Identification of topographic factors for gully erosion susceptibility and their spatial modelling using machine learning in the black soil region of Northeast China

• Published in: Ecological indicators Vol. 143; p. 109376
• Main Authors: Huang, Donghao, Su, Lin, Fan, Haoming, Zhou, Lili, Tian, Yulu
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 01.10.2022; Elsevier
• Subjects: agricultural development; algorithms; Catchment areas; China; forestry equipment; gully erosion; Gully erosion susceptibility; Northeast China; Random forest; Remote sensing; risk; sediments; soil; streams; surface roughness; Topographic attribute; topography; watersheds; China; Northeast China; Random forest; Gully erosion susceptibility; Topographic attribute; Catchment areas; Remote sensing
• DOI: 10.1016/j.ecolind.2022.109376
• ISSN: 1470-160X

[Display omitted] •Gully erosion susceptibility (GES) of black soil region in China was assessed.•Ten topographic factors were used to predict gully erosion using random forest model.•Results revealed considerable potential for gully erosion (35.36–42.69 %)•Landforms close to channel network systems were more prone to gully erosion. Gullies are primary sediment sources that restrict sustainable agricultural development by reducing the quality of soil and destroying farmlands. In the black soil regions of Northeast China, the intensive exploitation and unreasonable cultivation result in severe soil erosion and malignant expansion of gully erosion. However, few gully erosion susceptibility (GES) assessments have been performed in this area, despite being an effective tool to explain the potential of gully occurrence. In this study, four adjacent catchments were selected in Keshan County, the black soil region of Northeast China. The locations of gullies were identified through extensive field surveys and interpreting remote sensing images. We used the random forest machine learning method to establish the spatial relationship between the gully occurrence and ten topographic factors (slope aspect, catchment area, channel network distance, elevation, LS-Factor, plan curvature, profile curvature, stream power index, surface roughness index, and topographic wetness index) and mapped the spatial distribution of GES. The mean decrease accuracy was calculated to identify the importance of the selected variables. The efficiency of the results was tested using the area under the receiver operating characteristic curve (area under the curve, AUC), accuracy, and kappa coefficient. The results indicate that 35%–42% of the total area in the study region presents high or elevated levels of GES. Although the importance of the topographic factors differed for the four catchments, the LS-Factor and channel network distance were the most important factors that affected gully spatial distribution. The AUC (0.805–0.846), accuracy (0.705–0.754), and kappa coefficient (0.715–0.788) indicated that the random forest model provided a reliable spatial distribution of GES in the study area. Our study demonstrates the potential risk of gully erosion in the black soil region of Northeast China.