A new study in the journal SOIL shows that combining three agroecosystem models, DAYCENT, DNDC, and ECOSYS, predicts crop yields and soil organic carbon across US croplands more accurately than any single model. The ensemble approach reduced error for corn, soybean, and soil carbon estimates while also making structural uncertainty visible through the spread between models. The researchers simulated a corn-soybean rotation at 4 km resolution and found that individual models disagreed sharply, with ECOSYS underpredecting corn yields and DNDC overestimating soil carbon in some regions. The ensemble smoothed those biases into a more reliable national picture, but the authors stress it is not ready for farm-scale carbon credit quantification. Still, the work offers a template for national greenhouse gas inventories and carbon farming programs that need transparent, defensible estimates.
