Karen Clark & Company (KCC), a catastrophe risk modelling firm serving re/insurers and insurance-linked securities investors, says artificial intelligence could significantly change how catastrophe models are developed and updated, with implications for reinsurance pricing, underwriting and capital management.
In a white paper published in August 2026, KCC says catastrophe modelling is moving from traditional statistical approaches towards physical models enhanced by AI.
The company believes this shift is particularly important for more frequent weather perils, including severe convective storms, winter storms and wildfires, where historical data may not fully capture increasingly complex risk.
KCC says first-generation catastrophe models were developed primarily for low-frequency, high-severity events such as hurricanes and earthquakes. These models used statistical distributions and historical data to generate large numbers of hypothetical events and estimate potential losses.
However, the company says the growing importance of secondary perils, climate change and the pace of scientific development are placing greater demands on models used by re/insurers.
KCC has developed high-resolution physical models for perils where traditional statistical approaches are less effective. These models use atmospheric equations and large volumes of environmental data to simulate weather systems as they develop, providing information that can be used for catastrophe pricing, reinsurance decisions, claims management and loss reserving.
KCC says its Severe Convective Storm (SCS) model processes more than 30 gigabytes of satellite, radar and weather data each day, producing hail and tornado or wind intensity footprints. Its LiveEvents™ process, which has operated since 2018, has also enabled the company to compare model estimates with actual claims and losses.
According to KCC, this process has generated an archive of more than 100 terabytes of high-resolution atmospheric data alongside tens of billions of dollars in claims data. The company sees these datasets as a foundation for its next generation of AI-informed physical models.
For re/insurers, KCC says one of the main benefits of AI could be the speed at which catastrophe models are updated. The company says AI-informed models could allow updates to be made in days rather than months, helping insurers and reinsurers base pricing and underwriting decisions on more current information.
KCC is focusing particularly on severe convective storms, including derechos, which can generate significant losses but are difficult to predict accurately. The company says its researchers are using tens of thousands of meteorologist-labelled radar images to train machine-learning systems to recognise different types of convective activity.
Wildfire risk is another area where KCC sees potential. The company says machine learning can analyse large volumes of atmospheric data to identify the wind patterns that can drive major wildfire events, including downslope winds such as Santa Ana and Diablo winds.
Climate change adds further complexity. KCC says some climate-related changes affecting hurricanes and wildfires can be quantified and incorporated into models, while greater uncertainty remains around severe convective storms and winter storms. The company says it therefore continually refreshes models for these perils with new atmospheric data.
For the re/insurance market, KCC’s argument is that the value of catastrophe models increasingly depends on how quickly they can reflect new scientific evidence and emerging risk patterns. Rather than replacing the established catastrophe modelling framework, the company says AI can enhance its underlying components and allow models to be updated more frequently.
KCC concludes that AI-informed physical models could give re/insurers more current catastrophe risk information for pricing, underwriting, claims and capital decisions. As secondary perils become an increasingly important consideration for the market, the company says keeping models current will be as important as their underlying accuracy.
