Skip to main content
Article

The first MENA flood model specific to the region

Climate|Reinsurance
Climate Risk and Resilience

By John Alarcon , Anand Kulkarni and Charlotte Miller | September 29, 2021

Our recently released flood model represents the latest example in the chain of products to support defining a View of Risk, which also incorporates the possibility of exploring climate change impact.

We are committed to developing risk solutions and models for the region such as our proprietary MENA Earthquake model and new Flood solution.”

John E. Alarcon
Executive Director, Willis Re International

Commercially available catastrophe risk models in the Middle East and North Africa (MENA) region are limited. Hence, in order to support our clients better understand their exposure to natural perils, in conjunction with the Willis Research Network (WRN), Willis Re has been committed to developing risk solutions and models for the region, such as previously done with our proprietary MENA Earthquake model.

As a result, a new Willis Re MENA flood model has been released in 2021, the first of its kind for the region to be developed. It enables (re)insurers to quantify for the first time their risks to this peril in the region using a probabilistic catastrophe model and the ability to stress test for climate change impacts from the hazards.

Currently the model covers Morocco, the United Arab Emirates and the Kingdom of Saudi Arabia (KSA), with Oman, Qatar and Egypt being released during the third and fourth quarter of 2021. Further country developments will follow in 2022.

Why the MENA flood cat model matters

Floods in the region are relatively common, representing a high frequency, low severity risk for the region in comparison to the tail earthquake risk quantified already by the Willis Re MENA Earthquake Model. With the predicted impact of climate change, these events are likely to become more extreme in the future, thus making it important to quantify this risk.

Damage from floods comes from both pluvial and fluvial flooding in the region, with fluvial flooding being more predictable due to proximity to river networks. Pluvial flooding is more unpredictable due to the arid desert environment, which after intense rainfall may cause runoff and flash floods. This is exacerbated in urban areas where the soil is covered by impermeable man-made surfaces. Wadis (dry river-beds) also amplify this by channelling water after intense precipitation, causing unpredictable and localised flooding.

These flood events have the capability to cause important economic and insured losses. Figure 1 shows the scale of the economic losses from recent floods in the region. Jeddah is a particular city of note, having suffered from multiple large events in recent years. The 2009 Jeddah floods, for example, caused $900m economic losses and 191 deaths, whilst the 2011 floods caused >10 deaths and $300m economic losses. Such floods are likely to impact the lower return periods of the exceedance probability curve, therefore impacting (re)insurers profitability and making it more and more vital for (re)insurers to mitigate this profit volatility by modelling the risk.

This graph illustrates Flood Economic Losses in the MENA Region In US dollars covering Algeria, Egypt, Lebanon, Morocco, Qatar and Saudi Arabia.
Figure 1: Flood Economic Losses in the MENA Region (Various Sources)

Flood model applications

There are several business applications from the model outputs, including:

  • Structuring and pricing reinsurance layers for purchasing suitable protection
  • Assessing capital adequacy and responding to regulatory solvency requirements
  • Portfolio management and optimisation
  • Responding to rating agency requests

How do we model flood for the region

The recently launched model was developed through the collaboration of Willis Re’s in-house peril experts with the leading flood specialists KatRisk – our global partner for flood solutions-, and resources from the Willis Research Network (WRN). The very latest flood hazard layers and loss aggregation methodologies are used to generate location level losses, as well as portfolio loss outputs using a 50,000 year simulation period. The latter aims to cover all possible flood scenarios.

Willis Re uses KatRisk’s flood hazard layers for different return periods, including both pluvial and fluvial sources of flooding. These hazard layers provide water depths for a number of return periods, as shown in the example presented in figure 2 which highlights KatRisk’s quality against others.

This figure shows the footprints of the flood layers from three different sources, one of them being KatRisk.
The figure highlights the better quality and more natural distribution of the flood cover distribution from KatRisk compared to the other two sources.
Figure 2: Sample comparisons of hazards layers with other source layers

KatRisk has developed these hazard layers based on modelling stochastic precipitation from historical precipitation records. A rainfall to runoff transformation model is then used to estimate the runoff and route the surface runoff from all the contributing catchments. 

Using the hazard and vulnerability information, Willis Re performs event simulations and computes the location level return period (RP) losses and location level Exceedance Probability (EP) curves.

How do we address coarse geocode level exposures

High resolution hazard vs low resolution exposure

The high-resolution nature of the flood peril compared to the low-resolution exposure data often associated with MENA territories presents a unique challenge for flood modelling. The Willis Re model offers a thorough solution to this challenge through our innovative exposure disaggregation module. In order to accurately capture the flood hazard at a high resolution the exposure needs to be disaggregated down to the appropriate geocoding level.

Willis Research Network sources for remote sensing, industry databases and the European Space Agency (ESA) satellite imagery was used to derive a representative view of the distribution of property risks at a 90 x 90 m grid resolution. Differing weights are applied for residential, commercial and industrial lines of business that depict the various spatial distributions across the territories (see figure 3).

