Documentos de Académico
Documentos de Profesional
Documentos de Cultura
Abstract: The collection of time-sensitive data on real-life embankment failures and their analysis are essential steps to model breach
Downloaded from ascelibrary.org by Hacettepe Universitesi on 03/29/17. Copyright ASCE. For personal use only; all rights reserved.
processes and the consequential flood, which are of great importance for preparing emergency action plans. In this paper, data on 14 earthen
embankments that failed or were damaged in the Midlands of South Carolina resulting from the historic 1,000-year storm during
October 2015 are presented and analyzed. The investigation includes measurement of breach dimensions, collection of undisturbed soil
samples, soil classification, embankment erodibility tests using the submerged jet erosion test (JET) method, calculation of peak discharge,
estimation of maximum height of overtopping, and calculation of maximum reservoir volume behind each embankment at the time of failure.
Using this information and results, single-variable and multivariable parametric breach models for the breach depth and breach width are
developed and compared with selected breach models available in the literature. All the models except for one estimate the breach width
satisfactorily, with the proposed single-variable equation giving a relatively better prediction. The collected data set of this study may be used
by others for model verification. DOI: 10.1061/(ASCE)HY.1943-7900.0001315. 2017 American Society of Civil Engineers.
Author keywords: Overtopping; Embankment breach; Empirical model; Erodibility; Dam failure.
Fig. 1. (Color) Study area including digital elevation model of South Carolina and selected watersheds in Lexington and Richland Counties, location
of the visited dams, watershed boundaries, and hydrography
Fig. 2. (Color) Dam breach at (images by authors) (a) Barr Lake Dam; (b) Cary Lake Dam; (c) Lake Elizabeth Dam; (d) Gibson Pond Dam; (e) Old
Mill Pond Dam; (f) Ulmer Pond Dam; (g) Beaver Dam; (h) Lower Rocky Ford Dam; (i) Spring Lake Dam; (j) Upper Rocky Ford Dam
Fig. 3. (Color) Digital elevation, land use, and soil map (designated by the hydrologic soil groups, with A and D indicating the lowest and highest
runoff potential, respectively) for the watershed for Cary Lake Dam
dam sites. Table 2 lists the field and laboratory investigations done and Atmospheric Administration (2015) during October 34, 2015,
for the visited embankments, along with their hazard classification is 22.2 cm (8.74 in.).
according to the South Carolina Department of Health and Envi- The TR-55 (SCS 1986) graphical peak discharge method is used
ronmental Control (SCDHEC). The watershed within which each to calculate the peak discharge for each dam. The input require-
dam is located is also reported in this table. Further calculation is ments for this method are the time of concentration (tc ), the drain-
necessary in order to delineate the subwatersheds. The watershed age area (A), type of rainfall distribution (Type II), 24-h rainfall
delineation for each dam location is performed utilizing the state- depth (P), curve number (CN), and the percentage of pond or
wide 7.5-min hypsography digital elevation map (DEM) and the swamp areas (Fp ) (Akan and Houghtalen 2003). With this infor-
hydrology tools in ArcGIS. Land-use data, such as open water, mation, the peak discharge, Qp , resulting from a storm event is cal-
deciduous forest, or grassland, are obtained from the National Land culated as
Cover Database (Homer et al. 2015) and the soil type map is ob-
tained from Natural Resources Conservation Service (2011), where Qp Qu AQFp 3
the hydrologic group (A, B, C, and D) is specified. The DEM, land
use, and soil map for the Cary Lake Dam are shown in Fig. 3. The unit peak discharge, Qu , is obtained from the figure for
The weighted curve number is calculated for each dam. the SCS graphical peak discharge for Type II distribution rainfall
Richland and Lexington Counties are located within the Soil Con- (Akan and Houghtalen 2003), utilizing the time of concentration
servation Service (SCS) Type II region. The SCS runoff curve num- and the I a =P ratio. The pond percentage factor, Fp , is deter-
ber method (SCS 1986) is used to calculate for runoff, Q, in inches mined from the table for the adjustment factor for ponds and
swampy areas that are spread throughout the watershed (Akan and
P I a 2 Houghtalen 2003). The peak discharge Qp is then calculated for
Q 1 each dam.
