A fine evaluation of population density at slope scale in mountainous areas considering the spatial variation of environmental factors

Accurate population density evaluation in mountainous regions is critical for landslide risk assessment. Current methods using mixed urban-rural approaches often fail to capture the unique characteristics of low-density mountain areas, leading to significant inaccuracies. Additionally, traditional assessment units like administrative or grid units are misaligned with slope units used in landslide risk assessment, limiting practical integration. To address these issues, this study proposes a novel evaluation framework for fine-scale population mapping by integrating slope unit-based spatially variable environmental factors, featuring several key aspects: (1) Slope units are adopted as the fundamental evaluation units to enable seamless integration of population exposure data; (2) Population density sampling is focused on mountainous regions with an urban density threshold; (3) Random forest (RF) model is employed to effectively capture complex relationships, with input features engineered to represent the spatial variability of environmental factors within each slope unit; (4) The optimal model is interpreted using SHAP analysis to reveal the driving factors of population distribution. Experimental results show the RF model with spatial variability achieves high accuracy (R2 = 0.94, RMSE = 2.82 persons/hectare). This method provides high-resolution, interpretable population data to improve risk assessment and support disaster preparedness and planning in mountainous regions.