Enter Note Done
Go to previous page in this tab
Session
  • Presentation | A51B: AI-Driven Innovations in Earth and Atmospheric Sciences II Oral
  • Oral
  • Share Icon
    Map Icon
    Bookmark Icon
  • A51B-02: BenchFlux: Scalable AI Benchmarks for Terrestrial Carbon Fluxes to Advance Research, Education, and Resource Management
    • Schedule
    Notes
  • Location Icon344-345
    NOLA CC
    Set Timezone
    •  
    View Map

Generic 'disconnected' Message
Author(s):
Stefan Metzger, AtmoFacts (First Author, Presenting Author)
Maoya Bassiouni, University of California, Berkeley
Samuel Bower, AtmoFacts
Ankur Desai, University of Wisconsin Madison
Matthew Fortier, Mila - Quebec Artificial Intelligence Institute
Sophie Hoffman, University of Wisconsin Madison
Yanghui Kang, Virginia Tech
Trevor Keenan, University of California, Berkeley
Danielle Losos, Yale University
Alvaro Moreno, Image Processing Laboratory, Universitat de València
Christopher Pal, Ecole Polytechnique de Montreal
Sadegh Ranjbar, University of Wisconsin Madison
Oliver Sonnentag, University of Montreal
Paul Stoy, University of Wisconsin-Madison, Madison
Emma Izquierdo, BOKU University
Jingfeng Xiao, University of New Hampshire, Institute for the Study of Earth, Oceans, and Space, Earth Systems Research Center


The BenchFlux project uses artificial intelligence (AI) to improve how we track the exchange of carbon dioxide (CO₂) between land and atmosphere. These exchanges are critical to understanding and managing our planet’s atmosphere and natural resources. BenchFlux combines data from specialized towers that measure atmospheric CO₂ exchange rates directly at the ground level with satellite and other Earth observation data. This allows us to build better models that work across local to global scales.



Unlike traditional methods that often overlook differences in scale, BenchFlux applies a new “scale-aware” approach. This helps capture complex patterns in how CO₂ moves through ecosystems and provides more accurate, reliable estimates. The project develops advanced AI models that include uncertainty estimates, making the results more trustworthy for decision-making.



BenchFlux also supports open science and education. The team is creating free tools, data sets, and online tutorials so that scientists, students, and resource managers around the world can use and learn from them. By improving how we measure and model carbon fluxes, BenchFlux provides decision support for improved stewardship of Earth’s ecosystems and atmosphere.




Scientific Discipline
Suggested Itineraries
Neighborhood
Type
Main Session
Discussion