Topographic gradients shape territorial-space evolution and their nonlinear drivers in a karst mountain region of Southwest China

Abstract Karst mountainous areas are characterised by fragmented terrain and high ecological sensitivity, which make territorial-space evolution strongly terrain dependent. However, existing studies often treat terrain merely as a general control variable, leaving the spatial transitions and nonlinear driving mechanisms across different terrain gradients insufficiently understood. Taking the karst mountainous area of northwestern Guangxi as a case study, this study integrates land-use, natural environmental, socioeconomic and locational data from 1980 to 2020. A transition matrix, terrain distribution index (TDI), and explainable machine-learning models were employed to identify the terrain-gradient effects and driving mechanisms of territorial-space evolution. The results show that vegetation-dominated ecological space (VES) consistently accounted for more than 80% of the study area. Nevertheless, the study period was accompanied by the continuous expansion of urban space (US), whose area increased by approximately 397.66% from 1980 to 2020, while ecological space (ES) and agricultural space (AS) declined overall. Transitions from AS to ES were mainly concentrated in middle- to high-elevation and relatively steep areas, whereas transitions from AS to US were clustered in low-elevation and gentle-slope areas. Aquatic ecological space (AES), US and AS tended to occur in low-elevation and gentle-slope areas, whereas VES was dominant in high-elevation and steep-slope areas. The random forest (RF) model performed best, with an R 2 of 0.791 on the test set, outperforming ordinary least squares (OLS) and extreme gradient boosting (XGBoost). Shapley additive explanations (SHAP) analysis based on the RF model further indicated that the normalised difference vegetation index (NDVI), population density (POP) and distance to rivers (DRIV) were the leading explanatory factors. The dominant roles of natural environmental, socioeconomic and locational accessibility factors shifted across elevation and slope gradients and exhibited clear threshold effects. These findings show how terrain gradients are associated with territorial-space patterns, transition pathways and driving mechanisms, and provide a scientific basis for territorial-space optimisation in karst mountainous areas.