| Summary: | Accurate digital soil maps of soil organic matter (SOM) are needed to evaluate soil fertility, to estimate stocks, and for ecological and environment modeling. We used 5982 soil profiles collected during the second national soil survey of China, along with 19 environment predictors, to derive a spatial model of SOM concentration in the topsoil (0–20 cm layer). The environmental predictors relate to the soil forming factors, climate, vegetation, relief and parent material. We developed the model using the Cubist machine-learning algorithm combined with a non-parametric bootstrap to derive estimates of model uncertainty. We optimized the Cubist model using a 10-fold cross-validation and the best model used 17 rules. The correlation coefficient between the observed and predicted values was 0.65, and the root mean squared error was 0.28 g/kg. We then applied the model over China and mapped the SOM distribution at a resolution of 90 × 90 m. Our predictions show that there is more SOM in the eastern Tibetan Plateau, northern Heilongjiang province, northeast Mongolia, and a small area of Tianshan Mountain in Xinjiang. There is less SOM in the Loess Plateau and most of the desert areas in northwest China. The average topsoil SOM content is 24.82 g/kg. The study provides a map that can be used for decision-making and contribute towards a baseline assessment for inventory and monitoring. The map could also aid the design of future soil surveys and help with the development of a SOM monitoring network in China.
|
|---|