Monitoring carbon and greenhouse gas exchange between atmosphere, land, ocean and human systems underpins the Global Stocktake, national inventories, carbon markets and carbon dioxide removal (CDR). Yet bottom-up (inventories, process-based and data-driven models), top-down (atmospheric inversions) and bookkeeping (land-use change) estimates remain uncertain and often disagree, while the credibility of measurement, reporting and verification (MRV) rests on independent methods agreeing. Machine learning (ML) offers new ways to enhance, complement, emulate, integrate or even replace established observational, modelling and analytical workflows.

This session invites contributions using ML to advance carbon cycle monitoring and understanding across all compartments (atmosphere, vegetation, soils, permafrost, wetlands, inland waters, coastal and open ocean, anthropogenic sources) and all relevant trace gases, including CO2, CH4, N2O, CO, COS and isotopes. We welcome the full breadth of ML: foundation models, transformers, GNNs, CNNs, generative models (diffusion, flow matching), hybrid and differentiable modelling, Bayesian ML, symbolic regression, interpretable, robust and trustworthy ML, and tabular methods such as gradient-boosted trees. We particularly encourage work bridging science, policy and industry.
Topics include, but are not limited to:

- ML-enhanced bottom-up estimates: upscaling, mapping stocks and fluxes, inventories
- ML for top-down inversions: transport emulators, data assimilation, retrievals, point-source detection
- ML for bookkeeping and land-use change emissions
- Emulators and foundation models for process-based and Earth system models
- Hybrid modelling, and generative models for gap-filling, downscaling and ensemble generation
- Fusion of remote sensing with in-situ data (eddy covariance, flask networks, ocean pCO2)
- ML for carbon cycle process understanding and discovery: drivers, feedbacks and emergent constraints, with explainability, causality and uncertainty quantification
- Reconciling lines of evidence across methods, compartments and gases
- Monitoring for policy and industry: MRV, national inventories, carbon markets, CDR verification
- ML-based decision support and policy pathways (reinforcement learning, integrated assessment models)
- Benchmarks, datasets & open tools