Predicting Breach Formation
Mark Morris
HR Wallingford
The Science of Asset Management
London – 9th December 2011
www.floodrisk.org.uk
EPSRC Grant: EP/FP202511/1
“An overview of current options, practice and
future direction for predicting breach formation”
1. What are we talking about?
2. Current options / Latest developments
3. Practice
4. Future direction
5. Research gaps
[1] What are we talking about?
What are we talking about?
What are we talking about?
Why do we need to predict breach?
The way and rate at which an embankment breaches can
affect the timing of the breach, the rate and magnitude of
the flood water released and the size of the breach itself.
Therefore, breach affects the analysis of flood risk (ie.
FRAs) and can change the way in which flood events
might be managed.
Understanding the degree of uncertainty within the
process and any prediction is a very important aspect of
using breach predictions
Why do we need to predict breach?
Consider that different aspects of breach prediction are
important for different users:
Flood risk assessment
Planning
Scheme design
Emergency planning
Flood event management
Emergency repairs
Peak discharge? Flood volume? Rate of flooding?
Time to catastrophic failure? Size of breach?
Different types of breach processes
The stages in breach development comprise:
• Initiation
– Surface protection (grass) cover fails, soil starts to erode
• Formation
– Significant erosion of material through embankment body and
down to base
• Growth (widening)
– Open breach; flow continues to widen the breach
Can be driven by wave overtopping, water overflowing or
seepage through
Different types of breach processes
Soil type and state affects the soil erodibility
Soil erodibility affects the overall physical process
...which (for external erosion) may be headcut erosion
or surface erosion (or a combination / transition
between the two).
Typically:
• Sandy, non cohesive soil = surface erosion
• Cohesive, clayey soil = headcut
[not forgetting that soil state significantly affects the process...]
Different breach processes
Different breach processes
Photos: Greg Hanson - USDA-ARS
What’s the point here?
1. Rate of flow through a breach is controlled by the
upstream crest level
–
–
–
Breach formation occurs when this control erodes since flow
can then increase, so increasing erosion
The timing of crest erosion for headcut or surface erosion
varies
Hence the nature of flood hydrograph depends upon the form
of erosion, hence the soil erodibility, hence soil type and state
What’s the point here?
2. The nature of the flood hydrograph also depends
upon the reservoir stage/volume relationship and the
soil erodibility
Assuming the crest erodes (surface erosion)
Hydrograph shape = function of rate crest erosion and
reservoir surface area
• Can the reservoir level drop to match the rate of crest
erosion?
Effect of Soil Erodibility on Breach - Outflow
100
M1 - Kd=0.2
M1 - Kd=18
M1 - Kd=0.9
M1 - Kd=35
M1 - Kd=3.5
M1 - Kd=88
M1 - Kd=17.7
M1 - Kd=177
90
80
70
Flow (m3/s)
–
–
60
50
40
30
20
10
0
0
500
1000
1500
2000
2500
Time (Seconds)
3000
3500
4000
4500
5000
[2a] Current options for predicting breach
Different methods include:
1. Judgement (guess work?)
2. Simple predictive equations
3. Simple predictive models
4. More complex predictive models
5. Integrated breach and flow models
6. Probabilistic breach models
Current options for predicting breach
1. Judgement (guess work?)
–
–
Can vary from reasonable judgement to a blind guess hence
can be highly inaccurate
Use of historic records from same catchment / soil / structure
type helps
2. Simple predictive equations
–
–
–
–
Typically developed for dams or non cohesive soils
Regression analysis on limited data (and hence specific
conditions)
Can have large uncertainties
Typically give peak discharge, maybe breach width – not
hydrograph
Current options for predicting breach
3. Simple predictive models
–
Predefines and simplifies the breach process
•
Is the simplification acceptable?
4. More complex predictive models
–
–
Greater complexity – less uncertainty
typically a slower model
5. Integrated breach and flow models
–
Allows for drowning of breach – important but don’t overlook
what the model actually does
6. Probabilistic breach models
–
Gives more information regarding range of possible
behaviour
[2b] Some recent developments in
predicting breach
i.
ii.
iii.
iv.
