Iron-removal learning machine for multicolor determination of soil organic carbon

This study was conducted to accurately estimate soil organic carbon (SOC) for the protection of soil quality.

Soil color analysis using three colors (tricolor) has been suggested as a potential tool to estimate SOC, but its accuracy can be affected by other factors such as iron (Fe) content. This study used an iron-removal learning machine (ILM) method, developed using machine learning concepts, to analyze and remove the influence of total soil Fe on SOC estimates.

Compared with the random forest model, the SOC model developed using ILM exhibited better validation performance, with a coefficient of determination of 0.878, a mean relative error of 13.8%, a root mean square error of 2.316 g/kg, a ratio of performance to deviation 2.465, a ratio of performance to interquartile distance of 4.903, and a concordance correlation coefficient of 0.934.

Results indicate that the ILM method employed here has great potential for estimating SOC. To improve real-time estimates of SOC, future work should include Fe indices within ILM.