Prediction of soil organic matter using multi-temporal satellite images in the Songnen Plain, China

• Published in: Geoderma Vol. 356; p. 113896
• Main Authors: Dou, Xin, Wang, Xiang, Liu, Huanjun, Zhang, Xinle, Meng, Linghua, Pan, Yue, Yu, Ziyang, Cui, Yang
• Format: Journal Article
• Language: English
• Published: Elsevier B.V 15.12.2019
• Subjects: carbon sinks; China; crop management; crop residues; melting; moderate resolution imaging spectroradiometer; MODIS; Optimal input variables; plowing; prediction; rain; remote sensing; snowmelt; soil organic carbon; Soil organic matter; soil sampling; soil surveys; soil water; Spectral index; Temporal information; China; Soil organic matter; Optimal input variables; Spectral index; MODIS; Temporal information
• DOI: 10.1016/j.geoderma.2019.113896
• ISSN: 0016-7061

Due to confounding factors such as crop residue and soil moisture, soil organic matter (SOM) is usually estimated from soil samples in a laboratory or in the field at a local scale. In this study, laboratory and field data of crop residue, soil moisture, crop management practices, and SOM content were used in concert with multi-temporal MODIS images captured during bare soil periods over three years to construct spectral indices, which were then used as input variables to build a regional-scale SOM prediction model. Results showed that: (1) multi-temporal satellite images can be used to predict SOM content at a regional scale; (2) crop residue cover and time interval between snow melt, rainfall, and ploughing determined the optimal input variables for SOM prediction; (3) compared to a SOM model based on a single image, a multi-temporal model reduced the influence of soil moisture and improved both the stability and the accuracy of SOM prediction; (4) the best models generally used the ratio of MODIS Band 6 and Band 1 (R61) as an input variable, as R61 showed good correlation with SOM and less correlation with moisture; and (5) comparing different models in different years showed that models performed better in years with less crop residue. The study results can be used to improve the accuracy of quantitative estimates of the soil organic carbon pool and provide assistance in digital soil mapping. •SOM prediction was influenced by soil moisture, crop residue, and cultivation.•Imagery under uniform environmental conditions produced the best SOM predictions.•Multi-date spectral indices were better predictors of SOM than single-date indices.•Indices that accounted for soil moisture performed better in SOM inversion models.