. ville, and Modena. It was noted that the model could not be
The Brandywine basin has experienced major land use satisfactorily calibrated for the Honey Brook subbasin 共Sloto
changes since European colonization. Although the region was 1982兲. Model calibration problems for the Honey Brook subbasin
heavily forested before settlement in 1685, approximately 70% of were due to errors in the discharge data for Honey Brook, as
the land area was cleared by 1735 共Bolakas 1984兲. Agricultural reflected in the 1999 revision of the 1972 peak discharge.
practices during the 18th and early 19th centuries resulted in dra- Model analyses were carried out to predict the effects of a
matic changes to channels and valley bottoms of the basin 共Wol- series of proposed flood control structures. Two USGS studies
man 1955; Jacobson and Coleman 1986兲. During the period since included estimates of flood reduction by the proposed flood con-
1950, Chester County has experienced a linear growth rate in trol structures. Both studies estimated possible reductions in peak
population density of almost 2.5 people km−2 year−1. Population flows during 100-year flood events of up to 40% at Coatesville
density is expected to continue to increase, as developers seek and 20% at Modena 共Sloto 1982, 1988兲. Based partially on the
additional land for suburban development within commuting dis- studies conducted by the USGS, the Chester County Water Re-
tance of Philadelphia. sources Authority constructed five dams for flood control pur-
The Brandywine watershed includes nine United States Geo- poses. The characteristics of the dams are shown in
logical Survey 共USGS兲 stream gauging stations that were active Table 1. The capacities listed are total capacities, including any
in 1999 during Hurricane Floyd. The subbasins investigated in normal storage used for water supply or recreation. For multipur-
this report are: 共1兲 West Branch Brandywine at Honey Brook pose reservoirs, flood control capacity will be lower than that
共drainage area 48.4 km2, USGS stream gauging station listed.
#01480300兲; 共2兲 West Branch Brandywine at Coatesville Although studies were conducted prior to construction to
共119 km2, #1480500兲; 共3兲 West Branch Brandywine at Modena investigate the effects of the proposed flood control structures,
共142 km2, #01480617兲; 共4兲 East Branch Brandywine near Down- systematic examination of the benefits from the flood control
ingtown 共157 km2, #01480700兲; 共5兲 East Branch Brandywine structures has not been undertaken 共Sloto, personal communica-
below Downingtown 共233 km2, #01480870兲; and 共6兲 Brandywine tion, 2001兲. The purpose of this study is to determine the reduc-
at Chadds Ford 共743 km2, #01481000兲. Annual flood peak distri- tion in flood peaks and volumes due to the system of flood control
butions, expressed as unit discharge 共i.e., discharge divided by reservoirs within the Brandywine Creek watershed during Hurri-
drainage area兲 generally decrease with drainage area. Ten-year cane Floyd in September 1999.
Coatesville. This also contributes to a decrease in runoff and run- Ford. The size of the subbasin, representing approximately half of
off ratio from Honey Brook to Coatesville. Table 4 shows the the watershed, serves to dampen any variability from the up-
results of the water balance computations for each of the indi- stream subbasins. Flood response at Chadds Ford is strongly con-
vidual 共nonoverlapping兲 subbasins 共as opposed to Table 3, which trolled by conditions in the Chadds Ford subbasin.
covers the drainage areas defined by the stream gauging stations兲. Hurricane Floyd was a devastating storm, based on the mon-
It can be seen that the Coatesville subbasin has low values for etary damage caused and on the loss of 57 lives. It produced
runoff and runoff ratio, which is why the cumulative water bal- record flooding in the Brandywine Creek basin. At three of the six
ance through Coatesville shows a reduction from Honey Brook. stations considered, the discharge generated by Floyd was greater
The large runoff ratios for Modena are likely due in part to am- than any event since the introduction of flood control structures.
plification of runoff associated with urban and suburban The scale of Hurricane Floyd provides a good opportunity to
development. investigate the performance of those structures in preventing large
On the East Branch Brandywine Creek, the effects of the flood scale flood disasters in the Brandywine watershed.
