Some thoughts on the future of flood and Cat event modelling for you, by Dr Jannis Hoch, Senior Developer, Fathom and Harry Vardigans, Insurance Manager, Fathom.
The first flood models were built more than 30 years ago. Since then, they have evolved greatly. Currently, flood models can be run at a spatial resolution below 100 meters, thanks to an increase in both computational resources and advances in remote sensing. Most critically, the elevation data to accurately capture floodplain topography has reached unprecedented levels, and with TanDEM-X and derived datasets such as FABDEM, the accuracy of flood models is expected to make a giant leap forward. Currently ongoing remote sensing missions will further contribute to better observations of meteorological input to flood models (such as rainfall) or can help to validate model output. Together with powerful data science techniques such as Deep Learning, it can be expected that the skillfulness of remotely sensed data will only improve further.
All of this would not have been possible without continuous research-led efforts in both academia and industry. For any groundbreaking finding to have maximum impact, it is essential to maintain an open approach towards knowledge sharing across domains, as recently outlined by a report issued by the US President’s Council of Advisors on Science and Technology.
Despite these successes, there are still many challenges ahead. Specifically for flood modeling, knowledge about flood defenses is crucial, yet this knowledge is very scattered and often not available especially where it matters most.
Looking into the crystal ball
Another challenge is the insurance industry’s reliance on their own historic claims data for risk profiling. Doing so gives the insurer real ground-level information on the risk that a property – or portfolio of properties – faces to a certain peril. This claims data is often seen as the (re)insurers IP; targeting properties which are seemingly profitable, thus providing a competitive advantage over competitors without such detailed data.
However, although useful, we can no longer solely rely on historic claims data with the impact that a changing climate has on the risk landscape, particularly when considering flood. Using claims data which represents a past climate and therefore a different risk landscape, can often mean insurers could be under-pricing business.
Depending on the socio-economic landscape of the future, exposure may change drastically compared to the present day as well. Much work is done in climate-conditioning flood models, which is an important task to capture the intensification of the hydrological cycle. However, in some areas, changes to flood hazard can be dwarfed compared to changes in exposure as a result of socio-economic growth. Zooming in from a national to sub-national level, it is very difficult to estimate where exposure will be located in the future. While some research suggests that coastal areas will experience a net influx of people, such projections remain subject to a wide range of potential sources of error.
Asset pricing in an uncertain world
Other variables in a catastrophe (CAT) model can also change in the future. Take vulnerability curves. These curves are strongly reliant on data collected in the past years and decades. Constant innovation and technological developments, however, will most likely yield better resilience of assets to a hazard. This is not to mention that the way cities and infrastructure are planned and built may change at large, adapting to an increased threat to hydrological hazards. Hence, existing relations between hazard and losses may have to be updated regularly.
We are, and most likely will, live in a strongly connected world. Geo-political shifts in one location may have reverberations elsewhere in the world. The stark increase of inflation in the recent year is one example showcasing this. Whether and when comparable geo-political events will happen is close to impossible to predict. Such is their impact on volatile asset values.
To be able to include all these non-hazard uncertainties, future generations of CAT models should have a high degree of flexibility in calculating asset values across a portfolio. For example, inflation should be adjustable by the insurers themselves by calibrating exposure according to the local inflation levels. Flexible CAT models could not only be used to appropriately react to financial changes, but also be employed to explore different potential “what if” scenarios. Having multiple scenarios prepared will help insurers to react to unforeseen changes quicker.
Thus far, we merely discussed flood risk ex post. But what if we already could anticipate when and where floods occur? Such flood early warning systems could reduce the potentially devastating impact of a flood event by initiating anticipatory action. Small mitigation measures such as asking property owners to move valuable items up to the second floor prior to an event can already reduce the losses experienced by the insurer. Other actions such as local flood defenses can be the difference between a loss or no loss at all, especially when mitigating against events in the high frequency – low severity space.
But early warning is also beneficial at the larger scale. For instance the track and rainfall intensity of tropical cyclones, which typically affect large areas and thousands of people, can be predicted a couple of days ahead despite inherent uncertainties. The UK Foreign, Commonwealth and Development Office (FCDO), together with partners such as Fathom, the University of Reading, and ECWMF, issues event forecasts aimed at improving decision making and fund allocation for a faster recovery.
However, the recent 2021 flood in Europe is a sad example that early warning is not only a technical undertaking, but also has a clear human dimension to it: only if alerts are converted into right action can losses be reduced. This points to the issue that these early warnings and anticipatory actions are impossible to represent in a CAT model, as it is dependent on if action is actually taken or not. To what extent warnings are actually turned into action is nigh on impossible to include.
Overall, the insurance world faces a plethora of challenges when it comes to preparing for future flood risk. Building on a long history of model development and data collection, it is key to account for changes and uncertainties along the entire workflow: climate conditioning flood hazard, adapting vulnerability curves, as well as accounting for short-term and long-term changes to exposure. For next-generation future-proof CAT models, these aspects need to be addressed, providing the basis for continuing or expanding profitable business of insurers into the future.