Advances in rapid simulation – FRMRC2
Conclusions from the DSIG breach modelling project
Advances in more detailed modelling – HR BREACH
Advances in understanding soil erodibility
[i] Advances in rapid simulation:
FRMRC2 WP4.4 research
Goal of the research
WP4.4 (University of Oxford and HR Wallingford):
Rapid breach assessment - will develop simplified
equations for the rapid prediction of breach size for a
limited range of embankment structures. The methods
will be directly applicable to practicing engineers and
will replace the default and very approximate breach
modelling methods currently included within the RASP
family of tools.
Researcher = Myron van Damme – University of Oxford
In summary...
Myron has:
1. Reviewed breaching processes and models
2. Chosen to develop simple, physically based equations
–
NOT regression analysis; Not emulating other models
3. Equations simulate average erosion rates for the
different types of breach process (surface / headcut /
internal erosion)
4. Assumes predefined physical processes for breach
5. Validated the model [AREBA] using HR BREACH and
DSIG case study data [ builds upon international best]
Underlying assumptions (1/4)
General assumptions
 1D flow behaviour
 Rectangular spatially
constant cross section breach
 No equilibrium transport
conditions
 Constant soil erodibility
 Instantaneous failure grass
cover
 No erosion below the
foundation level of the
embankment
 Widening rate is proportional
to the downward erosion rate
Underlying assumptions (2/4)
Assumptions surface erosion
Landside slope remains equal
to the initial slope gradient
 Depth along landside slope
approaches the normal depth
starting from the critical depth
 Landside slope retreats at a
spatially averaged erosion rate
 Crest erodes downward while
landside slope retreats
Underlying assumptions (3/4)
Assumptions headcut erosion
 No downward erosion due to
flow over the crest
 Headcut starts at the top of
landside slope
Underlying assumptions (4/4)
Assumptions piping
 An initial pipe diameter widens in
an equal rate due to the flow
through the pipe
 After slumping of the soil above
the pipe, potential further failure
of the embankment is described
by the surface erosion failure
process
 The grass cover is assumed to
have failed with failure of the pipe
Performance and validation
• AREBA - A Rapid Embankment Breach Assessment
• AREBA gives promising results when being bench marked
against HR BREACH
• Validating AREBA against the DSIG data showed that the model
prediction lie within the bounds of uncertainty following from
the uncertainty in the input parameters
• Run speed AREBA is approximately 0.2s per run.
Data Requirements / Applicability
• AREBA predicts breach through simple, homogeneous
embankments
– Does not simulate composite or complex structures
– But speed of simulation does allow the user to ‘play’
• AREBA requires definition of:
–
–
–
–
–
Upstream load (time varying water level)
Embankment geometry
Embankment soil erodibility, Kd
Assumed failure mode (surface, headcut, internal erosion)
Downstream volume / stage (time varying water level)
[ii] Conclusions from the DSIG breach
modelling project
• International project – dam owners – industry needs focus
• Breach model review  3 selected for evaluation (as greatest
potential for industry uptake) = SIMBA / HR BREACH /
FIREBIRD
• Data review to assist model performance validation
– USDA ARS data / IMPACT project data / 2 dam failures
• FIREBIRD eventually rejected due to lack of usability and hence
testing
DSIG Conclusions (Cont.)
• SIMBA = Headcut; HR BREACH = surface erosion
• Performance of both models against the various test cases
varied, but neither was consistently better or worse
• Conclusions:
– Still to be formally reported by PM (Tony Wahl – USBR) ( not formal conclusions)
– Develop an industry model using both headcut and surface erosion
processes (ie SIMBA + HR BREACH)
– Modeller understanding and ability with the models is important
• For example, models simulate breach through simple structures; expertise is needed
to know how to apply the models to more complex, real structures
– Available data to support research and model validation is limited
– Uncertainty / lack of knowledge in soil erosion processes / simulation
[iii] Advances – HR BREACH model
•
HR Wallingford programme of research over past
decade
–
–
•
•
National / EU / International projects plus company R&D  HR
BREACH model
Mohamed (2002); Integrated InfoWorks (Wallingford Software
 Innovyze); Morris (2011)
1D flow; sections through embankment; 2D section
erosion plus block failure
Focus here on one area of advance – breach formation
through zoned flood embankments / embankment
dams
Zoned embankments
Trying to predict breach through
real rather than idealised
structures
How significant is this for flood
risk assessment and hence asset
management practice?