control dams can be easily noted in the subbasin defined by the
station near Downingtown. The lower runoff, runoff ratio, and
peak discharge can all be attributed to the three upstream flood
control structures. The largest reservoir, Marsh Creek, is 3 km Hydrologic Modeling Studies
upstream from the station. As is the case for the Modena subba-
sin, the subbasin below Downingtown is also more developed The Network Model 共Morrison and Smith 2001兲 is a distributed
than other parts of the basin. Increased development leads to in- hydrologic model, which is used in this study to analyze spatially
creased runoff in that subbasin. Important to note is the indepen- varying flood response in the Brandywine Creek basin. Discharge
dence of the Chadds Ford subbasin as compared to the others at any location along the drainage network is represented as
studied. The runoff and runoff ratio values are the same for the
Chadds Ford subbasin as for the full basin defined by Chadds
Table 3. Water Balance for Brandywine Watershed during Hurricane Table 4. Subbasin Water Balance for Brandywine Watershed during
Floyd Event Hurricane Floyd Event
Runoff Peak Area Runoff
Station 共mm兲 Runoff ratio 共m3 s−1 km−2兲 Station 共km2兲 共mm兲 Runoff ratio
Honey Brook 58.1 0.26 1.72 Honey Brook 48.4 58.1 0.26
Coatesville 45.4 0.20 0.87 Coatesville 70.2 36.6 0.16
Modena 64.9 0.29 1.21 Modena 23.8 60.8 0.73
Near Downingtown 46.6 0.21 0.98 Near Downingtown 157.0 46.6 0.21
Below Downingtown 56.6 0.25 0.88 Below Downingtown 75.8 77.5 0.25
Chadds Ford 59.5 0.27 1.02 Chadds Ford 368.1 59.4 0.27
The basin was modeled first with uniform soil properties 共an-
tecedent moisture content and saturated hydraulic conductivity兲
and then with three soil regions. The soil properties were modeled
as spatially variable among the three drainage basins defined
above the stations at Modena, below Downingtown, and between
those two and Chadds Ford. The model with three soil regions
produced better results, but the difference was not important for
the overall conclusions of the model analyses.
Calibration was accomplished by first matching the observed
Fig. 4. Drainage basin and network extracted from DEM timing of the flood response at each of the stations. This was done
by varying the overland and channel flow velocities of the model
until the correct timing of peak discharge was achieved. The an-
冕 冉 冊
tecedent moisture content was estimated by taking into account
d0共x兲 d1共x兲 rainfall prior to the period of concern during the Hurricane Floyd
Q共t兲 = M t− − ,x dx
A v0 v1 event. The saturated hydraulic conductivity of the soils was varied
to reproduce the water balance of the watersheds. The East
where A = domain of the drainage basin above the specified loca- Branch Brandywine was set first, with the objective being to
tion; x is an arbitrary location in A; M共t , x兲 = runoff rate 共mm h−1兲 minimize total absolute fractional error between the model and
at time t and location x; d0共x兲 = distance from x to the closest recorded data for the two East Branch stations. The same was
stream channel; and d1共x兲 = channel flow distance from x to the done for the West Branch, with the focus on minimizing error for
outlet of the basin specified by the region A. The total flow dis- the stations at Coatesville and Modena, as the data at Honey
tance from x to the basin outlet is d0共x兲 + d1共x兲. The runoff rate Brook are unreliable 共as noted previously兲. With the East and
M共t , x兲 is computed from the rainfall rate R共t , x兲 using the Green- West Branches calibrated, the saturated hydraulic conductivity
Ampt infiltration model with moisture redistribution 共see Ogden was set for the subbasin between those stations and the station at
and Saghafian 1997 for algorithm details兲. Runoff is assumed to Chadds Ford. This was done by minimizing the error at Chadds
move over hillslopes at a uniform velocity v0 and through the Ford. The final parameters used in the model are shown in Table
channel system at velocity v1 共see Rodriguez-Iturbe and Rinaldo 5.
1997 for discussion of similar models兲. Following calibration, the model was run both with and with-
The channel network of the Brandywine Creek basin was ex- out the flood control structures included, for both the spatially
tracted from the digital elevation model 共DEM兲 using an area- invariant and variable soil parameters. The model with spatially
threshold algorithm 共with an area threshold of approximately invariant soil properties produced accurate representations of the
0.08 km2兲. The drainage network and basin area 共Fig. 4兲 were hydrographs at Honey Brook, Modena, and below Downingtown,
then used for implementation of the Network Model. The three while overestimating peak discharge and runoff at Coatesville and
large sections indicated by solid lines in Fig. 4 mark the Modena near Downingtown and underestimating the same at Chadds Ford.
