Insurers remain reticent when it comes to granting cover for potentially catastrophic climate risks.
The possible impact of natural events, worsened by the changing climate and the rise and spread of
insured values, is enough to give even the most experienced underwriting director pause when
considering storm-related risks.
Sensors and historical data from the past 30 or 40 years have proven to be helpful, but without
longer time series, they have been insufficient in accurately modelling rare, extreme events. They are
not enough when attempting to assess the impact of changing environmental factors on loss
frequency and severity. New approaches and new tools are resolving the challenge. Artificial
Intelligence is unquestionably increasingly valuable in climate-risk modelling.
In the arena of parametric insurance, the emerging science of AI-powered satellite imagery analysis is
already an invaluable tool to help address the challenge of understanding and modelling some
specific natural perils like wildfires. Since better understanding leads to more effective coverage
structures, its use benefits everyone in the value chain. It is already helping to ease risk carriers’
reticence to cover many escalating climate risks.
Using AI for wildfire risk
The game has been changed by combining current and historical data with new, global third-party
data sources and AI-enabled assessments of advanced multispectral imagery. It allows the risk
modelling industry to enrich data with physical factors such as sea temperatures, wind strength, and
other global external data. Insurers are already using AI to improve climate risk assessment and
modelling, and to assess claims of in-force policies at the loss stage.
AI has proved enormously powerful in the analysis of imagery, whether aerial or satellite. For
example, it helps parametric insurers to better understand wildfire. During and after an event, it is
used to identify burned and threatened areas, allowing insurers to assess the damage caused. But
even before an event occurs, at the product design and pricing stage, AI has been deployed to help
underwriters distinguish between trees, shrubs, and grass, and therefore to deliver more responsive,
better-rated policies.
AI closes the gap
The volatile effects of climate change have left many traditional insurers unwilling to underwrite
wildfire, but AI-enhanced image-based data analytics has enabled parametric insurers to improve
understanding of the risk and to introduce new coverages to fill the gap. Combined with
supercomputing capacities, it can transform the way insurers model the risk.
Descartes, for example, was granted last year an unprecedented three million computational hours
by GENCI, a leading specialist in supercomputing associated with Artificial Intelligence, to run
complex physics-based models that simulate wildfire risk. The work delivered better understanding
both of climate change and of insurance risk modelling.
Wildfires in the US burned almost 2.7 acres in 2023, and 7.6 million in 2022, and more than 10
million in each of 2015, 2017, and 2020. Bushfires in Australia have been even worse. They ripped
through 27 million acres of the southeast in 2023, and burned more than 207 acres of desert and
savannah in Australia’s north. These numbers show the importance to use AI to better protect and
insure communities.
For example, satellite imagery from SERTIT, the French public laboratory, has enabled Descartes to
harness high-resolution Earth observation data – including optical satellite imagery, 3D data, and
thermal infrared data – to conduct comprehensive damage assessments in the aftermath of natural
disasters. Following wildfires, satellite imagery has been used to examine topography in the relevant
coverage areas to determine if policies have been triggered. If a burnt pixel is observed anywhere
within a coverage area, the relevant clients are informed of their entitlement to a payout, and
receive their pre-agreed claim almost immediately.
Using AI for hail risk
Hail is one of the natural risks least understood by insurers and scientists, but that’s changing with
the help of AI. The analytical toolset is having a profound impact on hail-event modelling. Already it
has facilitated a better understanding of the risk, which like wildfire has become more frequent in
some areas, and simultaneously more difficult to insure due to the insurance market’s lack of
understanding of the changing peril.
AI’s ability to model complex and nonlinear relations has begun to bridge the hail understanding gap.
Improved awareness of the relationship between recoded weather patterns, radar data, and ground
observations of hail has been facilitated by AI-driven models. These new insights help estimate the
location, size, and severity of hail events, which in turn allows retrospective reconstruction and
analysis of historical hail events.
The resulting database of historical events provides insights into the geographical distribution of hail
frequency and event intensity with a high spatial-temporal resolution which is invaluable for
modelling. It has led to the development of flexible coverage for hail-prone risks in areas where many
traditional carriers have withdrawn.
More to come
Development of many applications of AI to the modelling of the world’s evolving natural threats is
under way, such as GenCast, part of Google’s growing suite of next-generation AI-driven weather
modelling tools. In combination with improved satellite imagery and larger stream of data, AI models
have enormous potential. They have already improved our ability to mitigate the financial impact of
climate change with effective risk transfer tools. With this deeper insight, the insurance sector will
support people and enterprises as they face ever worsening hail, wildfires, severe convective storms,
flooding, and a host of other perils of the changing climate.