P Ia S
The peak discharge calculations for the watershed for each dam
where are shown in Table 3. The highest estimated peak discharge is in
Lexington County at the Old Mill Pond Dam. Within Richland
I a 0.2S County, the highest predicted peak discharge for this study is at the
Lake Katherine Dam, the dam furthest downstream within the Gills
S 1,000=CN 10 2 Creek Watershed. Along the Gills Creek Watershed, USGS Gauge
02169570 is located just downstream of the Lake Katherine Dam.
and P = rainfall in inches during a 24-h period; I a = initial loss in During the October 2015 flood, the gauge, shown in Fig. 4 (USGS
centimeters; S = maximum retention in centimeters; and CN = 2012), was inoperative between October 4 at 3 a.m. and October 5
curve number. The value for P measured by the National Oceanic at 11 a.m. because the flow was higher than the maximum flow rate
Gibson Pond Dam 7.80 69 5.66 11.35 9.80 0.75 8.14 0.57 1.72 98.62
Old Mill Dam 8.39 69 5.66 11.35 9.80 0.81 9.19 0.57 1.72 114.65
Note: A = drainage area; CN = curve number; Fi = impervious area fraction; Fp = pond and swamp adjustment factor; I a = initial loss; Qp = peak discharge;
Qu = unit peak discharge; R = runoff; S = maximum retention; tc = time of concentration.
Fig. 4. (Color) USGS streamflow gauge at Gills Creek Watershed (image courtesy of USGS)
of 56.63 m3 s1 (2,000 ft3 =s). The peak discharge, Qp , based on dam and at several points within the same embankment by using a
hydrologic calculation for the Lake Katherine Dam, is found to laser distance measurement device (Bosch, Stuttgart, Germany
be 73.31 m3 s1 (2,589 ft3 =s). ArcMap is used to measure the res- GLM40 with an accuracy of 1.5 mm) and the values were then
ervoir area behind each dam at the normal water surface elevation averaged. Multiple soil samples, some of which were undisturbed
and the results are shown in Table 4. samples obtained by using a drive tube with weighted hammer,
were collected from each site for geotechnical investigations in
the laboratory, such as soil classification, erodibility, and dry unit
Field and Laboratory Investigations weight measurements. The maximum head of water above breach
invert, hw , is estimated for each dam (Table 4) from both the high
Fourteen selected dams were visited during October to November water marks observed at the field and also from the report of
2015 immediately following the flood and before any construction Devereaux (2016).
and/or modification was performed. The original embankment The particle size distribution of the embankment material of
characteristics, i.e., dam height (NID 2013), crest width, upstream each dam for the sediment coarser than 0.075 mm was determined
and downstream slopes, and maximum height of crest overtopping, from standard sieve analysis. The grain size distribution of the
are presented in Table 5. Breach geometry, i.e., breach bottom material finer than 0.075 mm was measured with the Beckman
width, breach top width, and breach depth, was measured for each (Brea, California) Coulter counter MS4, which uses impedance
measurement technique. A submerged jet erosion test (JET) appa- breach invert at failure, hw = maximum head of water above the
ratus built in house was used to measure the erodibility of the un- breach invert, and g = gravitational acceleration, as
disturbed soil samples. The JET apparatus, operational procedures,
and data analyses are similar to those described by Hanson and Bavg fkd ; V; hw ; d ; g 5
Cook (2004). The test determines the erodibility coefficient, kd ,
and critical shear stress needed to initiate erosion, c , of a linear
detachment erosion model Using dimensional analysis, Eq. (5) is reduced to the following
nondimensional form:
kd ed c 4
p
Bavg =hw kd d ghw ; V=h3w 6
where = rate of erosion; and ed = applied shear stress.
The variation of scour depth with time is monitored during the
JET method and the erodibility coefficient, kd , and the critical shear The observations from the studied embankment dams are then
stress, c , are then determined using curve-fitting techniques on the used to determine the functional form of the fitted relation
observed time series of the scour depth. for Eq. (6).
