www.guycarp.com probabilistic flood modelling in eastern europe icar forum, 1 st -2 nd october 2007...
Post on 19-Dec-2015
219 views
TRANSCRIPT
www.guycarp.com
Probabilistic Flood Modelling in Eastern Europe
ICAR Forum, 1st -2nd October 2007
Silke Huebner, MunichInstrat® Cat Modelling CEE
2
Guy Carpenter
Agenda
Introduction
Types of Natural Catastrophe Models
Structure of a Probabilistic Flood Model
Flood Modelling Issues in Eastern Europe
Probabilistic Flood Modelling in Eastern EuropeProbabilistic Flood Modelling in Eastern EuropeProbabilistic Flood Modelling in Eastern EuropeProbabilistic Flood Modelling in Eastern Europe
3
Guy Carpenter
Guy Carpenter in CEE
Why is Guy Carpenter dealing with such topics?– GC is one of the leading reinsurance brokers– There’s a growing demand for natural catastrophe analyses– GC has its own analytical unit (Instrat®), with the following core services:
Actuarial analysis Reinsurance structuring Risk management consultancy Natural Catastrophe Modelling
– GC not only licenses the commercially available models (e.g. AIR, EQECAT, RMS) but also develops internal models on its proprietary G-CATTM platform
– CEE modelling support is provided by the Munich Instrat® team– The European Model Development team, with 10 members across Europe,
deals with the internal model developments
4
Guy Carpenter
Flood Hazard in Central and Eastern Europe
Source: UNEP/GRID-Geneva
Flood is a big topic!
5
Guy Carpenter
Agenda
Introduction
Types of Natural Catastrophe Models
Structure of a Probabilistic Flood Model
Flood Modelling Issues in Eastern Europe
Probabilistic Flood Modelling in Eastern Europe
6
Guy Carpenter
Types of Natural Catastrophe Models
Should I underwrite this risk?
What may happen in a specific scenario?
What‘s the probability for a loss
≥ x?
Zonation- or Rating Models
Deterministic Models
Probabilistic Models
7
Guy Carpenter
Types of Natural Catastrophe Models
All types of models have value
Zonation models are very valuable during the underwriting process of primary insurers or facultative reinsurers
Zone IZone II
Zone III
But for risk management purposes / reinsurance considerations– millions of risks have to be analysed,– losses have to be estimated for various probabilities and– it should be possible to do analyses for combined perils
Probabilistic models are the most suitable to meet those conditions
8
Guy Carpenter
Agenda
Introduction
Types of Natural Catastrophe Models
Structure of a Probabilistic Flood Model
Flood Modelling Issues in Eastern Europe
Probabilistic Flood Modelling in Eastern Europe
9
Guy Carpenter
Structure of a Probabilistic Flood Model
Analysis Module
HazardModule
Probabilistic event set
Import Module
Client data with location information
Built environment
Module
horizontal and vertical distribution
of the risks
LossModule
Loss in relation to flood intensity
Results
Losses per return period, event sets
FinancialModule
Consideration of deductibles, limits and reinsurance
structures
10
Guy Carpenter
Import Module
Module is used to import the formatted client dataset into the model environment:Location, number of risks, total insured value, deductibles, limits
Flood hazard can vary over short distances
Knowledge of exact risk location important
But: Sufficient address level exposure data is often unavailable in CEE countries
Therefore: Models must be able to cope with coarser data
Reliable results can still be achieved by using detailed land use data for disaggregation
11
Guy Carpenter
Structure of a Probabilistic Flood Model
Analysis Module
HazardModule
Probabilistic event set
Import Module
Client data with location information
Built Environment
Module
horizontal and vertical distribution
of the risks
LossModule
Loss in relation to flood intensity
Results
Losses per return period, event sets
FinancialModule
Consideration of deductibles, limits and reinsurance
structures
12
Guy Carpenter
Built Environment Module
Building census data
Land use / cover
Flooded area
Client portfolio
Modelling units
The built environment represents theinterface between flood maps and clientdata and has two functions:– To allow the spatial redistribution of
client data into the modelling units– To complete unknown client data
characteristics based on buildingcensus information
It works both horizontally and vertically
It not only tells where are risk is likely to be located but also if it’s rather a multi-storey building or a single family house
Bu
ilt
en
vir
on
me
nt
13
Guy Carpenter
Structure of a Probabilistic Flood Model
Analysis Module
HazardModule
Probabilistic event set
Import Module
Client data with location information
Built environment
Module
horizontal and vertical distribution
of the risks
LossModule
Loss in relation to flood intensity
Results
Losses per return period, event sets
FinancialModule
Consideration of deductibles, limits and reinsurance
structures
14
Guy Carpenter
Hazard Module
The hazard module defines the characteristics of modelled flood events:
– Water heights– Flood extents– Event frequencies
The probabilistic event set attempts to quantify the entire spectrum of risk by defining a representative subset of all possible future scenarios and their relative frequency
The historic event set is used to define what is representative Water heights and flood extents and can be modelled by using two
different approaches:– Simple GIS approach
– Hydraulic modelling techniques
Flood intensity
15
Guy Carpenter
Hazard Module
Simple GIS approachHydraulic modelling
techniques
Which area is flooded if the water level rises by x metres?
Disregards water flow constraints
Water levels at confluences are difficult to capture
How does the water propagate according to various parameters of main rivers and tributaries?
