Counterfactual designs for quantifying management effects on soil carbon change at multi-field scales

Agricultural policies for climate mitigation and adaptation, and farmer decisions about soil health, both rely on sustainable management of soil carbon. Controlled field experiments demonstrate that regenerative practices increase soil carbon, yet expected changes are often small relative to spatial and sampling variability. These sources of variation are assumed to hinder the quantification of management effects at the scale of working landscapes. However, spatial variability in soil carbon change has rarely been evaluated in the context of estimating management effects. Here, we evaluate these sources of variability in the context of a counterfactual design in an agricultural growing region of the Hudson Valley, NY, USA. We assembled groups of comparable fields from a population of eligible fields by stratifying by initial soil carbon and texture. Within each of these strata, multiple fields under conventional and regenerative practices were selected (80 fields in total) and sampled at two time points in the non-growing seasons of 2023–2024 and 2024–2025 for surface soil carbon concentrations (0–15 cm). We sampled each selected field at a density of 0.6 ha per soil core (∼10 locations per field; 796 sampling points; 1592 samples). Spatial variability in soil carbon was large both within and among fields (between field SD 0.59 %C). Using the difference in soil carbon between the time points removed persistent spatial signal and reduced between-field variability in change more than seven times (SD 0.08 %); within-field variability remained high. Minimum detectable differences from repeated measurement were roughly six-times smaller than those inferred from spatial data alone (0.33 vs. 0.056 %C), providing empirical evidence that spatial variance is a poor proxy for temporal variance. Estimated management effects depended on which fields were selected, whereby selecting fields with higher or lower soil texture altered the estimated treatment effect. Our results suggest that counterfactual designs based on remeasurement and comparable groups can quantify management effects under real-world agricultural conditions. Such studies at the scale of commercial agriculture are needed to build confidence about when, where, and which regenerative practices deliver benefits such as improved soil health, informing policies for more sustainable and climate-smart agriculture.