Learning vs. understanding: When does artificial intelligence outperform process-based modeling in soil organic carbon prediction?

Importance of on-farm research for validating process-based models of climate-smart agriculture

“…New modeling techniques incorporating artificial intelligence and multi-model ensembles may improve simulation accuracy and uncertainty estimations if the necessary data are available. For example, machine learning (ML) algorithms have the potential to incorporate large quantities of data to model ecological processes with superior performance to traditional process-based models [ 4 ]. Furthermore, multi-model ensembles that incorporate the strengths of several models simultaneously are shown to improve C flux simulations and are currently being pursued by collaborative groups throughout the process-based modeling community [ 51 , 56 , 64 ].…”

“…For example, uncertainties for simulating past SOC stock change in the Century model were as high as ± 118% and ± 739% at subregional and site scales, respectively [38]. Bayesian model analysis frameworks and intermodal comparisons have proven useful for reducing uncertainty [22,64], but the limited amount of measured data for model calibration and validation remain the largest bottle-neck for further reducing uncertainty in quantifying SOC stock changes [4,10,14,22,47].…”