Approach used by GC in the current CEE development projects
Both approaches can only produce reasonable results if an accuratedigital elevation model is used for modelling
16
Guy Carpenter
Hazard ModuleHydraulic Modelling of Flood Extents
Source: wxmaps.com
Rainfall
Rainfall drainage
Routingand water heights
Flood defence systems
Defense failure and lateral propagation
Discharge [m³/s]
0
2000
4000
6000
8000
10000
12000
Height [cm]
0
200
400
600
800
1000
1200
Advantage of starting the modelling with rainfall data:
Correlations between rivers are easier to determine
Advantage of starting with water height and discharge data:
Difficulties with the modelling of the rainfall drainage are avoided
17
Guy Carpenter
Hazard Module
Example of hydraulic flood propagation modelling used in a GC proprietary model:
18
Guy Carpenter
Structure of a Probabilistic Flood Model
Analysis Module
HazardModule
Probabilistic event set
Import Module
Client data with location information
Built environment
Module
horizontal and vertical distribution
of the risks
LossModule
Loss in relation to flood intensity
Results
Losses per return period, event sets
FinancialModule
Consideration of deductibles, limits and reinsurance
structures
19
Guy Carpenter
Loss Module
The loss module assigns the degree of loss if a building is affected
The loss degree depends on– Flood intensity– Type of risk (residential, commercial, industrial etc.)– Coverage (buildings, contents etc.)– Building type and occupancy
Vulnerability functions have to be calibrated for each country and preferably each client
The best calibration can be achieved if detailed loss data from recent events exists and is provided by the clients(best example: Czech Republic)
20
Guy Carpenter
Structure of a Probabilistic Flood Model
Analysis Module
HazardModule
Probabilistic event set
Import Module
Client data with location information
Built environment
Module
horizontal and vertical distribution
of the risks
LossModule
Loss in relation to flood intensity
Results
Losses per return period, event sets
FinancialModule
Consideration of deductibles, limits and reinsurance
structures
21
Guy Carpenter
Results of a Probabilistic Model
Probabilistic models have two main outputs:
“Event Set”:– Contains the frequency and modelled loss for each simulated event– Can be used in risk management / dfa-tools for comprehensive risk
analyses (e.g. testing of reinsurance structures, multi-peril analyses)
“EP-Curve” (EP: exceedance probability)– Gives the probability that loss x is exceeded– Either in tabular or graphical format
Analysis I Analysis II Analysis III Analysis I Analysis II Analysis III
5 14.727.104 23.713.066 9.276.871 16.262.650 31.845.940 18.631.404
10 65.854.092 58.020.271 16.281.915 70.855.336 70.983.665 27.958.899
25 201.559.370 135.084.090 37.475.027 213.446.363 156.926.396 49.715.822
50 316.440.125 225.012.203 62.673.739 328.389.772 252.025.210 75.952.504
75 368.197.040 288.982.531 82.594.442 393.486.468 318.325.227 96.462.415
100 424.264.333 334.788.550 99.819.777 457.453.067 365.847.504 114.022.482
150 507.597.027 411.915.607 129.222.490 525.654.311 441.278.402 143.833.144
200 561.134.055 476.148.538 154.163.072 577.057.361 504.888.184 168.983.025
250 602.519.918 537.789.523 175.945.806 638.314.717 563.657.726 190.921.548
500 782.828.342 692.894.465 257.498.219 836.787.271 729.387.938 272.817.736
1.000 926.986.055 907.975.913 358.247.804 988.970.599 944.627.727 373.957.146
Return Period (years)
Loss per Occurence (EUR) Annual Loss (EUR)
0
100
200
300
400
500
600
700
800
0 50 100 150 200 250 300 350 400 450 500
Mio
Return Period (years)
Lo
ss
(E
UR
)
Analysis I
Analysis II
Analysis III
22
Guy Carpenter
Agenda
Introduction
Types of Natural Catastrophe Models
Structure of a Probabilistic Flood Model
Flood Modelling Issues in Eastern Europe
Probabilistic Flood Modelling in Eastern Europe
23
Guy Carpenter
Special Flood Modelling Issues in CEE
Compulsory insurance schemes are discussed more and more and probabilistic (flood) modelling results represent a vital part for pricing and structuring
The CEE insurance market is growing and with it the demand for flood covers and analyses
Data availability and quality– Low insurance density and rather coarse client data recording in many
parts
Limited availability of detailed client and technical data provides challenges that can be overcome by detailed built environment modelling
Cross border correlation often one flood event affects more than one country
24
Guy Carpenter
Cross Border Correlation
Flood events do not stop at borders!
Especially in Central and Eastern Europe floods tend to affect more than one country
The cross border correlation needs to be considered for detailed multi-country analyses
But: It‘s certainly more important to have some risk measurement tool in place before starting to consider correlations
And: Rather take the time to build a model step-by-step than trying to incorporate everything in one go
Event Affected Countries
July 1997 Germany, Poland, Czech Republic
June 1999 Poland, Czech Republic, Slovakia, Romania
August 2002 Germany, Austria, Czech Republic, Slovakia, Hungary
July 2004 Poland, Slovakia, Hungary
March 2006 Germany, Austria, Czech Republic, Slovakia, Hungary
25
Guy Carpenter
Data essentials for the development of a good probabilsitic flood model
Digital elevation model with adequate resultion 1km is definetly too coarse for a reliable flood model
River network information
Sufficient history of rainfall or gauge station informationHeight [cm]
0
200
400
600
800
1000
1200
Loss information to calibrate the vulnerability functions- preferably from the country modelled- from other countries with similar building stock
Information on land use and building stock
GC provides the know-how to bring all this together!
With probabilistic models into the flood insurance future!
www.guycarp.com