AI and Satellites Verify Soil Carbon Sequestration

This article provides an overview of how regenerative agriculture platforms verify soil organic carbon (SOC) sequestration by combining multispectral satellite imagery with physical soil core measurements using artificial intelligence (AI). While physical soil sampling is accurate, it is too expensive to perform across entire farms, and satellite imagery alone cannot directly measure underground carbon. By fusing these two data types through machine learning, platforms can accurately map, predict, and verify carbon storage at scale, enabling farmers to participate in global carbon credit markets.

Physical Soil Core Measurements

The verification process begins on the ground with physical soil cores, which serve as the "ground truth" data. Agronomists use GPS-guided strategies to extract soil samples from specific locations across a farm, typically down to a depth of 30 to 100 centimeters.

These samples are sent to a laboratory and analyzed—often through dry combustion—to determine the exact percentage of organic carbon and soil bulk density. While this method provides indisputable data for a single point in the field, collecting enough samples to manually measure an entire landscape is financially and logistically impossible.

Multispectral Satellite Data Collection

To bridge the spatial gaps between physical soil cores, platforms utilize multispectral satellites, such as the European Space Agency’s Sentinel-2 and NASA’s Landsat. These satellites capture data across various wavelengths of light, including visible, near-infrared (NIR), and short-wave infrared (SWIR), revisiting the same fields every few days.

From this imagery, platforms extract specific environmental indicators that correlate with soil health and carbon inputs:

  • Vegetation Indices: Metrics like the Normalized Difference Vegetation Index (NDVI) track crop vigor, cover crop adoption, and overall biomass production, which directly drives carbon intake through photosynthesis.
  • Tillage and Residue Tracking: SWIR bands detect crop residue left on the soil surface and identify whether a field has been tilled, a practice that releases stored carbon back into the atmosphere.
  • Bare Soil Reflectance: When fields have no vegetation, satellites measure the color and moisture of the top layer of soil. Darker soils generally indicate higher concentrations of organic matter.

AI-Driven Data Fusion

Artificial intelligence is the engine that connects the sparse, highly accurate soil core data with the continuous, surface-level satellite data. Platforms use machine learning algorithms—such as Random Forests, Gradient Boosted Trees, and Convolutional Neural Networks—to fuse these datasets alongside localized climate and topographical data.

The AI process works in three distinct steps:

  1. Training: The AI model is fed the exact SOC measurements from the physical soil cores alongside the historical satellite data, weather patterns, and elevation metrics for those exact GPS coordinates.
  2. Correlation: The algorithm identifies complex, non-linear relationships between what happens on the surface and how much carbon is stored underground. For example, it learns how five years of consistent cover cropping (seen via satellite) in a specific clay-heavy soil type and rainfall zone translates to a specific increase in underground carbon (measured via soil cores).
  3. Extrapolation: Once trained and validated, the AI model looks at the satellite data for the rest of the farm where no physical cores were taken. It then predicts the SOC levels for every pixel of the land, generating a comprehensive, high-resolution carbon map.

Verification and Uncertainty Reduction

To issue credible carbon credits, regenerative agriculture platforms must prove to registries that the carbon has actually increased over time. Using AI data fusion, platforms establish a baseline carbon measurement and monitor changes annually.

The AI models also calculate uncertainty margins for their predictions. If the AI identifies an area of a farm where its prediction confidence is low, it directs soil samplers to that exact location the following season. This creates a continuous feedback loop: new soil cores make the AI smarter, reducing the number of physical samples needed in the future while maintaining rigorous verification standards.