The research driving that shift is worth understanding because the tools emerging from it are already beginning to change how serious organizations think about physical asset exposure and supply chain vulnerability.
Traditional climate forecasting evaluated variables in isolation — temperature here, precipitation there, extreme events as discrete occurrences. That approach works reasonably well for single-location analysis. It breaks down when the question is how risks compound and travel.
At Boston University, researcher Elizabeth A. Barnes and her team are taking a different approach. Working at the intersection of data science and atmospheric science, Barnes's group is developing AI models capable of forecasting hurricane activity across seasons and years, and hybrid systems that combine machine learning with traditional analog forecasting to improve multi-year climate projections. The specific focus is on compound events — heat combined with drought, excessive rainfall coinciding with supply chain stress — which are consistently more damaging than single variables and significantly harder to predict using conventional methods.
The research also builds in uncertainty quantification, so models can communicate not just what may happen but how confident those predictions are. For operational planning, that distinction matters more than most forecasting tools currently acknowledge.
One area where this research has direct industry relevance is agricultural and food supply exposure.
Recent modeling on global agricultural systems found that in extreme scenarios, more than 20% of caloric supply in nearly one-third of countries could face compound climate hazards in a single year. More importantly, that exposure doesn't stay contained. It propagates through global trade networks, meaning a disruption in one production region affects availability across markets that had no direct weather event of their own.
For companies with geographically concentrated supply chains or single-source dependencies, this isn't an abstract finding. It's a description of how a bad crop year in one region becomes an operational problem in another — and why forecasting only the weather at your facility misses most of the actual risk.
Companies that have been running annual risk assessments against fixed climate assumptions are working with tools that the underlying science has already moved past.
AI-enabled models are beginning to offer actionable forecasting windows rather than long-range projections, system-level risk mapping rather than isolated site analysis, and scenario planning that updates as conditions evolve. That capability is still maturing, but it's moving from academic research into applied tools faster than most risk management functions are tracking.
For organizations managing infrastructure, physical assets, or geographically distributed supply chains, the question is no longer whether climate forecasting is precise enough to act on. In a growing number of applications, it is. The gap is on the implementation side — specifically, whether risk models are built to absorb that information or still treating climate as a background condition rather than a variable with a forecast.
Learn more about the research here.
