AI for climate and weather risk forecasting
Terrafuse is building data fusion technology to re-engineer the climate and weather risk prediction infrastructure around next-generation AI models and hardware and real-time observational data. Its technology substantially reduces the time and cost to build and run physics-based models, while natively integrating remote and ground sensing data. This results in several orders of magnitude speedups and enables real-time, hyperlocal, scalable climate hazard estimation. It is developing this technology for use-cases in energy, insurance, and financial sectors.
Adrian Albert is the founder of Terrafuse, where he leads AI technology development. Previously, he was a machine learning research scientist at Lawrence Berkeley National Lab, where he conducted research on designing machine learning algorithms incorporating physical knowledge for scientific applications. He completed postdoctoral research at MIT working on deep learning for remote-sensing imagery, and obtained his Ph.D. in electrical engineering at Stanford with a thesis on machine learning for energy grids. He was one of the first machine learning scientists at startup C3.ai, working on AI methods for energy and industrial systems.
Since its founding in 2018, Terrafuse has received I-Corps and SBIR grants from the National Science Foundation, a Microsoft AI for Earth grant, and has been a finalist in the MIT VMS Demo Day, a startup competition for top MIT startups.
Forecasting environmental variables, both at the multi-year time scales of climate and at the real-time to monthly time scales of weather is extremely slow, computationally and energy-inefficient, and hard to scale to large areas at resolutions adequate to inform targeted decision making including in energy, public sector, and insurance. Current risk models and associated financial instruments require detailed data on infrastructure that is expensive to collect and maintain at scale.
The current climate and weather risk prediction infrastructure is oriented around numerical physics-based models and large supercomputer facilities, typically at national labs and large corporations. Several new technologies migrate these computationally-intensive, high-maintenance and typically not interoperable workflow models to cloud computing environments. In contrast, Terrafuse re-casts risk prediction as AI workflows that are intrinsically much more efficient, as well as easily deployable on commodity AI hardware. This technology will enable climate and weather risk prediction that is several orders of magnitude faster and more compute- and energy-efficient than current approaches, is more scalable to large geographical areas, allows for real-time, data-driven forecasts, and is more granular in spatial and temporal scales than current risk models.
Terrafuse envisions a world where climate-change risks and their impacts to people's livelihoods and property are accurately reflected in financial decision-making, and can be mitigated by accurate, timely forecasts. Orders of magnitude faster, cheaper, globally-scalable, and more accurate forecasts will be used to create real-time applications and insurance products that enable vaster pools of capital to address climate-change risks.
hello [at] terrafuse [dot] ai