Flood risk assessment and flood modelling Jane Toothill

1
Flood risk assessment and flood modellingJane
Toothill

• Reinsurance Association of America, London,
  June2007

2
Reasons for uncertainty in flood model results

• Probabilistic flood models are very complex and
  sources of uncertainty are larger than for other
  perils
• Are heavily reliant on input data quality
• Can use a wide range of methodologies
• Are not yet as widely used other peril models
• There is as yet little available European loss
  data against which to calibrate results

• Agenda
• Flood modelling methodologies
• Which components are the largest potential
  sources of uncertainty?
• What can we do to reduce uncertainty?
• The future of flood modelling?

3
Flood hazard module

• Many advanced scientific techniques are available
  for modelling flood
• Flood models in the insurance industry broadly
  fall into two categories
• Developed from rainfall data

4
Flood hazard module

• Many advanced scientific techniques are available
  for modelling flood
• Flood models in the insurance industry broadly
  fall into two categories
• Developed from rainfall data
• Developed from gauge station data

• Both approaches require vulnerability and
  exposure modules to enable loss calculation

5
What is off-plain flood?

• No model can capture streams of ALL sizes using a
  river-based approach
• Small streams and other sources of flooding (e.g.
  Related to heavy precipitation) are considered
  separately

• Two approaches are used to model off-plain flood
• Statistical
• Simple and potentially effective but how does
  it relate to the on-plain component?
• Developed from run-off modelling
• Scientific approach that fits well with
  rainfall-based river modelling. But more time
  consuming to develop... does it generate better
  results?

6
What is the effect of the off-plain component on
the modelled results?

• Lack of consensus as to the relative level of
  importance of on and off-plain flood!

7
Flood defences are complex structures and vary
along the river
8
Flood defence failure modelling

• It is not practical to walk the length of the
  river!
• Hence assumptions must be made

• Defence failure is typically modelled using
  defence fragility curves which associate the
  level of water below the defence crest with
  failure probability

9
Flood defence failure modelling

• What is the input information? Does it provide
  adequate information to assign the right curve to
  the defence?
• Input data is ideally
• Location, height, structure and standard of
  maintenance
• Plus engineering assessment to define failure
  curves

• But more typically is based on assumptions
  relating to
• Design return period of defences (may be related
  to population)
• Information from Digital Terrain Model
• Can these curves be derived and validated?
• Some defences are even harder to model...

10
Flood defence failure modelling

• Effect on results can be very large, especially
  locally
• Tendency for defences in model to fail at set
  return periods, leading to sharp rise /
  flattening of loss exceedance curve
• Worse if propagation model is poorly defined

• Can flood defence modelling realistically be
  improved?

11
Vulnerability functions

• Todays flood models relate damage to hazard
  intensity, typically water depth
• There is little historical information to help
  validate these functions

12
The relationship between water depth and claims
information

• Water depth is not the only parameter to affect
  the level of damage
• The source of the water depth information to
  which claims data are compared is not generally
  from the site of damage

13
The relationship between water depth and claims
information

• Effect on results
• Engineering vulnerability functions tend to be
  too conservative
• Business interruption is rarely well modelled
• Lack of consistency between and within models

• Can the current approach to vulnerability
  modelling be improved?
• Short term Greater reference to claims
  experience can help calibrate existing damage
  functions
• Longer term The validity of the current water
  depth-based approach requires consideration

14
What can users to reduce uncertainty?

• Make sure you understand the model youre
  using...
• What is the scope?
• Does it match the contract terms?
• Is it the best match to your requirements?
• Work on your portfolio data
• Better data better results!
• Check for differences between model assumptions
  and your portfolio
• Is a bespoke analysis required?
• Do you have claims information that can be used
  to calibrate vulnerability functions in the
  models?
• Consider multi-model use
• Use past event experience and detailed
  deterministic models to benchmark the loss
  exceedance curve

15
Role of deterministic models in benchmarking

• High quality deterministic models can use more
  detailed data than is possible in countrywide
  probabilistic solutions
• Can complement commercial models by providing
  highly accurate results for
• Calculation of total exposure to the worst case
  scenario
• Expected loss to specific scenarios linked to a
  return period
• Hence ability to calibrate probabilistic vendor
  models
• Modelling assumptions can be tailored to match
  specific individual portfolios

16
Flood modelling tomorrow?
Increased interaction with scientists and water
management authorities to enable access to
improved data for modelling (esp. defences)
Improvements in data quality and processing power
leading to more detailed and accurate models
Increased usage of flood models and settling of
results
Improved methods of modelling vulnerability
functions
Increased provision of claims data to modelling
companies for model calibration
Increased recognition that one model does not fit
all
17
Conclusions

• Flood losses in Europe are comparable to
  windstorm losses
• More events but less severe
• Low flood cover penetration might change in
  future causing overproportional increase of
  losses
• Flood models are complex and expensive
• Many potential source of uncertainty
• However, many flood risk assessment tools for
  primary insurance and reinsurance UW are now
  available
• Probabilistic models show some degree of
  consistency on a market level
• However, larger differences remain for regional
  portfolios and for different LOBs
• Multi-model use helps to reduce uncertainty
• Models are greatly improved over 5 years ago
• However, there remains plenty of room for future
  developments / improvements