We studied how farming practices like cover cropping, reduced tillage, and crop diversity relate to soil carbon on 28 California organic farms. Fields with more continuous plant cover and less tillage tended to have more particulate organic carbon, with weaker but positive signals for slower-cycling mineral-associated carbon. Management explained trends in particulate and mineral-associated carbon more than soil properties, suggesting that farming practices could help build soil carbon.

This study compares neural networks and a random forest model for predicting soil erosion in agricultural cropland using long-term data from northern Germany. All models captured general erosion patterns, while more complex neural networks slightly improved the distinction between soil loss classes. A permutation importance analysis identified slope and machine direction vs. aspect as the most influential predictors across all models.