Exploring influence factors in mapping soil organic carbon on low-relief agricultural lands using time series of remote sensing data

Abstract

Soil organic carbon (SOC) content is of critical importance in managing agricultural fields and understanding carbon fluxes at various scales. At regional and global scales, digital mapping of SOC has been successful using environmental variables such as climatic, land use and terrain factors. However, such a mapping for agricultural lands in low-relief areas cannot be resolved due to weak spatial variations of the environmental variables. An alternative approach, based on time series of remote sensing images, has recently been emphasized in estimating SOC since crop growth status clearly revealed by satellite images is strongly related to SOC. However, there is still a lack of full understanding of impacts of factors such as data source, regression techniques, and spatial resolution on the accuracy of SOC mapping using time series of satellite images. In this study, prediction variables including four spectral bands (i.e., blue, green, red, and near-infrared bands) and the normalized difference vegetation index (NDVI) from Sentinel 2, Gaofen (GF) 1, and Landsat 8 data, acquired between November 2017 to March 2018, were individually used to estimate SOC by stepwise regression (STR), partial least squares regression (PLSR), and extreme learning machine (ELM). Besides, we further explored the influence of satellite parameters, such as scanning time, spectral bands and vegetation indices on predicting SOC using Sentinel 2 time series images. Results showed that: (1) the time series of remote sensing images can be used to map SOC in small- and large-extent low-relief areas with low evaluated root mean square errors (0.199 % and 0.249 %) and high R2 (0.555 and 0.528), (2) SOC estimations from the Sentinel-2 time series images, compared to those from Landsat 8 and GF 1 datasets, exhibited a better accuracy, and the ELM yielded a better prediction accuracy than PLSR and STR, (3) the satellite parameters of the spatial resolution, the acquisition time and the spectral bands significantly affected the prediction accuracy. This study offers more comprehensive information than terrain factors for revealing the influence of satellite parameters on predicting SOC, and provides great help for constructing one more suitable and higher precision soil mapping strategy for agricultural lands in low-relief areas.