Landslide susceptibility assessment and attribution analysis in Yunnan Province based on weighted information value-logistic regression model

Landslide susceptibility assessment and attribution analysis of triggering factors are essential for regional risk management. However, existing methods face challenges such as subjectivity in determining factor weights and insufficient capacity to reveal the complex nonlinear mechanisms and causal relationships of landslide occurrences. To address these issues, this study proposes a GeoDetector-based Weighted Information Value-Logistic Regression (WIV-LR) model for Yunnan Province, combined with the Geographical Convergent Cross Mapping (GCCM) method to explore the complex causal relationships between susceptibility and influencing factors. The results show that: (1) the WIV-LR model achieves high predictive accuracy (AUC = 0.886), effectively predicting landslide occurrences in Yunnan; (2) landslide susceptibility exhibits significant spatial heterogeneity, with very high and high susceptibility zones mainly distributed in western, central, and northeastern Yunnan, accounting for 41.14% of the total area; (3) GCCM reveals significant bidirectional causal relationships between elevation, slope, soil moisture, rainfall, and landslide susceptibility, while lithology and seismic magnitude show unidirectional causal relationships. Elevation, slope, and relief control the distribution of gravitational potential energy and serve as the main driving forces for landslides. This study provides a scientific basis for landslide risk assessment, targeted prevention, and disaster reduction planning in Yunnan and similar regions.