Optimal soil organic matter mapping using an ensemble model incorporating moderate resolution imaging spectroradiometer, portable X-ray fluorescence, and visible near-infrared data

Abstract

Digital soil mapping of soil organic matter (SOM) is necessary because of its importance for carbon sequestration, soil health, and food security. However, large-scale digital soil mapping of SOM remains a challenge to improve accuracy and provide more local information. To address this knowledge gap, a strategy that incorporatesremoteandproximalsensingdataintoanensemblemodelforthe digital soil mapping is proposed. In this study, moderate resolution imaging spectroradiometer, portable X-ray fluorescence, and visible near-infrared spectroscopy data from 402 soil samples were used to map the SOM at a resolution of 90 m, model the SOM with random forest, cubist, and ensemble models, and evaluate its environmental factors in the northeastern and northern Chinese plains, a typical agricultural plain (approximately 66,000,000 ha). The digital maps of the SOM were derived along with their uncertainties. The results show that the visible near-infrared and portable X-ray fluorescence data play an important role in the SOM distribution. Inclusion of these data in the model, improved the R2p by 6.25–35.42 %, and reduced the root mean square error of the prediction by 0.30–1.54 g kg−1. The ensemble model, which included remote and proximal sensing variables, outperformed the results of previous studies with a root mean square error of 6.68 g kg−1 and provided more detailed information. Thus, this study confirmed the effectivenessoftheproposedstrategy. The SOM product obtained by this strategy was able to accurately control soil management, and the structural equation model showed that human activities exerted a direct influence on SOM comparable to the overall effect of natural factors on SOM. The contribution of straw mulching to human activities was great with path coefficient >0.50. This suggests that land management, especially straw mulching, should be improved in this area. In summary, this study proposed a novel strategy for accurately and efficiently obtaining SOM products for agricultural decision-making, terrestrial carbon cycling, and carbon stock estimation. Future advancements will focus on more accurate and detailed global SOM maps or carbon storage by integrating multiple models with remote sensing data and proximal sensing data from available soil spectral libraries.