This figure shows satellite images of three different zones in the UAE, about how at a fine grid resolution Willis Re assigns the Line of Business (i.e. Residential, Commercial and Industrial) coverage.
This is for data dissagregation since most portfolios come at very coarse resolutions, so for flood modelling it is important to dissagregate this coarse information geographically.
Figure 3: Sample weights being assigned for exposure disaggregation at individual locations

Two key primary modifiers impacting model results - number of building storeys and construction type are another common unknown in the exposure data in the MENA region. Willis Re undertook exposure characterisation leveraging data from the Global Earthquake Model (GEM) exposure database and local sources such as the Dubai Statistics Center and Housing census data, in order to assign representative vulnerability curves to each grid cell used in the exposure disaggregation. This methodology allows the incorporation of these important primary modifiers, overcoming to a good degree the limitations in the original data.

Vulnerability curves in data scarce regions

Vulnerability functions have several characteristics that relate to how flooding affects built structures. The principal causes of flood losses include deterioration of building components, rebuilding of defective components and replacing contents that are contaminated by the inundation of water. Even if the inundation depth is the same at two given locations, the flood damage may be different due to factors such as type of business, construction type and the height and age of the building.

Willis Re has identified and calibrated the dominant occupancy-construction type classes from more than 73 occupancy classes and 13 construction types available in the framework, based on the individual country’s building and construction types for residential, commercial and industrial lines of business. This facilitates the robust vulnerability modelling of the different countries and building types. To account for the loss uncertainty arising out of damage variability to the exposures, the framework employs a probability distribution when computing losses. 

Does the model match up to real events?

The final and key component of building a model is how the modelled losses compare to real events and market experience. Willis Re has used a combination of methodologies in order to calibrate the flood model to overcome the limited claims availability in the MENA region.

Willis Re has used a combination of methodologies in order to calibrate the flood model to overcome the limited claims availability in the MENA region.”

John E. Alarcon | Executive Director, Willis Re International

Available claims data was used alongside cross-validation with expert judgement. As an example (figure 4) the 2009 and 2011 Jeddah flood events were used to validate the model in the KSA. The claims data (indexed to present day) was compared against the flood model EP curve to determine whether the modelled losses produced an appropriate return period for the two events. We have found that the new flood model aligns well with the claims benchmarks, which we would gladly discuss individually with our clients/prospects in the region.

This figure shows an overall Exceedance Probability (EP) curve for the Kingdom of Saudi Arabia (KSA), and where the flood events from 2009 and 2011 sit on the curve.
The figure does not have axis values as to protect our IP.
Figure 4: Historical claims superimposed on the Willis Re flood curve, indicate good alignment with the Jeddah events

Model response to climate change sensitivities

Climate change studies indicate that intensity of precipitation extremes increases more strongly with global mean surface temperature (e.g. Myhre et al ,2019). Willis Re can use such information to assess the impact on losses for increases in temperature for a range of temperatures. Figure 5 shows the impact on the Occurrence Exceedance Probability (OEP) losses for a disaggregated Global Exposure database for KSA for two scenarios:

  1. Current scenario and
  2. 2°C rise in temperature.
This figure shows two Exceedance probability curves, one showing the current conditions and one considering the estimated impact of climate change with a two degree Celsius rise.
The curve with the climate change included shows higher loss estimates at all return periods.
Figure 5: Impact on OEP curve for climate change temperature sensitivity

This analysis only covers the rescaling of hazard severity by return period, and no other parameters.  

It can be observed that there is noticeable increase in losses for a 2°C rise in temperature across all return periods. Such analysis will be increasingly important for performing climate sensitivity stress tests for (re)insurers in order to determine how climate change could impact their portfolios.

Concluding remarks

Willis Re is fully committed to servicing our clients in the region and thus, over the last years have invested heavily in research and development in order to remain at the forefront in our advisory role. Our recently released flood model represents the latest example in the chain of products to support defining a View of Risk, which also incorporates the possibility of exploring the climate change impact. We are very proud of putting this additional tool at the service of our clients in the region.

With acknowledgements to our team members Ila Chawla, Shivangi Singh, Kishor Ghatage, Alex Saunders and Gunin Gogoi who have contributed to the development of our model, including inputs from ESA and GEM via the Willis Research Network.

Authors

Executive Director Catastrophe Analytics, Willis Re International

Dr John E. Alarcon holds the position of Executive Director in Willis Re, where he carries out a couple of complementary roles, one of them being the Head of Catastrophe Analytics for the Middle East and North Africa (MENA). He has a PhD on Engineering Seismology and over 17 years of post-graduation industrial experience in the fields of catastrophe modelling from natural hazards and as Reinsurance broker.


Technical Support Team Supervisor | Model Development & Platforms | Willis Re

Dr Anand Kulkarni is a senior associate in Model Development and Platforms team at Willis Towers Watson, WTW GDS India Pvt Ltd since 2019. He is a Flood Scientist who has over 9 years of experience working in hydrology and hydraulics and with 6 years in Catastrophe Model development.

After graduating from IIT Bombay in 2015, Anand has worked as Flood cat model developer with RMS Risk Management Solutions India Pvt Ltd, with focus on developing hazard layers across different geographies of the world.


Catastrophe Risk Analyst - Willis Re

Charlotte has worked at Willis Re for 3 years focussing on catastrophe analytics in the MENA region, quantifying client’s risks and employing various solutions to meet client’s needs. She has also contributed to the Earthquake and Flood View of Risks for the region. Her background is in Geophysical hazards and Earth Science having completed a Masters Degree at University College London (UCL) and a BSc from the University of East Anglia.


Related Capabilities

Contact Us