ALIVE
Paul Stoy · Sadegh Ranjbar
Low-latency, multi-scale flux estimation using geostationary remote sensing; builds gap-free EO layers and compares multiple ML methods for continental-scale, hourly datasets with uncertainties and sensitivities.
The Benchflux project is a consortium of individual research teams working on their own pipeline. Below, you can explore each team's individual contribution.
Paul Stoy · Sadegh Ranjbar
Low-latency, multi-scale flux estimation using geostationary remote sensing; builds gap-free EO layers and compares multiple ML methods for continental-scale, hourly datasets with uncertainties and sensitivities.
Álvaro Moreno-Martínez · Emma Izquierdo-Verdiguier
High-res, gap-free remote sensing (incl. OCO-2/3 SIF) and GEE-optimized pipelines; focuses on explainable and uncertainty-aware ML to deliver global decameter datasets with quantified uncertainty.
Oliver Sonnentag · Christopher Pal · Matthew Fortier
Curates ML-ready FGT-EO data (incl. uncatalogued EC sites) with terrain/soil/meteorology features; develops deep multimodal models and harmonized baselines with uncertainty.
Yanghui Kang · Trevor Keenan
Integrates optical/thermal/microwave EO, climate/soil and TBM simulations; advances spatiotemporal deep learning and knowledge-guided ML with UQ; delivers baselines, metrics and NbCS evaluation toolboxes.
Ankur R. Desai · Stefan Metzger · Samuel Bower
Next-gen EC processing and flux spatialization for explicit space-time matching and nesting, boosting information gain and enabling ownership-level comparisons with top-down inversions.
Jingfeng Xiao
Long-running satellite + reanalysis upscaling program; applies multiple ML methods to generate low-latency, global carbon flux estimates.
The Benchflux project aims to integrate these working groups' data into a scale-aware data pipeline. This pipeline and its tutorials will be hosted on our Github Organization.
Benchflux team
Machine learning benchmark methodology for estimating terrestrial carbon flux with remote sensing + eddy covariance integration, using the datasets above.