Trends in accuracy and bias of soil organic carbon mapping methods: A global synthesis of 37 studies

This is the authors' abstract. We don't add key points for this paper.

Spatial information on soil organic carbon (SOC) stocks is essential for assessing soil fertility and ecosystem services. However, the influence of different mapping methods on SOC map accuracy and bias remains insufficiently quantified. This study synthesizes evidence from 37 published studies comparing four main approaches: basic interpolation, geostatistical methods, statistical and machine learning models, and remote sensing. Overall, SOC mapping accuracy varied substantially across methods, with a median mean absolute error (MAE) of approximately 40% of mean SOC stocks. Statistical and machine learning approaches, as well as hybrid methods, were generally associated with comparatively lower errors, although results were strongly context-dependent and influenced by study design heterogeneity. Bias across methods was generally low, but differences among approaches were observed. Mapping errors were strongly influenced by spatial scale, increasing with study area size and reaching very high values in large regions. Errors also showed moderate relationships with environmental factors such as precipitation and soil carbon levels, indicating that both methodological choice and environmental context contribute to prediction uncertainty. These findings highlight substantial variability in SOC mapping performance and demonstrate that scale and environmental conditions strongly influence reported model errors. The study emphasizes the need for standardized evaluation metrics, improved reporting consistency, and broader integration of environmental and management drivers in future SOC mapping efforts. Interpretation is limited by study heterogeneity and potential publication bias inherent in the 37 available studies.