A new digital soil mapping method with temporal-spatial-spectral information derived from multi-source satellite images

Abstract

Detailed soil maps are essential for effective agricultural practices and environmental protection. Yet, despite the increasing accuracy of digital soil mapping (DSM) in recent years, generating regional-scale, high-accuracy soil maps remains a challenging task. Our study objective was to propose a new DSM method based on temporal-spatial-spectral (TSS) data and to compare the DSM result with a legacy soil map of China. In this study, 13 Landsat multispectral data from 2000 to 2019 were fused by discrete wavelet transform (DWT) to obtain temporal information, a shuttle radar topography mission digital elevation model (SRTM-DEM) was used as spatial information, and Gaofen-5 satellite hyperspectral data was used as spectral information. TSS information was obtained after the DWT and spectral band segmentation methods were used to fuse the temporal and spectral information, combined it with the spatial information. Then, the TSS information and random forest model were used for DSM. The results indicated that 1) The mapping result based on TSS information was highly correlated with the legacy soil map. In different soil classes, there were minute differences in the core area and large differences in adjacent soil classes between the two maps. The overall accuracy and kappa coefficient of DSM based on TSS information were 88.11% and 0.82, respectively. 2) With the same values of the soil moisture, the overall accuracy and kappa coefficient of DSM based on hyperspectral data were 6.80% and 0.02, respectively, higher than those based on multispectral data. 3) With increasing temporal information, the DSM accuracy continuously increased, and the mapping accuracy based on multi-temporal multispectral data was higher than that based on mono-temporal hyperspectral data when the number of multi-temporal multispectral images reached 6. 4) The DSM accuracy was effectively improved when terrain factors were considered, and the terrain factors featuring a strong separability for various soil classes differed. The DSM method proposed in this study based on TSS information from multi-sources remote sensing data greatly improved the DSM accuracy and provided new insight for future DSM research.