Research
Key questions about the soil carbon cycle — how carbon forms, how long it persists, and how organic and inorganic carbon respond to environmental change and land management — remain unresolved. That gap keeps science from informing the policy and industry that need verifiable methods for soil carbon sequestration. We work to close it.
Process-based soil and land carbon models, from soil columns to Earth system models.
Global soil profiles, radiocarbon, field trials and remote sensing, brought together through data assimilation.
Process-guided, interpretable machine learning that learns mechanisms instead of just fitting patterns.
Soil organic carbon
Formation, persistence, and responses to climate change.
Soil organic carbon is the largest actively cycling carbon pool on land, yet we still lack a quantitative picture of how it forms, why some of it persists for millennia, and how it will respond to warming and land management. We combine global soil-profile and radiocarbon data with process-based models to resolve these mechanisms at global scale.
Current directions
- Global soil organic carbon age and turnover informed by radiocarbon data
- Soil microbial physiology — e.g., microbial carbon use efficiency and energy limitation — and its environmental controls
- Deep soil carbon sequestration and its limits
Read more
- Nature 2023 — microbial carbon use efficiency promotes global soil carbon storage
- Nature Geoscience 2026 — limited energy for microorganisms constrains soil carbon accrual
- Nature Communications 2024 — multiscale insights on microbial carbon use efficiency
- Global Change Biology 2024 — model convergence after data assimilation
Enhanced rock weathering
Scalable, verifiable carbon dioxide removal.
Spreading crushed silicate rock on farmland accelerates natural weathering and can lock CO₂ away as inorganic carbon for thousands of years. We ask what limits this process in real soils, where the captured carbon ends up, and how organic and inorganic carbon interact — the science needed to measure, report and verify removals.
Current directions
- Key limitations on CO₂ sequestration by enhanced rock weathering (theory and data syntheses)
- Long-term fate of sequestered carbon across soil–river–ocean systems
- Interactions between soil organic and inorganic carbon
- Other soil-based carbon removal pathways, such as woody debris preservation
Read more
- JAMES 2026 (accepted) — COUSINE, a mechanistic model of CO₂ removal by enhanced rock weathering
- Global Change Biology 2024 — inorganic and organic synergies in enhanced weathering
- Science 2024 — size, distribution and vulnerability of global soil inorganic carbon
- Communications Sustainability 2026 — scaling up enhanced rock weathering equitably
Cross-scale carbon cycle models
From soil columns to Earth system models.
We develop land carbon cycle models that couple organic and inorganic soil carbon, link to Earth system models, and stay fast enough for data assimilation and scenario analysis — so that management decisions can be tested before they are deployed at scale.
Current directions
- COUSINE: a COUpled Soil Inorganic-organic carbon model for eNhanced wEathering, representing interactions with soil chemistry, biology and management
- Matrix-based approaches for fast spin-up and uncertainty attribution in land models
Read more
Process-guided AI
Interpretable AI and foundation models for biogeochemistry.
Our AI is built to respect, and to discover, the physics and biology of soils. We embed mechanistic models inside neural networks so that learned parameters stay interpretable, and we are building PASCA — a Process-guided, AI-assisted Soil Carbon foundAtion model for verifiable carbon dioxide removal.
Current directions
- PASCA: a process-guided, AI-assisted soil carbon foundation model
- BINN: Biogeochemistry-Informed Neural Networks
- ScIReN: Scientifically-Interpretable Reasoning Networks for discovering hidden relationships
- PRODA: PROcess-guided deep learning and DAta-driven modelling
Read more
For an overview of our research, watch Feng's talk at the Climate Change AI seminar.