PubMed · 2026-06-30
Researchers built machine learning models to predict how much biochar improves a soil's ability to hold nutrients (cation exchange capacity). The CatBoost model proved most accurate, and analysis revealed that biochar made at higher temperatures with a small surface area delivers the best soil improvements.
The CatBoost algorithm predicted soil nutrient-holding capacity with 96.3% accuracy (R² = 0.963), outperforming LightGBM, Deep Neural Networks, and Random Forest models.
Higher pyrolysis temperature in biochar production correlated with greater soil CEC improvement, while biochar with high inherent CEC counterintuitively reduced soil CEC gains, likely by competing for cations in the soil solution.
Biochar with a specific surface area below 50 m²/g produced at high pyrolysis temperatures achieved the best soil improvement outcomes across studied conditions.