Reconstruction of flood magnitude,
inundation extent and flow patterns
in an urban flood
Andrew J. Miller, University of Maryland, Baltimore County
James A. Smith, Princeton University
Mary Lynn Baeck, Princeton University
Paul Bates, University of Bristol
Tim Fewtrell, University of Bristol
Jeffrey Neal, University of Bristol
Katherine L. Meierdiercks, Siena College
Acknowledgements
• National Science Foundation EAR-0208269
• UMBC Center for Urban Environmental Research and
Education (CUERE)
• Institute of Ecosystem Studies (IES) and Baltimore Ecosystem
Study (BES)
• Maryland Department of Environment/FEMA
• USGS/MD-DE-DC Water Science Center
• Willis Research Network
• Peter Nelson, Eliot Holland, Mike McGuire, Matt Ballantine,
Jane Diehl, Tamara Newcomer, Peggy Preston, Craig Nelson,
Andy Brett, Andy Lapetina, Alex Lester, Mike McNey, Alex
Ntelekos, Garth Lindner, Matt Baker
Flash flood hazards in the
Baltimore metropolitan area
Boy, workers swept to deaths as
storm tears across state
Water in culvert rose from inches
to feet in a flash, police report
3rd man missing, feared dead as rising water hits crew in
Woodlawn culvert
Published on: November 20, 2003
Workers say two men were swept out by swift current that
barred rescue
Published on: November 22, 2003
Frank D. Roylance
Laura Barnhardt and Sara Neufeld
An 11-year-old North Baltimore boy died in a stream
while walking home from school and three workers were
swept away by floodwaters at a culvert repair site in
Woodlawn yesterday as heavy rains and thunderstorms
turned urban streams into raging torrents.
The child, identified by police as Darryl McTier Jr. of the
6000 block of The Alameda, was walking home from
school when he was caught in swollen Chinquapin Run.
Firefighters found his body wedged in rocks about 300
feet south of Woodbourne
The water flowing through the culvert under Interstate 70
where three construction workers were making repairs
Wednesday was only 2 inches deep at first. About 2:30
p.m., as the rain started falling harder, the water rose to
about 4 inches.
Within a minute, the project foreman told Baltimore
County police, the pipe, which measures 8 feet in
diameter, was three-fourths full.
The workers were still inside.
In a flash, the first worker was swept out. It looked as if he
was swimming with
Articles from the Baltimore Sun, November 2003
Older urban watershed with buried streams and storm drains
Drainage area = 9.1 km2
Moores Run storm of July 6-7, 2003
16 minutes
10
100
90
Streamflow
8
7
Basin-average rainfall rate
16 minutes
80
70
6
60
5
50
4
40
3
30
2
20
1
10
0
20:00
0
20:30
21:00
21:30
22:00
Rainfall rate in mm/hr
Discharge in m3/s/km2
9
Baltimore area watersheds of
drainage area 3 to 20 km2 with
USGS stream gage records
Gwynns
Falls
Gwynns Falls at Delight
Dead Run
Moores Run
Fall Line
Blue – Surface
drainage
Orange – Storm Drain
Dead Run watershed
Natural and artificial
drainage networks
Watershed hydrologic
response is enhanced
by impervious cover
and a dense network
of storm drains
1950’s-era suburban development with high impervious cover and
limited stormwater management
Drainage area = 14.3 km2
Dead Run at Franklintown
Storm of August 9, 2003
0.8
40
70 minutes
35
m3/s/km2
0.6
Discharge
Basin-average rainfall rate
0.5
30
25
0.4
20
0.3
15
0.2
10
0.1
