New Strategy for Building Resilience: Integrating Climate Risk into Machine Learning Models

By Jeffrey Bohn, Chief Strategy Officer, One Concern

Climate change has triggered a new era of severe weather events that requires an evolved approach to address and differentiate how climate change affects property, infrastructure, businesses, and communities from a resilience modeling perspective. As climate change continues to reshape the 21st century’s threat landscape, decision-makers must have access to predictive analytics, more granular models, and a more accurate picture of the distribution of probable outcomes related to risk. Better data and models can guide more effective natural disaster mitigation efforts. Failure to enable visibility into the bigger picture of risks, resilience, and consequences, from the individual asset level to the market level, will materially hinder our efforts to address climate change.

To use a recent example, a conventional approach to risk modeling sees a group of buildings on a coastline and assumes that they are vulnerable to coastal weather events, whereas structures located further inland are less at risk from storm surges. This is the assumption one would arrive at based on using only historical event data.

However, climate change has changed the nature of risk and resilience evaluation, mainly as weather events outside historical norms create a rising number of unanticipated outcomes and domino effects. That’s why when we do risk and resilience modeling at One Concern, we look beyond a particular property. We also look at the full networks of lifelines they depend on, including infrastructures and communities.

Our Approach

We transform risk analysis with a digital twin of the underlying networks of lifelines, such as power and transportation. The digital twin facilitates visualization of the impacts of lifeline failure. Moreover, we forecast future events through forward-looking, predictive models rather than relying only on historical data to map a changing world.

Using this dynamic approach, we identified an overlooked risk factor impacting global communities: climate change makes previously resilient inland areas sometimes more susceptible to floods due to their proximity to rivers. Modeling this network effect at the building level with new climate data in mind uncovers the true risk obscured by outdated methods.

Caption: Present conventional approach to risk modeling shows Florida’s panhandle is at minimal risk of flooding.

Caption: In 2050, when considering the RCP4.5 trajectory adopted by IPCC, inland communities in Florida’s panhandle will experience greater flooding at 105% depth ratio.

Overlooked risk factors like the example above represent the new reality of severe weather, which requires an expansive approach to address and differentiate climate change impact on risk and resilience models. We must model these emerging risks and outcomes based on new parameters that reflect a future characterized by climate change.

The Science of Probabilistic Modeling to Parameterize Risk and Resilience

Climate change has thrust conventional risk model frameworks into a world of unknown unknowns — where you don’t even know what data or risk categories you might be missing.

With One Concern’s advanced resilience intelligence platform, users have a framework where climate change risk becomes a known unknown for resilience modeling, enabling risk mitigation at scale.

We can characterize severe weather phenomena with known parameters. We can then use machine learning (ML) to “fill in the data gaps” to account for a broader range of scenarios that can be assessed at an individual property level — even when datasets are incomplete.

Risk and resilience models rely on machine intelligence ranging from conventional statistical approaches to more sophisticated approaches reflected in ML and artificial intelligence (AI). Conventional approaches are easier to scale and typically less data-intensive, while more sophisticated ML & AI can produce more granular analyses at the costs of needing much more data and requiring more sophisticated model support teams. Dealing with the complexities inherent in integrating climate change risk into granular property risk and resilience models for robust decision support requires a deep understanding of the data, models, and use cases for mitigation.

One Concern has the expertise, experience and data access to decide which machine-intelligence approach makes the most sense for robustly supporting practical decision-making at scale. Armed with better data and machine learning, we can simulate a large number of scenarios before a natural disaster to identify property weaknesses, dependent network lifelines, vulnerable communities, and critical infrastructure in harm’s way.

Ultimately, we’re trying to facilitate better resilience management through detailed mapping of the distribution of probable outcomes.

Pricing Risk to Bridge the Resilience Divide

By accurately modeling risk and resilience, One Concern can enable enterprises, communities, and governments to price, mitigate, and transfer risk through insurance, insurance-linked securities, and innovative resilience securities. Historically, impoverished and otherwise disenfranchised communities have suffered disproportionately from disasters due to socioeconomic factors and a lack of investment in disaster mitigation — a key element is underinsured assets.

For example, in the past decade, large earthquakes hit eastern Italy and Christchurch, New Zealand. In New Zealand, most assets were insured; in Italy, only a small fraction. Today Christchurch has fully recovered, while communities in Italy continue to deal with the destructive aftermath.

Even when dealing with the factors that lead to underinsurance, insurance companies, investors, property owners, and governments, then and now, struggle with quantifying and pricing risk and resilience. And if you can’t price something, you can’t ensure or trade it. Without understanding the dependencies a business faces, it is almost impossible to underwrite risk at scale properly. This makes it challenging to improve resilience systematically. In the case of Italy’s seismic impacts, underinsurance led to long-term economic repercussions, while New Zealand was able to recover thanks to proper insurance.

Advanced risk modeling improves both valuation and management. Using One Concern’s available data and machine learning systems, we can parameterize and define the distribution of probable outcomes, which enables more resilient communities through proper risk transfer and mitigation.

The insurance industry and capital markets are vital to building a workable approach for the broader private sector. If we can price risk and resilience, they can be managed. We can plan adequately and address issues before disaster strikes.

Jeffrey Bohn is Chief Strategy Officer for One Concern and a Board Member of the Consortium for Data Analytics in Risk (CDAR) at U.C. Berkeley

--

--

--

We’re advancing science and technology to build global resilience, and make disasters less disastrous

Love podcasts or audiobooks? Learn on the go with our new app.

Recommended from Medium

P2P Lending for Home Flippers and Minorities

ProjectReport

A Junior Data Scientist’s Guide to A Better Life

Big Data Architecture Concepts

INTRO POST

Meeting Mollie’s Data Demands, Part 1: Specialisation

Download In ^*PDF Good Charts: The HBR Guide to Making Smarter, More Persuasive Data Visualizations…

Aesthetic Map: Bivariate Choropleth for Association Context on Spatial Data

Get the Medium app

A button that says 'Download on the App Store', and if clicked it will lead you to the iOS App store
A button that says 'Get it on, Google Play', and if clicked it will lead you to the Google Play store
One Concern

One Concern

We’re advancing science and technology to build global resilience, and make disasters less disastrous

More from Medium

Visualizing Spatial and Temporal Trends in Earthquakes

Radiologists’ nightmare — segmentation masks

Predicting March Madness Upsets with Regression Models in R

Series: AI For Drug Discovery Part 2