Land management could help mitigate climate change by sequestering atmospheric carbon dioxide as soil organic carbon (SOC). The impact of a given management change on the SOC content of a given volume of soil is generally unknown, but is likely moderated by features of the land that collectively determine its sequestration potential. To maximize sequestration, management interventions should be preferentially applied to fields with the highest sequestration potential and the lowest cost of application. We present a design-based statistical framework for estimating sequestration potential, average treatment effects, and optimal management policies from a randomized experiment with baseline covariate information. We review the myriad and nested sources of uncertainty that arise in this context and formalize the problem using potential outcomes. We show that a particular regression estimator -- regressing field-level SOC on management indicators and their interactions with covariates -- can help identify effective policies. The regression estimator also gives asymptotically valid inference on average treatment effects under the randomized design -- without modeling assumptions -- and can increase precision and power compared to the difference-in-means $T$-test. We conclude by discussing the saturation hypothesis in relation to sequestration potential, other study designs including observational studies of SOC, models for policy costs, nonparametric inference, and broader policy uncertainties.

翻译：土地管理可通过将大气二氧化碳固存为土壤有机碳（SOC）来帮助缓解气候变化。特定管理措施变化对给定土壤体积中SOC含量的影响通常未知，但可能受到土地特征的调节，这些特征共同决定了其固存潜力。为最大化固存效果，管理干预应优先应用于具有最高固存潜力和最低实施成本的田地。我们提出了一种基于设计的统计框架，用于从具有基线协变量信息的随机化实验中估计固存潜力、平均处理效应和最优管理政策。我们回顾了该背景下产生的多重嵌套不确定性来源，并使用潜在结果形式化该问题。我们证明一种特定的回归估计量——将田块级SOC对管理指标及其与协变量的交互项进行回归——有助于识别有效政策。该回归估计量还能在随机化设计下（无需建模假设）提供关于平均处理效应的渐近有效推断，并且相较于均值差异$T$检验可提高精确度和统计功效。最后，我们讨论了与固存潜力相关的饱和假说、包括SOC观测性研究在内的其他研究设计、政策成本模型、非参数推断以及更广泛的政策不确定性。