Assessing urban expansion into landslide susceptibility zones using machine learning methods: A case study of Yunnan Province, China

Landslide exposure assessment is critical for urban development in landslide-prone regions. However, field-based methods are often prohibitively expensive and impractical for long-term monitoring applications. This limitation can lead to inadequate risk quantification and unplanned urban expansion in vulnerable areas. Taking Yunnan Province, China, as a case study, we employed multiple machine learning models to assess regional landslide susceptibility and integrated multi-temporal GAIA impervious surface data (1990–2018) to quantify urban expansion into high-susceptibility zones. The proposed framework achieved superior performance in both mapping accuracy (AUC = 0.86) and spatial coverage compared to previous studies. Between 1990 and 2018, the extent of impervious surfaces within high-susceptibility zones increased from 193.29 km² to 948.89 km²—a rise of 390.92%—with accelerated growth observed after 2000. Exposure disparities among cities have widened over time, shaped by geographic location, economic structure, and policy orientation. This study produced benchmark landslide susceptibility maps for the province and enabled continuous monitoring of landslide exposure risk over nearly three decades. The findings underscore growing threats to urban safety and offer empirical evidence to inform landslide risk mitigation policies and support sustainable urban development strategies.