5
0
0
6:30
7:00
7:30
8:00
8:30
9:00
9:30
mm/hr
0.7
10:00
Dead Run at Franklintown
Storm of Sept 18-19, 2003
72 minutes
62 minutes
3
70
60
50
2
40
1.5
30
Discharge
Basin-average rainfall rate
1
20
0.5
10
0
1:00
0
1:30
2:00
2:30
3:00
3:30
4:00
4:30
5:00
5:30
6:00
6:30
7:00
mm/hr
m3/s/km2
2.5
Research on flood dynamics
• Problem: Need for improved forecasts of flood-wave
propagation, inundation extent, flood probabilities –
especially in urban watersheds
• Research tools and methods:
– Rainfall fields (high-resolution weather radar,
corrected using point measurements)
– Observational analysis and mesocale modeling (WRF
model) of warm-season thunderstorms
– High-resolution topographic data to characterize
channels and flood-prone areas
– Field surveys of high-water marks and inundated area
after floods
– 2-d hydraulic modeling to match observations and
make predictions
Storm and
flood of
July 7, 2004
Dead Run
watershed
Dead Run Watershed
Contours based on
storm-total rainfall at
rain-gage network
Point totals used in
bias-correction of
WSR-88D record with
6-minute time
intervals
WATER RESOURCES RESEARCH, VOL. 44, W08446, doi:10.1029/2007WR006346
From Ntelekos et al., 2008, WATER RESOURCES RESEARCH, VOL. 44, W08446,
doi:10.1029/2007WR006346
Basin-average rain rate
100
90
80
Millimeters
70
60
50
40
30
20
10
0
13:00
14:00
15:00
16:00
17:00
Time
18:00
19:00
20:00
Cumulative Rainfall
Dead Run Watershed - July 7, 2004
140
120
Millimeters
100
80
60
40
20
0
13:00
14:00
15:00
16:00
17:00
Time
18:00
19:00
20:00
Maximum precipitation accumulation and recurrence intervals
Dead Run Watershed, July 7, 2004
Duration
(minutes)
Maximum
accumulation
(mm)
Recurrence interval
(years)
15
30
60
120
27.9
49.8
79.4
120.9
10
37.5
108
384
Dead Run at Franklintown stream gauge characteristic hydrograph response
to a rainfall pulse is 60-75 minutes
Suggests that flood peak most likely responds to the 60-minute maximum
accumulation and may have comparable recurrence interval
Dead Run Stage Hydrographs for July 7, 2004 Flood
14
13
12
11
Stage in feet
10
Field-surveyed high-water
mark at Franklintown gage
DR4
DR2
Time based on field notes
taken onsite during flood
DR3
Franklintown
9
8
7
6
5
4
3
2
1
0
15:00
15:30
16:00
16:30
17:00
17:30
Time
18:00
18:30
19:00
19:30
20:00
To model flood hydraulics, we need detailed topography.
Coverage of LiDAR points along Dead Run – note prominent gaps
in the channel…
…which we fill by
conducting field surveys
with a total station.
Dead Run upstream of Franklintown gage
Channel data = field survey + LiDAR
Bank/floodplain data = LiDAR
We then identify and survey
high-water marks in the field
DR5 gage – high water line
Flood of 9-27-08
Flood of
July 7, 2004
We georeference them to photo
image points and LiDAR elevations
Yellow – July 7 Inundation Area
Light blue – FEMA 500-yr Floodplain
Dark blue – FEMA 100-yr Floodplain
TELEMAC-2D1
• Numerical solution of 2- dimensional depth-averaged form
of the Navier-Stokes continuity and momentum equations
• Finite-element mesh
• Specify topography and boundary conditions, including
inflow hydrograph
• Model predicts inundated area, depth and velocity vectors
for each node for each time step
1
Hervouet, J-M., 2007. Hydrodynamics of free-surface flows: modelling with
the finite element method. Wiley, 341 pp.
