Blog summary: By pre-computing millions of years of tropical cyclone activity, along with the climatic conditions driving them, we can query our database to match current observable, and future projected trends in global climate signals. This opens the door to a new chapter in catastrophe risk modelling, as Cat models can now be used as predictive tools. This new generation of cat models can provide dynamic views of risk, that get refined continuously as the climate changes.
In a previous series of blog articles, we discussed the fundamental aspects of the next generation of Cat models we are building at Reask, and why we believe they will change the way the industry is assessing risk from natural disasters. In brief:
- Cat models need to be explicitly connected to climate conditions, and driven by physics – not just by statistics from historical events.
- Only a fully global modelling system can facilitate the assessment of interconnections between regions and perils.
- There is nothing special about history, and it should not be the focus of model calibrations.
Our Unified Tropical Cyclone model (UTC) embraces these principles, offering a vast range of advantages over traditional Cat models when providing a long-term reference view of risk. However, there is another key benefit from this new generation of Cat models: the possibility to provide climate-adjusted probabilistic views, ranging from seasonal forecasts to future climate risk projections.
At its core our UTC event set contains over 15 thousand years of different climate conditions, and these were used to drive over 10 million years of tropical cyclone activity globally. With our TC events explicitly connected to their climate drivers (e.g. sea surface temperature, surface pressure, wind shear, steering flow patterns), the UTC database represents a true climate driven view of TC risk. As a result, it can be sub-sampled to extract climate adjusted views of risk: ones that match (1) current observable trends in global climate signals (e.g. a strong la Niña event) or (2) projected trends for the coming decades.
How does Reask’s sub-sampling methodology work?
With every single stochastic year of the UTC event database being associated to known climate conditions, we can use our AI engine to extract years that closely match a target set of climate conditions, such as the one we are currently observing, or expect to observe in the near future (Figure 1).
Instead of relying on the same reference long-term view every season (portraying the risk under an average historical climate), this dynamic view of risk can be refined to account for near-term evolution in climate signals (as shown in Figure 1 – bottom panel). The long-term reference view of risk can then be adjusted on-the-fly (in this example monthly) to enable decision making through the course of a tropical cyclone season, as global climate signals evolve.
Beyond seasonal forecasts, the applications of our technology are broad; the impact of climate patterns (such as the phase of ENSO – El Niño Southern Oscillation – for tropical cyclones) can be investigated, change of risk induced by decadal variabilities might be inferred, and importantly, the impact of climate change and how regional risk might evolve for the next century can be examined. Finally, the method is fully extendable to any perils building a consistent solution to link risks globally.