How data modelling could predict future climate catastrophes

10 Feb 2022

Image: © William Messing/

Aon’s head of catastrophe insight explains the role data science plays in quantifying natural disaster risk and the future of climate modelling.

Click here to view the full Data Science Week series.

As part of Data Science Week on, we’ve been examining the ways in which data science can make a difference in the world.

We heard from AI thought leader Aruna Pattam about the many real-world applications of data science, including how it can be used to predict the effects of different climate crisis mitigation or pandemic management strategies and highlight those most promising.

We’re seeing this kind of data science being employed by start-ups and major companies alike. UK start-up Cervest, for example, has a climate intelligence platform that uses AI to analyse climate data. It is targeted at businesses and governments that need to make sense of climate and extreme weather data to assess risks and effects on physical assets.

On a larger scale, risk management and insurance multinational Aon recently released a 2021 report on weather catastrophe and climate insights using sophisticated data models. It highlighted trends and the real-time impact of the climate crisis.

Aon’s head of catastrophe insight, Steve Bowen, has a unique position within the insurance industry when it comes to climate.

‘Addressing the uncertainty in climate data should be a high point of focus for data scientists’

He has a bachelor’s degree in meteorology from Florida State University and a master’s degree in business analytics from the University of Notre Dame.

In his current role, he aims to help identify and communicate where natural disaster risk is accelerating and incorporate the latest scientific research to help answer questions regarding short- and long-term potential risk.

He told that data science has an important part to play when it comes to addressing the direct and indirect impacts from the climate crisis.

Climate and catastrophe modelling

“As more robust data sets become available and computational power increases, the hope is that we’ll be able to not only reduce the uncertainties and sensitivities that exist within global climate model output, but also obtain realistic information on future scenarios at a more precise geospatial level,” he said.

“Climate change impacts are not globally uniform or linear and exhibit unique differences on a peril and regional basis. Better addressing the uncertainty in climate data should be a high point of focus for data scientists.”

Aon’s weather, catastrophe and climate report highlighted “global natural hazards” from 2021 and looked to quantify and qualify how issues such as the climate crisis are driving new and emerging types of risk.

Data science plays a major role in building reports such as this using catastrophe models. These comprise elements such as physical hazard scenarios, building or vehicle exposure, and a financial loss engine to quantify risk.

According to the report, insured losses from natural disasters reached $130bn in 2021, well above the 21st-century average of $74bn.

Bowen explained that catastrophe models are different from climate models in that the former typically include backward-looking analysis to best fit observed weather or other natural hazard loss scenarios.

Climate models, on the other hand, take various atmospheric, land surface, ocean and sea ice components to compute complex simulations that aim to discover how changes in any of these areas can lead to larger-scale climate disruptions. These simulations are meant to project future climate environments on a decadal or centurial timescale.

“What is exciting is that a new generation of catastrophe model output is now starting to directly implement event scenarios that have been conditioned by climate models,” Bowen said.

“Adjusting for future weather events and future exposure can help identify where physical risk and associated losses will change.”

The challenges of data modelling

Even though the insights that can be gleaned from this modelling are impressive, Bowen noted that model outputs are only as good as the data that’s fed into it.

“While data availability and quality has exponentially grown in recent decades, we’re always in search of more to better calibrate and validate model results,” he said.

“Prediction modelling has made significant forward progress. But there remains a lot of uncertainty in much of the modelling around climate, which is common with virtually every other subject matter as well.”

Outside of data quality, Bowen said communication is another challenge, especially when it comes to the general public.

‘Disaster events in other parts of the world can lead to cascading effects’

“Most people are not aware of the individual natural hazard risks that exist at their home and they are even less aware of how risks may evolve during the life of their mortgage. Any advances that can be made by data science to help more clearly define where people may be increasingly affected by future events would be a major step forward,” he said.

“We also need to recognise that prediction models in the weather, climate or natural hazard space must not only highlight physical risk, but the non-physical risk that is having increased impacts on our daily lives.

“Disaster events in other parts of the world can lead to cascading effects to supply chains, humanitarian aid distribution or other asset-related disruptions. These secondary and tertiary effects will become even more compounded as our interconnected world is even more dependent on non-local resources.”

In spite of these challenges, Bowen said he remains optimistic “that we’ll begin to narrow the uncertainties and further grow our confidence in what future climate conditions we may face”.

Don’t miss out on the knowledge you need to succeed. Sign up for the Daily Brief, Silicon Republic’s digest of need-to-know sci-tech news.

Jenny Darmody is the editor of Silicon Republic