Responsible machine learning is the key to unlocking Climate Intelligence

Our climate is changing, impacting the way we live and work. And these changes will only intensify with time. This creates wide-reaching physical, financial and social risks for everyone, everywhere, impacting every industry, directly and indirectly. To understand climate risk and factor it into strategy, compliance and day-to-day operations, businesses need access to Climate Intelligence.

CI is asset-level intelligence on climate risk to inform decision-making. Climate change brings with it complex challenges that require us to look at how systems are interconnected. At the same time, to enable decision-making that addresses material risks, CI has to be relevant and localized enough to enable specific asset-level decisions.

Climate Intelligence connects the latest climate science, modeling and datasets to real world outcomes at the asset level using machine learning (ML) techniques to create a physical representation of our climate system. This is how CI is able to provide insight across multiple scales, time-frames, hazards and geographies to provide relevant, decision-useful intelligence at the asset level. ML is the only way to tackle physical climate risk in a way that is fast, scalable and multivariate, enabling asset level granularity while still tying together multiple risks.

ML is a powerful tool, and like any powerful tool, it needs to be used responsibly. It’s critical the climate tech industry applies ethical, responsible and trustworthy approaches to machine learning when developing climate services. Businesses applying CI to decisions need to know that algorithms have arrived at decisions through processes humans can understand and trust. Below are the four key principles we at Cervest believe are essential for creating and working with machine learning algorithms.

1. Machine learning decisions need to be explainable

ML algorithms perform complex calculations that can consist of thousands, or even millions, of individual data points. Complex ML algorithms with many decision points can become too convoluted to track, even by the programmers who wrote the source code. Explainability means keeping algorithms interpretable so you can explain how and why the machine made the decisions it’s made.

2. Informed by climate expertise

One of the temptations of ML is to just throw a bunch of data at it and hope it will figure out how to solve whatever problem you're trying to solve by itself. People are right to be skeptical of black box approaches, where ML is used to figure out what to do with that data. Applying ML to data without understanding the use cases of the Insights you're providing and how people will use those to inform decisions is going to muddle rather than provide clarity.

This is why guidelines such as responsible machine learning principles advocate for designing “human-in-the-loop” review processes. Having in-house access to diverse expertise is key when evaluating climate data models. At Cervest, we incorporate climate science expertise into our model as part of our validation process and make adjustments in our model approach based on human expertise and interpretation, rather than relying solely on backtesting against historical data.

3. Quality-controlled data and processes

Machine learning can only be as good at the data it’s given. Climate data feeding into CI needs to come from reliable sources, such as peer-reviewed climate science (e.g. CMIP6).

Generating CI at scale requires highly effective data management to realize good data quality. Climate datasets tend to be very large, fragmented and come from multiple sources in multiple formats. For businesses to trust CI, they need to know that the right quality controls on data processes are in place, and that data can be traced throughout its journey along the data pipeline.

4. Part of the right data architecture

ML techniques are just one piece of the CI puzzle. Getting the foundational elements of an ML platform right is key to accelerating the accuracy, speed, and quality of decision-making it can be used for. Flexible, scalable and networked data infrastructure for CI is key to the accuracy, speed, and quality of decision-making.

Access to asset-level intelligence on climate risk is fundamental to navigating business performance, compliance and staying competitive in the 21st century. Businesses need to know that the CI they are using to manage their climate risk is science-backed, based on responsible machine learning methodologies and has the right data architecture in place, assimilating and rendering complex climate science in ways that decision-makers can easily understand and act on. EarthScan™ places climate at the core of strategic, operational, and financial decisions by creating shared visibility on how climate change has, is and could affect your assets.

Visit our Climate Intelligence Academy to build the skills and knowledge your organization needs to become climate intelligent.