Dead Run Watershed
DR2 Gauge
DR2 Gauge
Comparison of
surveyed and modeled
inundation areas
Transect at DR2 Gauge
Modeled maximum water surface vs. high water marks
120
119
Stage in meters
Bed elevation
118
Maximum water surface
High Water Marks
117
116
115
114
20
30
40
50
60
Horizontal distance in meters
70
80
Thalweg longitudinal profile at DR2 Gauge
Modeled maximum water surface vs. high water marks
118
Bed elevation
Stage in meters
117
Maximum water surface
High Water Marks
116
115
114
113
112
10
60
110
160
Longitudinal distance in meters
210
DR2 Stage Hydrograph - July 7, 2004 Flood
117.0
Stage in meters
116.5
116.0
Recorded stage
Modeled stage
115.5
115.0
114.5
0
1
2
3
Time in hours
4
5
DR2 Discharge Hydrograph - July 7, 2004 Flood
50
Peak flow rate: 46.6 m3/s (1646 cfs)
for a watershed of 1.9 km2 (0.74 mi2)
45
1600
40
1400
Runoff volume of 106 mm, or
85.5% of DR2 basin-average
storm-total rainfall
30
1200
1000
25
800
20
600
15
400
10
5
200
0
0
0
0.5
1
1.5
2
2.5
Time in hours
3
3.5
4
4.5
5
Discharge in cfs
Discharge in m3/s
35
• At the Dead Run at Franklintown USGS gauge, the July 7, 2004 event was the
flood of record and exceeded the flood peak during Tropical Storm Agnes in 1972
• The gauge was inundated during the flood and stopped recording for 95 minutes
when the flood was at its peak. High-water marks were surveyed by USGS and
used in the official indirect estimate of peak flow. (8700 cfs or 246 m3/s).
Flow
Franklintown
Gauge
We used TELEMAC to estimate the
peak flow with the best match to
the ensemble of USGS high-water
marks (5650 cfs or 160 m3/s)
We also estimated the timing and
shape of the missing portion of the
hydrograph
Dead Run at Franklintown
July 7, 2004 hydrograph
10000
9000
Adjusted based on model results
8000
Discharge in cfs
7000
6000
5000
4000
3000
2000
1000
0
Interpolated from USGS indirect
Runoff volume based on the
USGS estimate is 93.5 mm or
80% of rainfall volume
Runoff volume based on our
model results and adjusted
hydrograph is 75 mm or
64% of total rainfall volume
Gauge overtopped
and did not record
from 16:00 to 17:35
Dead Run at Franklintown
10000
July 7, 2004
Annual flood peaks in cfs
9000
8000
Tropical Storm Agnes
The three largest floods
in the record are outliers
Higher values are official
USGS discharge estimates
7000
6000
Lower values are based
on recalibration of the
rating curve assuming
that the July 7, 2004
flood peak is consistent
with our model results
5000
4000
3000
2000
1000
0
1960
1970
1980
1990
2000
2010
Exceedence probability
99.5% 99% 98%
95%
90%
80%
70%
50%
30% 20%
10%
5%
2% 1%
0.5%
100000
Bulletin 17B Log Pearson III Flood frequency analysis
Dead Run at Franklintown official peak flow record
Recurrence interval of July 7 flood: ~68 years
95% confidence interval: 30 – 180 years
Discharge in cfs
10000
1000
100
3.0
2.0
1.0
0.0
z-statistic
-1.0
-2.0
-3.0
Exceedence probability
99.5% 99% 98%
95%
90%
80%
70%
50%
30% 20%
10%
5%
2% 1%
0.5%
100000
Bulletin 17B Log Pearson III Flood frequency analysis
Dead Run at Franklintown adjusted peak flow record
Recurrence interval of July 7 flood: ~46 years
95% confidence interval: 21 – 109 years
Discharge in cfs
10000
1000
100
3.0
2.0
1.0
0.0
z-statistic
-1.0
-2.0
-3.0
Franklintown gauge results
• Uncertainty in peak flow estimation and flood
frequency analysis
• Sensitivity of hazard assessment to uncertainty in
estimates for even a single event
• In urban watersheds, “ordinary” storms may
generate big floods; but floods generated by rare
events may not be as rare as the storms themselves
• Importance of models as tools for investigating
uncertainty and filling gaps in observation
DR3
DR4
250
Dead Run at DR3/DR4 site - final inflow hydrographs
200
8000
7000
150
5000
DR3
DR4
DR3+DR4
4000
100
3000
2000
50
1000
