No One Plan can have all that Power: Efficient Designs and Valid Inference for Soil Carbon Measurement and Monitoring on Heterogeneous Landscapes

Accurate and precise measurement of soil carbon (C) is crucial to research and policy supporting soil C sequestration. Unfortunately, spatial heterogeneity of soil C and in-lab analytical variability limit the ability to reliably detect real soil C sequestration when it exists. Worse, errors in sampling design and statistical analysis can lead to false detection of non-existent sequestration. This paper examines sources of uncertainty and recommends improvements for field-scale soil C measurement and monitoring. Using simulations and data from two intensive field sampling campaigns on one rangeland site (n=662) and seven cropland (n=455) sites across California, USA, we show that spatial heterogeneity is the primary driver of uncertainty in agricultural soil C estimates, while overall contribution to error from dry combustion in laboratory settings is relatively low. Furthermore, Students t-test (and associated confidence intervals) can be unreliable for drawing statistical inferences about soil C changes, depending on the baseline distribution of soil C and the effects of management. Typical sample sizes of 10-30 samples cannot guarantee valid quantification of uncertainty with the t-test and are insufficient to detect changes in soil C of the magnitude expected over a few years from recommended management interventions (i.e., no-till, improved grazing management, compost application). We advise: (1) stratification, random sampling, and the use of spatial data for a priori power analyses to determine necessary sample sizes, (2) less compositing of field samples, (3) the use of highly precise analytical lab instruments, and (4) increased use of non-parametric statistical methods to address shortcomings of the t-test, ANOVA, and related parametric methods. These practices would greatly improve the reliability of both soil C research and verification schemes for soil C offsets in C markets.