Last year we had the opportunity to attend the 2017 Cat Risk Management and Modelling Conference in London, where we were invited to participate in a flood model comparison session.
The session explored the underlying methodologies and key features of the three leading UK Probabilistic Flood Models on the market. This comparison enabled clients who do not currently licence models, and those that don’t license all the models, to understand the key differences in results that may impact their business, allowing them to make a more informed decision when choosing a flood model in the future.
The Current Cat Modelling Approach
It was interesting to see that, in common with other models available in the market, JBA’s model uses similar input datasets from organisations such as the UK Environment Agency, CEH, Met Office, SEPA, DEFRA and Ordnance Survey, as well as broadly similar methodologies. For example, each model contains hundreds of thousands of simulations of future flood events and contains a combination of river, surface water and coastal flood type.
All use depth as the key flood hazard parameter driving damage (as opposed to velocity, debris flow or duration) and this is derived for each type using 1D or 2D hydraulic models. This hazard is then related to damage by a series of depth-damage vulnerability functions and users can output losses for a (varying) range of deterministic scenarios.
On the surface then, it is not surprising that the models gave broadly similar results at portfolio level. At a more granular level, results varied more significantly, indicating that the models had significant differences in data vintage and the underlying resolution at which loss calculations were carried out.
Understanding JBA’s latest catastrophe model
Our 2018 update to our UK Flood Model builds on the methodologies and features of the 2015 model to solidify its position as the most advanced probabilistic flood model on the market. The model has several key features that sets it aside from other models available.
High-resolution analysis
This model is the first to go beyond postcode resolution and differentiate flood risk by individual building. The UK Flood Model incorporates our market-leading river, surface water and coastal flood maps at 5m resolution.
The surface water maps are modelled using a direct rainfall approach capturing the routing, ponding and pooling of water at 5m resolution. Analysis at building resolution is particularly key for surface water as flood water tends to be channelled along roads and pools in natural depressions, which are usually further away from where buildings are constructed (pictured). This high-resolution ensures only flood risk that actually affects the building is considered and gives users of the model an advantage over their competitors as risk selection and risk pricing can be much more accurate.
Sophisticated application of science
Our stochastic event set is based on an advanced methodology which builds on academic spatial dependence models. Importantly, the methodology is based on continuous rainfall simulation that drives the generation of river and surface water flood events in a correlated and hydrologically robust way.
The continuous rainfall simulation offers two further important advantages. First, it captures antecedent conditions where prolonged rainfall may cause the ground to become saturated. Seasonal changes in flood event type and severity are taken into account so that summer downpours are captured alongside winter storms. Secondly, the method allows flow in ungauged catchments to be modelled using rainfall-runoff models from neighbouring catchments with similar hydrological characteristics.
This enables intense, localised flood events, such as the Boscastle event in 2004 (pictured) and the Cockermouth flood in 2009, that are not previously captured in ungauged locations to be included within the event set. This allows our model to much more accurately estimate attritional flood losses.
Data to reduce uncertainty
Finally, by including a greater range of historic events including North Sea flood 1953; Easter 1998; Autumn 2000; May, June and July 2007; Storm Xaver 2013; Thames 2014; and Storms Desmond, Eva and Frank 2015/16, users can have much greater confidence in the modelled loss output, as they can run their own portfolios against the events and compare to observed losses.
It’s clear that competition between flood models is heating up, with each vendor presenting their best estimate of flood risk. Understanding the key differentiators between models is vital for users to understand how the models have been built, how up to date the datasets are within it are and how well the model estimates both portfolio level and individual property level losses. Without this information, it’s impossible for users to formulate their own accurate view of risk.
We welcome another comparison session to showcase our new UK Flood Model and to demonstrate to clients the key differentiators between leading models. If this is something you’d like to see, or if you have any questions regarding the model, get in touch and let us know!
