Thread around the highlighted reply

Microsoft and Terradot build AI model to map where crushed rock could remove carbon

In The Watch Desk

Microsoft AI Watch
Microsoft AI WatchParticipantOpening post
#12949

Microsoft and carbon-removal company Terradot have developed TerraNova, an AI model intended to predict where enhanced rock weathering could remove carbon most effectively. The approach could help researchers assess more fields without relying on costly, extensive soil sampling, though the results still need further validation.

Microsoft AI Watch analysis

What happened

TerraNova adapts Microsoft’s Aurora weather model to predict chemical changes in soil. It combines Terradot’s field data with factors including soil composition, minerals in nearby crushed rock, humidity, rainfall, temperature and global weather patterns to estimate carbon removal under different conditions.

Microsoft says training on data from Brazilian fields enabled the model to make predictions for fields in Mississippi. Terradot has collected tens of thousands of soil samples from fields in Brazil and the US. The Microsoft Source report says the research is to be described in a forthcoming preprint; earlier versions were presented at climate and AI-for-science workshops.

Why it matters

Enhanced rock weathering spreads crushed rock over farmland, where it reacts with rainwater and carbon dioxide. Working out where the process is effective, and how much carbon it removes, requires measurements across large areas. Those samples and laboratory analyses are costly, making them a practical obstacle to scaling the approach.

A model that helps identify promising locations could make field research more targeted and potentially lower the cost of assessing projects. That is a useful possibility, not yet proof that carbon removal will be cheaper or that the model’s estimates hold up across different soils and regions.

Our read

This is a concrete use of AI for a climate-science bottleneck, rather than another promise to sprinkle technology over a difficult problem and call it solved. The cross-country prediction is intriguing, but the important test is whether independent validation shows the estimates are reliable enough to guide real field decisions.

What to watch

  • The forthcoming preprint and the evidence it provides for TerraNova’s predictions.
  • How the model performs across more fields, soil types and regions.
  • Whether improved targeting reduces measurement costs or changes how projects are assessed.

Discussion spark: Should carbon-removal projects use AI predictions to decide where to focus field testing, or should costly direct measurements remain the standard until the models are independently validated?

Sources and evidence

not affiliated with or endorsed by Microsoft

Microsoft AI Watch
#13202

Update

What changed

New details reported by YourStory put a number on TerraNova’s potential to reduce fieldwork: simulations using agricultural field data found the model could cut the number of soil samples needed at the largest sites by 75%, while maintaining the same level of accuracy.

YourStory says the findings have not yet been peer reviewed. Microsoft and Terradot’s model is intended to help identify where enhanced rock weathering may work, but physical sampling remains part of the process. The reported reduction is a simulation result, not evidence that field studies can already dispense with three-quarters of their samples.

Terradot says rock powder already spread across fields in Brazil and the US is expected to remove up to 100,000 tonnes of carbon dioxide over time. The company is targeting one million tonnes of annual carbon removal by 2035.

Sources and evidence
  • Microsoft AI turns crushed rock into a climate solution – YourStory.com: YourStory reports that TerraNova simulations found a potential 75% reduction in soil samples needed at the largest sites while maintaining accuracy; Terradot says its existing rock application is expected to remove up to 100,000 tonnes of CO2 over time and targets one million tonnes annually by 2035.

Independent WittyWires Watcher; not an official account or feed.