共western兲, Downingtown 共eastern兲, and Chadds Ford 共southern兲 Despite the apparent deviation between the model and actual dis-
subbasins, and the dashed outlines are the subbasins defined by charge for the model with spatially invariant soil, the model re-
flood control dams. Rainfall input for the model was determined sults are generally good in representing peak discharge and runoff
from WSR-88D weather radar estimations of rainfall rate over the 共see Tables 6 and 7兲.
watershed from the station at Fort Dix, N.J.
The modeling objective was to assess the performance of the
flood control system during Hurricane Floyd. The first step in the Table 6. Calibration Data for Model with Spatially Invariant Soil: Peak
Discharge
modeling analyses was to reproduce the actual discharge hydro-
graphs of the Hurricane Floyd event. After that had been Actual Model Error
achieved, “natural discharge” without the flood control system Station 共m3 s−1兲 共m3 s−1兲 共%兲
was computed. In order to reproduce the Hurricane Floyd event, Honey Brook 84 72 −14
however, model analyses had to account for the effects of the Coatesville 108 125 15
flood control structures. This was accomplished by effectively Modena 172 161 −7
removing the basin area upstream of dams from the model analy-
Near Downingtown 153 95 −38
ses. Observed discharge values from the reservoirs during Hurri-
Below Downingtown 204 190 −7
cane Floyd were small as compared to measured discharge values
Chadds Ford 762 821 8
in the major branches of Brandywine Creek.
Fig. 5. Discharge hydrograph at Honey Brook for model with spatially variable soil
Fig. 6. Discharge hydrograph at Coatesville for model with spatially variable soil
that, in order to model the effects of the flood control system to a though the model does capture the response of the watershed, the
large scale event, it is not necessary to completely capture the analysis of the flood control system is not very sensitive to minor
response of the watershed to a rainfall event. Also, because the changes in the model.
comparison is based on a large scale flood event, minor differ- The analysis shows that the flood control structures were ef-
ences in saturated hydraulic conductivity are not of great concern. fective at reducing peak flows and runoff volumes on the East
With rainfall rates near 30 mm h−1, a difference in Khsat of 1 or Branch Brandywine Creek, particularly for the station near
2 mm h−1 would not lead to a significant change in flooding. Even Downingtown. Reductions of approximately 10% in peak dis-
Fig. 7. Discharge hydrograph at Modena for model with spatially variable soil
Fig. 8. Discharge hydrograph near Downingtown for model with spatially variable soil
charge and runoff were also experienced on the West Branch Summary and Conclusions
Brandywine Creek. The benefits were most significant immedi-
ately downstream of the flood controls, with the highest reduc- Hurricane Floyd generated record flooding in watersheds through-
tions at the stations near Downingtown and at Coatesville. The out the East Coast, from North Carolina to New England. Over a
station at Chadds Ford also benefited from the system of flood 35 h period, between 170 and 220 mm of rain fell in the Brandy-
control structures, although to a lesser degree, likely due to its wine watershed, with peak rainfall rates of approximately
distance downstream from the reservoirs. 30 mm h−1. Floyd produced a record flood peak at Chadds Ford
Fig. 9. Discharge hydrograph below Downingtown for model with spatially variable soil
Fig. 10. Discharge hydrograph at Chadds Ford for model with spatially variable soil
Table 8. Calibration Data for Model with Spatially Variable Soil: Peak and peaks not seen since 1973 on the East Branch Brandywine.
Discharge Hydrologic modeling analyses of flooding in Brandywine Creek
Actual Model Error were used to examine the performance of a system of small flood
Station 共m3 s−1兲 共m3 s−1兲 共%兲 control reservoirs in a rapidly urbanizing region.
