Precision Soil Carbon Mapping on Agrobrazil Field T3 (78 ha)
We evaluated our AI model on a 78 ha Brazilian field named T3. With a single local sample and 600 internal analogues, we nearly halved the prediction error and generated a high-resolution SOC map.
Our breakthrough
Before local calibration, the model MAE on T3 was 1.49 g/kg. After adding one representative sample and 600 internal samples, MAE dropped to 0.77 g/kg, producing reliable soil-carbon insight with minimal field work.
How it works
- Smart data integration: multispectral imagery, terrain and climate data, plus global and Brazilian soil samples train the baseline model.
- Local calibration: one field-representative sample and 600 similar internal samples fine-tune the model to local soil conditions.
- High-resolution mapping: the recalibrated model generates a 30 m SOC map with a median SOC of 12.24 g/kg and visible management zones.
Why it matters
- Better accuracy: MAE below 1 g/kg gives farmers confidence in the map.
- Actionable insight: median SOC values support targeted soil-building and fertilization.
- Lower cost and time: minimal sampling still delivers high-quality maps with fewer lab costs and field trips.
See it in action
Want more information or a personalized solution?
Contact us