Representing zones in the model
1
1
1
2
3
2
4
1
1
2
3
2
4
1
1
2
3
2
4
1
1
2
3
2
4
2
3
Effect of Kd versus As
•
Soil erodibility versus reservoir stage area
Homogeneous embankment breach – two different
erodibilities
Peaky
Hydrograph
Flat
Hydrograph
Effect of zoned embankment (2 layers)
•
•
2 flat layers – different erodibility (also fn of reservoir area)
Extremes show peaky versus flat hydrograph (trends in-between)
Homogeneous embankment breach – two layers with
different erodibilities
Erodible top
layer
Resistant
top
layer
Effect of layers on outflow hydrograph
•
2 layers – different erodibility
Advances – HR BREACH model
Conclusions:
• The combined effects of zoning and soil erodibility
can significantly affect the breach flood hydrograph
• Layering can provide significant degrees of protection
against more erodible material beneath
–
•
•
Ie. the Dutch / German embankment design
Erodibility plus reservoir area affects flood
hydrograph
Each of these points is significant for design /
construction / maintenance / risk assessment
[iv] Advances in understanding soil
erodibility
 Greg Hanson  
[4] Industry Practice
Industry practice:
• Often guesswork / judgement or simple equations
Why?
• Confusion over choice of best (appropriate) method?
• Avoid the cost
–
–
•
By avoiding perceived complexity of analysis
By avoiding purchase of software
Blame uncertainty
Systems analysis:
• Risk models run many thousands of simulations 
need simple and / or fast methods of prediction
[5] Future Direction
1. Industry versus academic solutions – the challenges
are different
–
Fast – simple   Complex – slow(er)?
2. Need to integrate flow, with soil erosion, with
structure response / behaviour
–
–
–
Complex / significant interdependencies
Can reduce the uncertainty from flow by moving to a 2D, 3D
flow model (using existing models and improved computer
power)
Need to keep a balance in terms of analysis accuracy &
relevance for different processes
3. Need to improve understanding of soil erodibility
[5] Future Direction
4. Need to improve understanding of natural and man
made variability in soils / structures in order to assess
likely erosion / failure processes
5. Different soils erode by head cut or surface erosion –
currently as separate models  merge into a single
model with analysis sufficient to determine and flip
between processes
[5] Future Direction
5. Putting model speed aside, use more accurate flow
models combined with a better analysis of soil
behaviour... merge a soils analysis model (ie. Levee
stability analysis) with a breach model to provide the
complete assessment of levee response and failure
6. Given the uncertainty in soil parameters / distribution,
consider how this may be done probabilistically as
well as deterministically
[6] Research Gaps
(i) Soil Erodibility:
1. Understand which parameters affect erodibility (close)
2. Understand how those parameters vary in-situ and
under climate change conditions
3. Link parameters to field observations for asset
management
4. Link parameters to design and construction
specifications
[6] Research Gaps
(ii) Breach model:
1. Meshing headcut and surface erosion, with analysis to
flip between processes
2. Meshing ‘geotechnical analysis’ model with a ‘breach’
model
3. Dealing with uncertainty distribution?
[6] Research Gaps
(iii) Data:
We always seem to be chasing better data!
1. Despite the frequency of dam and levee failure there
remains very little high quality, field scale data against
which the different processes can be studied and
models validated. Solution?
2. Soil erodibility data – If we accept process
dependence upon soil erodibility, and we build models
using measures of erodibility, how can we provide
best guidance on value selection?
More information on breach?
Website information
The EU FP6 Integrated Project – FLOODsite - Tasks 4 & 6:
www.floodsite.net/html/work_programme2.asp?taskID=4
www.floodsite.net/html/work_programme2.asp?taskID=6
The CEATI Dam Safety Interest Group breach modelling project:
www.ceati.com/collaborative-programs/generation/dam-safety
The FRMRC II project website: www.floodrisk.org.uk
Contact: Mark Morris mark.morris@samui.co.uk