The Brandywine Watershed Work Plan led to the construction
Honey Brook 84 65.7 −21
of five flood control dams and their associated reservoirs between
Coatesville 108 115.0 6
1970 and 1994 共Brandywine Valley Association 2000兲. The flood
Modena 172 148.5 −14 control system reduces discharge from larger tributaries into the
Near Downingtown 153 121.5 −21 main stem, and consequently the discharge downstream along the
Below Downingtown 204 245.3 20 main stem is reduced 共Leopold and Maddock 1954兲. Dams along
Chadds Ford 762 736.2 −3 tributaries to the East and West Branch Brandywine Creek pro-
vide primary protection to areas along those two streams and
secondary protection to the main stem below their confluence.
Studies completed prior to construction of flood controls on
Table 9. Calibration Data for Model with Spatially Variable Soil: Runoff the West Branch Brandywine Creek estimated the protection af-
Actual Model Error forded by Hibernia Dam, the only such structure actually built in
Station 共mm兲 共mm兲 共%兲 the West Branch. This reservoir was estimated to provide approxi-
Honey Brook 58 56 −3 mately 6% reduction in flood peaks at Coatesville and Modena
Coatesville 46 53 15
for a 100 year flood, with reductions decreasing with decreasing
flood magnitude 共Sloto 1982, 1988兲. For Hurricane Floyd, de-
Modena 67 57 −15
creases in peak flow are estimated to have been 12 and 10% at
Near Downingtown 49 34 −31
Coatesville and Modena, respectively. Decreases in runoff vol-
Below Downingtown 58 50 −15
ume were also comparable, with 11 and 9% reductions at Coates-
Chadds Ford 61 56 −8
ville and Modena, respectively.
No studies were conducted on the East Branch flood control
structures prior to construction, preventing comparison between
Table 10. Effects of Flood Control Structures: Peak Discharge expected and actual performance. For Hurricane Floyd, peak dis-
Uncontrolled Controlled Change charges near Downingtown and below Downingtown were re-
Station 共m3 s−1兲 共m3 s−1兲 共%兲 duced 53 and 36%, respectively. Runoff volume also decreased
Honey Brook 66 66 0 56 and 40% at the two stations, respectively. Further downstream,
Coatesville 130 115 −12
at the outlet at Chadds Ford, benefits from the flood control struc-
tures were less, but still substantial. Peak discharge and runoff
Modena 165 149 −10
volume at Chadds Ford were reduced 16 and 17%, respectively.
Near Downingtown 260 122 −53
With complete coverage of the United States by WSR-88D
Below Downingtown 382 245 −36
weather radars and advances in hydrologic modeling, a system
Chadds Ford 881 736 −16
can be developed to supplement the current USGS streamflow
measurement program. Errors in discharge estimates at high Leopold, L. B., and Maddock, T., Jr. 共1954兲. The flood control contro-
flows, which is when measurement accuracy is most important for versy: Big dams, little dams, and land management, Ronald Press
flood protection design calculations, are unavoidable. Radar esti- Company, New York.
mates of rainfall and hydrologic models can be used to augment Morrison, J. E., and Smith, J. A. 共2001兲. “Scaling properties of flood
streamflow measurements by the USGS. Such a system can serve peaks.” Extremes, 4共1兲, 5–23.
to increase accuracy and provide the basic hydrologic data dis- Ogden, F. L., and Saghafian, B. 共1997兲. “Green and Ampt infiltration with
cussed by Leopold and Maddock 共1954兲 as necessary for flood redistribution.” J. Irrig. Drain. Eng., 123共5兲, 386–393.
control system design. Pasch, R. J., Kimberlain, T. B., and Stewart, S. R. 共1999兲. “Preliminary
report: Hurricane Floyd 717 September 1999.” National Hurricane
Center, National Oceanographic and Atmospheric Administration,
具http://www.nhc.noaa.gov/1999floyd.html典 共April 2, 2001兲.
Acknowledgments
Rodriguez-Iturbe, I., and Rinaldo, A. 共1997兲. Fractal river basins:
Chance and self-organization, Cambridge University, Cambridge,
This research was supported in part by the National Science U.K.
