Best practice for upscaling soil organic carbon stocks in salt marshes

Abstract

Calculating the amount of soil organic carbon (SOC) stored in coastal environments, including salt marshes, is needed to determine their role in mitigating the Climate Crisis. Several techniques exist to calculate the SOC content of a unit of land from the upscaling of soil cores. However, no comprehensive assessment has been made on the performance of commonly used SOC upscaling techniques until now. We measured the SOC content of soil cores gathered from two Scottish salt marshes. Two SOC values were used for upscaling; SOC content for a 1 m standardised depth (as recommended by the IPCC), and SOC content of the modern marsh deposit (identified in the stratigraphy as a transition from organic-rich (marsh) to mineral-rich (intertidal flat) soil. Twenty-two upscaling techniques were used (SOC content × area, interpolative, and regression-based extrapolative calculations). Leave-one-out cross-validation procedures and prediction interval widths were used to assess the accuracy of each technique. Digital Terrain Models and Normalized Difference Vegetation Indices were used as covariate surfaces in the regression models. We found that marsh-scale SOC stocks varied by as much as fifty-two times depending on which sampling depth and upscaling technique was used. The largest differences emerged when comparing SOC stocks upscaled from 1 m deep and modern marsh deposits. Using the IPCC recommended 1 m sampling depth inflated the SOC stocks of salt marshes, as intertidal flat environments were included in the calculation. Ensemble regression models from the weighted average of seven machine learning algorithm outputs produced the highest upscaling accuracies across marshes and sampling depths. Simple SOC content × area calculations produced marsh-scale SOC stocks that were comparable to stock values produced by more advanced ensemble regression models. However, regression models produced detailed maps of SOC distribution across a marsh, and the associated uncertainty in the SOC values. Our findings are broadly applicable for other environments where large-scale SOC stock assessments and uncertainty are needed.