ML for Carbon Capture and Storage

Carbon deposit estimation is complex, highly variable, and dependent on multiple environmental factors. Terrasphere needed a reliable way to quantify carbon for monetisation, but existing methods were manual, inconsistent, and lacked predictive accuracy.

Solution

We built hybrid ML models - combining Random Forest with fine‑tuned modern techniques and Graph Neural Networks (GNN's) to learn patterns from soil, climate, and weather datasets. These models powered an automated MRV pipeline capable of predicting carbon deposits with strong accuracy (as shown in the screenshots).

Impact

• Accurate carbon predictions enabling Terrasphere to monetise deposits confidently.
• Automated MRV workflow replacing manual estimation with scalable, data‑driven modelling.
• Research‑grade results validated through UKRI funding and university collaboration.