Detection of soil erosion hotspots with accelerated carbon losses in the black soil region of Northeast China as driven by Sentinel-2 multispectral remote sensing

Global efforts to restore the world’s degraded croplands require knowledge on the degree and extent of accelerated soil organic carbon (SOC) loss induced by soil erosion. However, methods for assessing where and to what extent erosion takes place is still inadequate for precise detection of erosion hotspots at high spatial resolution.

 

In this study, we attempted to develop a spectra-based soil erosion mapping approach to pinpoint eroded hotspots in a typical catchment located in the black soil region of Northeast China as characterized by undulating landscapes. We built a ground-truth dataset consisting of three classes of soils representing Severe, Moderate, and Low erosion intensity because of their inter-class contrasts in estimated erosion rates from ¹³⁷Cs tracing. The spectral separability of different erosion classes was first tested by a combined principal component and linear discriminant analysis (PCA-LDA) against laboratory hyperspectral data and then validated against Sentinel-2 derived broadband spectra.

 

We will present results on the performance of the PCA-LDA model to classify soil erosion intensity classes based on laboratory and satellite-based soil spectra. We further identified distinctive spectral features representative of shifting soil albedo and biochemical composition due to erosion-induced SOC mobilization. A classification scheme comprising the spectral features was applied to the Sentinel-2 bare soil composite for pixel-wise soil erosion mapping, from which 15.9% of the catchment area was detected as erosion hotspots while the Moderate class occupied 65.4%.Our study highlights the potential of the spectral-based remote sensing approach for  better targeted cropland management to combat soil degradation.