100
Embankment Breach Dimensions, Soil Classification, 90
and Erodibility 80
Beaver Dam
Lake Elizabeth
The measured breach parameters, i.e., breach depth, breach top 70 Upper Rocky Ford
Percent Finer (%)
Spring Lake
width, and breach bottom width, are presented in Table 4. Fig. 6 60 Cary Lake
shows a definition sketch of these parameters. The majority of the 50 Lake Katherine
breached dams studied here were low dams with the original height Forest Lake
40
of the dam less than 10 m. For dams that did not fail, the dimen-
30
sions of the eroded section of the embankment are presented
instead of the breach dimensions. No significant deposition was 20
noticed in the vicinity of the dams. The bed erosion in the embank- 10
ments can be estimated by comparing the original dam height with 0
0.001 0.01 0.1 1 10
the measured breach depth. For Spring Lake Dam, which did not
proceed to failure, maximum advancement of the headcut on the (a) Sieve Size (mm)
100
90
Lower Rocky Ford
80 Arcadia Lake
70 Ulmer Pond
Percent Finer (%)
Fig. 6. (Color) Breach profile with breach parameters Fig. 7. Grain size distribution of the material of studied embankments
Gibson Pond Dam 0.1 0.4 0.7 1 10.00 1.60 SW 79.7 6.4
Old Mill Pond Dam 0.01 0.19 0.5 0.72 72.00 5.01 SC-SM 74.5 16.5
Note: Cc = coefficient of curvature, Cc D30 2 =D10 D60 ; Cu = coefficient of uniformity, Cu D60 =D10 ; Dx = grain size such that x% of the sediment is
finer; SC = clayey sand; SM = silty sand; SP = poorly graded sand; SW = well-graded sand.
dimensions, i.e., breach depth and breach width, and the erodibility proposed to estimate the breach dimensions, i.e., average breach
of the embankment material, i.e., kd . Moreover, no correlation is depth, Davg , and average breach width, Bavg
observed between the breach dimensions and the embankment
compaction, i.e., d . Several possible reasons for this exist Davg 0.54B0.71
avg 7
including
1. A high amount of sand in the embankment material, which
makes the studied dams less cohesive and may cause a poor Davg 0.87hw 8
applicability of the detachment relation [Eq. (4)] to model the
erodibility of those embankments.
2. Various embankment slopes and embankment vegetation cover. Bavg 5.59h0.85
w 9
From observations, some of the embankments had steeper side
slopes than the others. In addition, the embankments were cov- Multivariate regression analysis of the field and laboratory data
ered with different vegetation types, e.g., bare soil, mature trees, leads to the following form of Eq. (6) with R2 of 0.88
shrubs, or grass.
3. Different flood magnitudes. The embankments had various Bavg p V 0.012
0.57 2.11kd d ghw 0.001 3 p
drainage areas, amount of impervious area, and amounts of hw hw kd d ghw
precipitation. 0.415 V p
4. Different initial integrity of the embankment prior to the flood. 2.53 106 3 kd d ghw 10
V hw
Some of the embankments were poorly maintained as compared 3 hw
to others. Also, the spillway pipes were corroded and damaged
in some cases, e.g., Beaver Dam. Fig. 11 shows the predicted breach width using the proposed
5. Different spillway capacities for the embankments and different relations [Eqs. (9) and (10)] versus the observed data. Also plotted
spillway designs. in Fig. 11 are predictions by several selected breach models devel-
oped by others. Except for the relation by USBR (1988) that under-
estimates the breach width, most of the predicted results of other
Predictive Breach Models
relations fall within the 30% deviation bounds. The best agreement
Using the regression analysis of the observed embankment breach between the observations and predictions is observed when us-
characteristics [Figs. 10(ac)], the following breach models are ing Eq. (9).
6
10.00
Beaver dam 4
Very Erodible
Cary Lake
Upper Rocky Ford 2
Gibson Pond
0
Old Mill Pond
Erodible 0 1 2 3 4 5 6 7 8 9
1.00 Lake Elizabeth
(b)
Kd (cm3/N-s)
Lower Rocky Ford Maximum estimated water height above breach invert(m)
Ulmer Pond
Barr Lake 45
10.000
50
45
40
Predicted breach width (m)
1.000
Kd (cm3/N-s)
35
30
y = 17.745x-1.543 25
0.100
R = 0.7296
20 Proposed relation (Eq. 9)
Proposed relation (Eq. 10)
15
USBR (1988)
Von Thun and Gillette (1990)
10
0.010 Froehlich (1995)
0 5 10 15 20 25 30 5 Agreement Line
%Silt and Clay Bounds of 30% deviation
0
0 5 10 15 20 25 30 35 40 45 50
Fig. 9. (Color) Variation of erodibility coefficient with the percentage
Observed average breach width (m)
of fine grain size (silt and clay)
Fig. 11. (Color) Predicted and observed breach width using proposed
relations [Eqs. (9) and (10)] and some of the available models
Summary and Conclusions