0
0
0
2000
4000
6000
8000
10000
12000
Model time in seconds
14000
16000
18000
Discharge in cfs
Discharge in m3/s
6000
Match of modeled stage with recorded stage
105
DR3 recorded stage
104
DR4 recorded stage
Modeled DR3 stage
Stage in m
103
102
101
100
99
Modeled DR4 stage
Ponding at high stage extends
300-400 m upstream of bridges
and road embankments
Simulated vs observed
thalweg flood profiles
July 7, 2004
Dead Run July 7, 2004
Peak discharge vs. drainage area
300
Peak discharge [m3/s]
DR3/DR4 combined
Franklintown gauge
(USGS indirect)
250
200
150
100
Franklintown gauge
(model results)
50
0
0
5
Drainage area [km2]
10
15
Dead Run July 7, 2004
compared with historic mid-Atlantic floods
Peak discharge [m3/s]
1000
100
Smethport 1942
10
Dead Run 2004
Camille 1969
Sauls Run 2003
1
0.01
0.1
1
Drainage area [km2]
10
100
Comparison of LISFLOOD-FP1 1D/2D hydraulic model
with TELEMAC-2D for the July 7, 2004 flood
Dr. Timothy Fewtrell1, Dr. Jeff Neal2 and Prof. Paul Bates2
1 Willis
Research Network, University of Bristol, University Road, Bristol, BS8 1SS, UK
2 University of Bristol, University Road, Bristol, BS8 1SS, UK
• Channel is approximated as 1D and rectangular and solved by
the diffusive wave equation
• Floodplain flow is calculated in 2D using the continuity and
momentum equations
• LISFLOOD is orders of magnitude more efficient than standard
2d hydraulics models
1
Bates, P.D., M.S. Horritt, and T.J. Fewtrell, 2010. A simple inertial formulation of the
shallow water equations for efficient two dimensional flood inundation modelling.
Journal of Hydrology doi:10.1016/j.jhydrol.2010.03.027
Comparison of surveyed inundation lines with LISFLOOD-FP
results based on Miller input hydrographs for DR3/DR4 site
Comparison of surveyed inundation lines with LISFLOOD-FP
results based on Miller input hydrographs for DR3/DR4 site
Comparison of surveyed inundation lines with LISFLOOD-FP
and TELEMAC model results based on Miller input
hydrographs for DR3/DR4 site
106
DR3 Maximum water level, July 7 2004 flood
105
Elevation in meters
104
103
102
101
100
Channel bed
99
Maximum free surface elevation
Observed high water marks
98
Lisflood maximum water level
97
0
200
400
600
Longitudinal distance in meters
800
1000
104
DR4 Maximum water level, July 7 2004 flood
103
102
Elevation in meters
Channel bed
Maximum free surface elevation
101
Observed HWM
Lisflood maximum water level
100
99
98
97
0
100
200
300
400
500
Longitudinal distance in meters
600
700
800
Comparison of TELEMAC-2D and LISFLOOD-FP
• Both are computationally stable
• Both provide good match to mapped flood inundation area
(assumes a good topographic base)
• LISFLOOD-FP missed backwater effects of infrastructure; but
these were not built into the model
• LISFLOOD does well with high-water marks outside of
backwater zone
• LISFLOOD can run this extreme-flood scenario in minutes
rather than hours
• If a model can be set up in advance, might allow real-time
prediction of flood inundation for given hydrologic inputs
Conclusions
• The July 7, 2004 Dead Run flood is a valuable case study for
application of new tools and methods for flash floods
• The flood was comparable to the most extreme floods
measured at comparable drainage areas in the mid-Atlantic
region
• 2-d hydraulic models can generate results that closely match
high-water marks, inundation extent, and water-level records
• Modeling results highlight uncertainty in flood-frequency
analysis and sensitivity to the accuracy of estimation of the
largest events
• Urban infrastructure exerts a controlling influence on flood
hydraulics and needs to be accounted for in modeling studies
• With proper advance preparation, LISFLOOD-FP has the
potential to be used in real-time prediction of flood
inundation extent