If you can’t make it, or even if you can, and have questions or comments, please reach out. We’d love to hear from you.

Q: In your work with predictive analytics, what behavior or outcome do your models predict?

A: At Sust Global, we transform complex climate science to transparent and actionable signals to customers across finance, real estate and industry. Our models provide asset-level predictions of acute and chronic physical risks like wildfires, floods, heatwaves, drought and sea level rise over the short term as well as decadal time horizons. Using geospatial machine learning techniques, we predict risk exposure at high spatial resolutions, techniques widely known as super resolution, that allows for more accurate localized predictions. These techniques have only recently been applied to climate risk.

By incorporating asset level metadata on physical locations, we can also estimate losses from such events. In 2020, climate-related disasters accounted for roughly $100 billion in damages in the US alone. Predictive capabilities form the first step towards climate adaptation and loss mitigation measures.

Q: How does predictive analytics deliver value at your organization – what is one specific way in which it actively drives decisions or operations?

A: Asset-level, physical climate risk predictions constitute the central pillar of climate-aware, sustainable capital allocation. To distill further, when looking at climate change risks on a large collection of assets, be it a real estate investment trust (REIT) or a mortgage portfolio, the spatial non-stationarity of climate risk exposure means that not all assets have the same level of risk. Our predictive capabilities allow our customers to better understand where is the concentration of their climate risk, what is the level of their risk vulnerability and how they could take the necessary steps to allocate capital in a climate-aware, sustainable fashion.

Q: What surprising discovery or insight have you unearthed in your data?

A: When thinking about climate risks, today’s solutions are natively biased by the past. When thinking of wildfires, we think of California, when talking about floods, we think of Japan or Germany or Bangladesh and when the concern turns to hurricanes , our thoughts race to Florida. Our products enable the exploration of such hazards in the coming decade and the projected footprint of such hazards over the years to come is significantly different from the footprints in the recent past, say the past 20 years. For example, we are seeing increasing fires in Siberia and Alaska which seems less likely if we looked at historic data from 1980-2010. When assessing such risks, we need to understand that climate is a planetary phenomena with global impacts, some with the potential to be catastrophic, some less so (for example, fewer cold waves). This will have a significant impact on human settlement and industrial development in the coming decades as the planet warms and the occurrence of such hazards increases..

Q: Sneak preview: Please tell us a take-away that you will provide during your talk at Predictive Analytics World.

A: Based on our actions now and in the near future, the world could look very different over the coming decades, both in terms of the impact of humans on the climate (transition risk) and the impact of the climate on humans (physical risk). In my talk, I will describe analysis on predictive climate risk data across different scenarios. We’ll also look at how such analysis can be used by businesses to better understand operational risks and plan for the future.

Q: Additional Questions: What adjacencies do you see to machine learning for making predictions useful in your domain?

A: If the goal is to change human behavior through spatial inference and predictive analytics, we need to build synergies across different product and technical functions. That would include interaction design and interpretable metrics. Prediction alone is often insufficient to influence behavior. To better understand and interpret predictive analytics, a user needs to understand what the associated metrics represent. For example, error tolerance on the predictions and performance benchmarks. For the data to be actionable, users often need to be able to interact with the data. The functions of interaction design and metrics presentation are ever critical for predictive analytics on spatial data to be successful.

Posted from the original by: Eugene Kirpichov, Sasha Luccioni & David Rolnick, Conference Chairs, Predictive Analytics World for Climate

Want to learn more about Sust Global? Fill out the form below and we'll be in touch shortly!