Determination of future gully erosion risk and its spatially quantitative interpretability of driving factors in regional scale using machine learning algorithms

Abstract

Gully erosion has already caused severe damage to land production and ecosystem safety. It is pivotal to identify areas sensitive to and at risk from gully erosion for guiding gully erosion prevention and control efforts. Existing studies typically use gully erosion susceptibility modeling (GESM) to assess the severity of gully erosion at the watershed scale, focusing primarily on whether a gully forms, but not accounting for the risk of further gully development. This study proposed a new gully erosion risk modeling (GERM) method by combining GESM reflecting gully erosion potential and gully density modeling (GDM) reflecting current gully erosion situation in the whole rolling hilly region of northeast China with an area of 177,584 km². A stratified random sampling (SRS) method was used to investigate the current state of gully erosion in study area. Gullies were interpreted as line features based on Google Earth images with the resolution of 0.3–0.6 m and further confirmed by field investigation, so all gullies in study area is spatially accurate. The results showed that the mean gully length and gully density were 335.85 m and 250.35 m km⁻², respectively. The GERM was realized by four machine learning algorithms including Random Forest (RF), XGBoost, K-Nearest Neighbor (KNN), and Multi-layer perceptron of artificial neural networks (ANN-MLP). The results showed that XGBoost demonstrated the best overall performance (AUC = 0.999). The global Moran’s I index indicated significant spatial clustering of gully erosion risk and no risk. Although gullies tended to be stable in 44 % of the study area, the high risky zones accounted for 12.9 % which located in the eastern, central-northern, northwestern, and southern of study area. Additionally, we found that climate was a key factor driving gully formation, both topography and climate jointly impacted the current state or degree of gully erosion development, and the risk for further gully development was influenced by a combination of topography, climate, and human activities. This study revealed the future risk of gully erosion under the existing topography, climate, and human activities conditions in the rolling hilly region of northeast China, and explored the optimal algorithm for GERM, which can serve as a reference for the investigation, simulation, and management of gully erosion.