RMS

Most catastrophe (cat) models that describe weather related phenomena, such as hurricanes, floods, or storm surges rely on the same fundamental physical equations that are also being used by numerical weather and climate prediction models or by hydrological and hydraulic models. So what does that mean for cat modeling?

On the most basic level, our clients need a model to contain a credible set of weather extremes that might occur over a given period of time. They need to know that the model captures when the events might occur, where they could be, and how they would take shape. The process of building that event set is a combination of science, statistics, and a lot of instinct and intuition from people like us.

Let’s take one piece of the puzzle, and assume we’re looking at what events could occur over a period of one year, since this is the duration of many insurance contracts. We want to confidently capture the events that could realistically happen, so we’re going to have to run an ensemble of tens or hundreds of thousands of derivations of events that allow us to see the breadth of possibilities that might occur in the next year. A few years ago, our only choice was to run simplified atmospheric and hydraulic models for that many ensemble events. But today we can leverage the immense growth in computational power to really change the way that we think about building models smarter and faster.

A cat model must also be able to reconstruct past events. This drives the need to build and use more realistic and highly complex numerical models that capture every nuance of an event. Parts of our models have therefore become so complex and small-scale that they require smart coding on central processing units (CPUs) and graphics processing units (GPUs) that are commonly used for computer graphics that need to process large blocks of data in parallel. It’s an enormous computational task that rivals numerical weather prediction and climate prediction modeling at the National Oceanic Atmospheric Associate (NOAA), the National Center for Atmospheric Research (NCAR), and many other international weather/climate agencies.

But the process doesn’t end with finding the computing resources needed to support the data and analytics. There is a scientific challenge in understanding the physical driving mechanisms for extreme events, applying advanced statistics, and having intuition in judging complex results. Computers and models can only take us so far, and they will always have their biases. There are elements that human instinct and expertise have to correct or apply to make them as realistic as possible. This is what makes catastrophe modeling rewarding for our modeling teams, and also ever changing and evolving. For our clients, it’s a solution to a problem. For us, it’s an exciting mix of science and intuition that makes it an art form.