Foundation 共Grants EAR02-08269, EAR-0409501, and ITR- Sloto, R. A. 共1982兲. “A stormwater management model for the West
0427325兲 and the U.S. Army Research Office 共Grant DAAD19- Branch Brandywine Creek, Chester County, Pennsylvania.” Water-
99-1-0163兲. This support is gratefully acknowledged. Resources Investigations 81-73, United States Geological Survey,
Harrisburg, Pa.
Sloto, R. A. 共1988兲. “Effects of flood controls proposed for West Branch
References Brandywine Creek, Chester County, Pennsylvania.” Water-Resources
Investigations Rep. 86-4054, United States Geological Survey, Harris-
Atallah, E. H., and Bosart, L. F. 共2003兲. “The extratropical transition and burg, Pa.
precipitation distribution of Hurricane Floyd 共1999兲.” Mon. Weather Smith, J. A., Baeck, M. L., Meierdiercks, K. L., Nelson, P. A., Miller, A.
Rev., 131共6兲, 1063–1081. J., and Holland, E. J. 共2005a兲. “Field studies of the storm event hy-
Baeck, M. L., and Smith, J. A. 共1998兲. “Estimation of heavy rainfall by drologic response in an urbanizing watershed.” Water Resour. Res.,
the WSR-88D.” Weather Forecast., 13, 416–436. 41.
Bolakas, J. F. 共1984兲. “Investigation of flood plain development, Brandy- Smith, J. A., Baeck, M. L., Morrison, J. E., and Sturdevant-Rees, P.
wine Creek, Chadds Ford, Pa.” MS thesis, Univ. of Delaware, 共2000兲. “Catastrophic rainfall and flooding in Texas.” J. Hydrometeor.,
Newark, Del. 1共1兲, 5–25.
Brandywine Valley Association. 共1998兲. State of the Watershed Rep.: Smith, J. A., Baeck, M. L., Morrison, J. E., Sturdevant-Rees, P. L.,
Brandywine Valley 1997, West Chester, Pa. Turner-Gillespie, D. F., and Bates, P. D. 共2002兲. “The regional hydrol-
Brandywine Valley Association. 共2000兲. State of the Watershed Rep.: ogy of extreme floods in an urbanizing drainage basin.” J. Hydrom-
Brandywine Valley 1999, West Chester, Pa. eteor., 3共3兲, 267–282.
Cope, A. M. 共2001兲. “Radar-derived rainfall estimates for Hurricane Smith, J. A., Baeck, M. L., Steiner, M., and Miller, A. J. 共1996兲. “Cata-
Floyd over New Jersey.” Proc., Symp. on Precipitation Extremes: strophic rainfall from an upslope thunderstorm in the central Appala-
Prediction, Impacts, and Responses, American Meteorological Soci- chians: The Rapidan Storm of June 27, 1995.” Water Resour. Res.,
ety, Boston, 352–356. 32共10兲, 3099–3113.
Dawdy, D. R., Schaake, J. C., Jr., and Alley, W. M. 共1978兲. “User’s guide Smith, J. A., Baeck, M. L., Zhang, Y., and Doswell, C. A., III. 共2001兲.
for distributed routing rainfall-runoff model.” Water-Resources Inves- “Extreme rainfall and flooding from supercell thunderstorms.” J. Hy-
tigations 78–90, United States Geological Survey, Reston, Va. drometeor., 2共3兲, 469–489.
Durlin, R. R., and Schaffstall, W. P. 共1999兲. “Water resources data— Smith, J. A., Miller, A. J., Baeck, M. L.,Nelson, P. A., Fisher, G. T., and
Pennsylvania, 1999.” Water-Data Report PA-99-1, United States Geo- Meierdiercks, K. L. 共2005b兲. “Extraordinary flood response of a small
logical Survey, Harrisburg, Pa. urban watershed to short duration convective rainfall.” J. Hydrom-
Environmental Resources Research Institute. 共1996兲. “Pennsylvania dam eteor., 6共5兲, 599–617.
locations.” Pennsylvania Department of Environmental Protection Wolman, M. G. 共1955兲. “The natural channel of Brandywine Creek Penn-
GIS Compendium, 具http://www.pasda.psu.edu典 or 具http:// sylvania.” Geological Survey Professional Paper 271, United States
www.dep.state.pa.us典 共Feb. 15, 2001兲. Geological Survey, Washington, D.C.