Gully erosion susceptibility mapping considering seasonal variations of NDVI using a machine learning approach in the Mollisol region of China

Gully erosion is the most severe form of soil erosion, and mapping gully erosion susceptibility accurately and automatically is crucial for guiding policy decisions. Topography and vegetation cover were general factors for assessing gully susceptibility, yet little attention has been paid to the spatiotemporal variations in vegetation. This study aims to predict gully-prone areas using stable factors (topography, hydrology, soil, etc.) considering the monthly variability in vegetation based on a machine learning approach in the Mollisol region of China. A total of 1890 gully and non-gully points were extracted to establish an inventory database. Twelve treatments were conducted including the stable factors and individual NDVI from January to December, respectively. All potential factors were evaluated for contributing to the gully erosion prediction, and a set of rules based on accuracy, AUC, and kappa were used to evaluate the model performance. The results demonstrated that NDVI varied widely between gully and non-gully areas and the importance of NDVI varied in diverse months. NDVI in August was the most important explanatory factor (25 %) to gully occurrence mapping, followed by the plan curvature (14 %), and elevation (13 %), respectively. The gully-prone areas predicted by NDVI in August exhibited higher accuracy, followed by that in May and June. This was attributed to the greater difference in NDVI between the gully and non-gully areas in June (0.30), May (0.23), and August (0.16). Overall, the very low, low, moderate, high, and very high gully susceptibility levels occupied 35 %, 23 %, 18 %, 14 %, and 10 % of the study area, respectively. This study advances our understanding of spatial-temporal heterogeneity in NDVI among gully and non-gully areas that need to be considered in gully mapping. Further, an automatic and accurate gully mapping approach can provide valuable information to identify areas where urgent and appropriate measures should be applied.