Abstract:
The land absorbs nearly 30% of the CO2 that humans emit. With growing interest in nature-based climate solutions, we urgently need to know how, where and why that sink is changing. Eddy-covariance flux towers give the most direct measurements of how ecosystems exchange carbon, water and energy with the atmosphere. But the data are complex, and that has long limited their use: the last global FLUXNET release was in 2015. In this talk, I will describe the new FLUXNET Data System. It is a community-wide effort across regional networks worldwide. It delivers flux data that are processed and quality-controlled through a common pipeline, and openly licensed (CC-BY-4.0), along with new tools to find and access them. I will then show how longer, more consistent records let us ask questions we could not answer before. These range from detecting step changes, tipping points and gradual shifts in how ecosystems function to tracking long-term trends in productivity. Finally, I will discuss how AI is speeding up this work. Examples include foundation models that combine flux data with satellite imagery to improve carbon accounting for nature-based solutions, and AI-ready data infrastructure with agentic tools for exploring it. Together, these advances give stronger constraints for Earth system models and remote-sensing products, and a basis for near-term forecasting of the land carbon sink.
Bio:
Trevor Keenan is a Professor in the Department of Environmental Science, Policy and Management at UC Berkeley, where he holds the Chen Chancellor's Chair in Climate Science and Solutions, and a Faculty Scientist at Lawrence Berkeley National Laboratory. He leads the Quantitative Ecosystem Dynamics Lab, which combines eddy-covariance flux networks, satellite and lidar remote sensing, machine learning and Earth system models to understand how terrestrial ecosystems respond to, and shape, a changing climate. He chairs the FLUXNET Community Council, directs the FLUXNET Coordination Project, and co-developed the NASA/IBM Prithvi-EO-2.0 Earth observation foundation model. He is a recipient of AGU's Piers J. Sellers Global Environmental Change Mid-Career Award.
Summary:
Focus: monitoring and modeling ecosystem cycles using diverse local and remote sensing measurements + ML to analyze them
The ecosystem drives and cycles carbon through the atmosphere
Plant growth captures CO2 and respires O2
⅓ of our emissions are captured by the land sink, controlled heavily by terrestrial ecosystems
Land sink
Fraction of emissions captured has stayed consistently at ⅓ even as emissions have risen
The total emissions captured have been increasing on pace with increasing emissions
We don’t currently know why
Many interventions can improve land’s ability to absorb: land use management, re-/afforestation
We need to establish best practices for this
Measurement
We have diverse measurement tools across scales, most measurements are proxies for the data we really want (e.g. trees are proxies for ecosystems)
Plot/Tree: surveys, cores
Ecosystem: flux towers, drones
Landscape: Tall towers
Region: Tower networks
Global: Satellites
Flux towers: key measurement capability
Stand over the landscape
Sonic anemometer: measures wind direction and turbulence
Gas concentration meter
Can infer where changes in concentration are coming from (tracing the wind backwards to its origin)
10-20Hz measurement
30 min resolution on inferred fluxes
FLUXNET: global sensor network
Run by individual organizations, using diverse equipment and procedures
Range: arctic tundra to tropical forest
Massive dataset
Challenging to unify the data
AmeriFlux Data Processing Pipeline
Aggregates unstandardized data into a uniform form
OneFLUX processing pipeline: https://github.com/fluxnet/ONEflux
Fluxnet data explorer: https://www.keenangroup.info/fluxnet-data-explorer
Community research publications: https://www.keenangroup.info/fluxnet-paper.html
Next Steps:
Agentic tools and MCP servers for data analysis and exploration
Analysis
Long-term measurement data -> long-term dynamical insight
Measured the CO2 fertilization effect
Higher CO2 concentrations increase plant growth
Inferring plant biomass based on absorption of CO2 at the ground measured from the flux tower
AI enables global inference
Integration of flux towers with diverse satellite-based measurements
MODIS/VIIRS, LANDSAT, Sentinel, GEDI, SMAP, etc.
ECO2BENCH: standardized benchmarks for ecosystem model development
Enabled by geo foundation models (e.g. AlphaEarth)
Foundation model embeddings enable Ecosystem Productivity GPP modelling