The impact of a management intervention on the soil organic carbon (SOC) stored in a given volume of soil may be moderated by observable features that proxy that soil's SOC storage potential under that intervention. To maximize total SOC storage, interventions should be targeted to soils with the highest responses and lowest costs to intervening. We summarize key sources of uncertainty and present a causal framework for estimating SOC storage potentials and optimal management policies. The method models SOC measurements as functions of covariates within each treatment arm, using the fitted models to estimate SOC storage potential for each plot and finding the policy that maximizes the average of those estimates. The modeling can use linear regression or other machine learning algorithms to learn relationships between features and SOC storage. We demonstrate its use in a study of compost amendments applied to California rangelands, finding that storage potential is moderated by baseline SOC and that optimizing the policy could provide a slight gain over uniform application of the intervention with the largest average treatment effect estimate. We evaluate the methods further in simulated field experiments. Simple estimates fared better than machine learning, except in unrealistically large experiments. We conclude by discussing recommendations for practice, baseline SOC moderation, extensions to observational studies, and broader policy uncertainties.

翻译：暂无翻译