Cutting-edge flood modelling shows why historical data alone can’t prepare financial systems for climate-driven catastrophe.
display-middle-ads-here
At CeFPro’s Climate Stress Testing Europe conference, one presentation stood out for
its rigorous approach to a key challenge facing financial institutions: how to
model climate-related hazards in the absence of rich historical data.
Delivered by
a leading figure within an organization renowned for its global flood modelling
expertise, the session explored how scientific innovation, machine learning,
and robust terrain modelling can inform financial risk assessments in the face
of escalating climate threats.
The speaker
began by contextualising the growing severity of climate-related disasters,
pointing to 2023 as a record-setting year for catastrophic weather events.
From the
devastating floods in Slovenia and Kenya to wildfires in Hawaii and Hurricane
Otis in Mexico, it is clear that extreme weather is both intensifying and
affecting regions once considered relatively immune.
These trends
underline the limitations of relying solely on historical loss data.
Traditional
records do not capture the full variability of the climate system, nor can they
reflect emerging risks driven by a rapidly warming planet.
With that
backdrop, the speaker spoke of her organization’s mission to provide globally
consistent, high-resolution flood risk data that integrates the latest
evidenced and opinion in climate science.
From small
beginnings in an academic environment, her company has grown into a globally
recognised provider of flood intelligence.
Now a
subsidiary of a major global bank, the company blends academic rigour with
commercial relevance, maintaining a strong commitment to open science and
peer-reviewed validation.
A significant
portion of the talk focused on the technical underpinnings of flood risk
modelling.
The
foundation, it was explained, lies in accurate terrain data.
While
countries like the US and UK benefit from high-quality LiDAR datasets, the rest
of the world must rely on satellite-derived elevation models, which often
misrepresent ground-level features due to interference from tree canopies and
buildings.
To address
this, an innovative machine learning-enhanced global terrain model strips away
surface obstructions to better represent the Earth’s bare surface.
This model
underpins global flood maps and catastrophe models, offering a more realistic
basis for simulating water flow during flood events.
Yet physical
terrain is only part of the equation. The presentation also delved into the
complexities introduced by internal climate variability, particularly phenomena
such as the El Niño Southern Oscillation.
These natural
cycles influence storm formation, rainfall intensity, and drought frequency,
introducing significant year-to-year variability that is not captured in the
typical 40 to 60-year historical record.
Furthermore,
as the speaker noted, anthropogenic climate change adds another layer of
non-stationarity, making it critical to move beyond past observations and
embrace forward-looking climate models.
To that end,
the talk explored how output from global and regional climate models can be
integrated into hazard frameworks.
By combining
climate projections with hydrological and oceanic models, the company can
simulate future changes in pluvial, fluvial, and coastal flooding.
Importantly, these
projections are developed using ensemble modelling techniques, which allow for
a quantification of uncertainty – a crucial feature given the inherent
variability of climate systems and the diverse methodological assumptions
embedded in different model generations.
The audience
was shown how projected flood risk varies by emissions scenario and hazard
type.
Coastal
flooding, driven by sea-level rise and storm surge, shows particularly
consistent increases under all scenarios.
Inland flood
risk, especially from pluvial sources, is expected to rise due to the increased
capacity of warmer air to hold moisture, while fluvial flood risk presents more
regional variability and lower confidence in projections.