Verisk's Tolwinski-Ward: Modelling nat cat risk in a changing climate
Suz Tolwinski-Ward, Verisk’s director of climate statistics, explores how advances in climate science are improving models’ ability to quantify nat cat risk.
How does our increased understanding of climate change and variability affect natural catastrophe risk modelling?
There are two aspects of climate science that come into play in cat risk modelling: climate variability, or the typical variation across years in atmospheric peril activity, and climate change, which is well-understood to have impacts on the distribution of hazards today. Climate change shifts the underlying baseline of hazard, and thus risk, shifting the envelope of variability.
The shift that has already occurred from climate change is more striking for some perils than others. The canary in the coalmine is U.S. wildfire risk, for which the impact of climate change is obvious over the historical record of the hazard. On the other hand, for wildfire risk in Australia there is no discernible climate signal in the data.
As cat modellers, we have always carefully constructed our models to capture climate variability, a basic part of the value proposition. Now, we’re spending more time thinking deeply about climate change as part of our model development processes.
What’s the most exciting development you’ve seen in our ability to observe climate change impacts?
The first thing that comes to mind are recent innovations in the ability of global climate models to resolve tropical cyclone activity.
Climate models are physics-driven but have historically been too coarse to resolve tropical cyclones in a meaningful way, so they haven’t provided great insight into climate change impacts. The overall scale of a tropical cyclone is large compared to the resolution of these models; important dynamics controlling tropical cyclone evolution are much smaller. The storm eyewall, for example, requires much higher granularity than climate models have had.
Recently, a new breed of super-high-resolution climate models has shed light on emerging climate change signals for tropical cyclone activity. We now have physical evidence linking some changes in behaviour we’ve seen to climate change, such as more frequent rapid intensification in the Atlantic basin.
Are there challenges associated with incorporating observational and modelled data?
Absolutely. Observational records are spatially sparse; you don’t have a long statistical sample. At the same time, the quality of the instruments used to make observations, such as satellite technology, has changed vastly over time.
Observations are our baseline for reality, but they’re uncertain, so we have to capture that uncertainty and mesh it with information from physically based models, which all have biases. It’s a fantastic and rich challenge from a statistical point of view.
That challenge isn’t something (re)insurers have to contend with directly, because of providers like Verisk. Our cat models are like a Rosetta Stone that takes everything we know from science and translates it into usable guidance for insurance decisions. We have nearly 100 PhDs working nonstop to take the best of the science, weave it together and package it nicely so our users can get the output they need that’s relevant to their decision-making.
What are some emerging climate change signals in nat cat perils?
Every pair of peril and region is its own unique, special little snowflake! The difference in climate change response in wildfire in the U.S. versus Australia, which I mentioned earlier, is a good case in point. Same peril, different region, different impact.
Climate change signals are regional, and we need to pay attention to that and understand why. We build our models with global coverage, so any organisation providing insurance for more than one region or peril can use our toolset to offset risks across regions or perils and diversify their portfolio to ensure resiliency for them and the communities they serve. That’s the whole game and it’s baked into what we build.