Research on landslide ecological vulnerability assessment in alpine valley region considering spatial heterogeneity and feature optimization: a case study of Luding County

Introduction Landslide frequency and ecological fragility jointly constrain Luding’s development. Clarifying the regional landslide distribution pattern and associated ecological hazards is critical for scientific disaster prevention and ecological management. Methods This study first considers internal and external geodynamic factors and analyzes the spatial distribution pattern of landslides in Luding County. Secondly, the Geographical Detector Model (GDM) was employed to quantify the influence of various factors on the distribution of landslides. Then, the RX-Stacking ensemble learning model was utilized to assess landslide susceptibility in Luding County. Finally, using land use type changes, the weight of ecosystem service value, and landslide occurrence probability three key factors, this study quantified the ecological damage induced by landslides and constructed an evaluation model for landslide-induced ecological damage. Based on this model and the susceptibility assessment results, an ecological vulnerability assessment of Luding County was conducted. Results The following conclusions were drawn: (1) Landslides in Luding County are densely distributed, with an over-all distribution density of 0.19 sites/km 2 ; (2) Rainfall, distance to fault zones, and Bouguer gravity anomaly gradient had the most significant influence on the distribution of land-slides, with q values of 0.24, 0.18, and 0.10, respectively; (3) Interactions between factors exhibit a nonlinear enhancement effect, with any two-factor synergy significantly surpassing the influence of individual factors on landslide spatial distribution. Among these interactions, the one between rainfall and distance from the fault zone exerts the greatest influence, with a q value of 0.37; (4) Compared with the Random Forest (RF) model and Extreme Gradient Boosting (XGBoost) model, the RX-Stacking ensemble learning model has an AUC of 0.926, and its landslide susceptibility evaluation is better than the other two models, with good generalization; (5) The distribution of landslide susceptibility levels and ecological vulnerability levels exhibits a high degree of consistency. High/extremely high vulnerability zones are predominantly clustered in the eastern region with prominent ecosystem service functions, while low/moderate vulnerability zones are mainly clustered in the western region with weaker ecosystem service capacities. Discussion The GDM results confirm that rainfall distance to fault zones, and Bouguer gravity anomaly gradient factors are the core controllers of landslides in Luding County, providing a basis for identifying high-risk landslide areas. The non-linear enhancement of factor interactions highlights the necessity of multi-factor synergy analysis in landslide risk assessment, avoiding the limitations of single-factor analysis. The superior performance of the RX-Stacking model ensures the reliability of susceptibility results, which is a key prerequisite for accurate ecological vulnerability evaluation. The spatial consistency between high susceptibility and high ecological vulnerability indicates that landslide prevention in the east should be integrated with ecological protection. In contrast, the western region can adopt targeted management strategies based on its weaker ecosystem service capacity. Overall, this study provides a scientific framework for landslide prevention and ecological management in Luding County, and its methods can be referenced for similar mountainous areas facing combined geological and ecological risks.