Landslide Susceptibility Mapping Assessment Method Based on the IVM-BiTCN–Transformer Model

Landslide susceptibility assessment acts as a core technical tool for geological disaster governance, ecological protection and long-term risk mitigation strategies. This modeling approach quantifies the possibility of slope-collapse events and delivers objective decision-making support for regional geologic environment supervision. To overcome the low computational efficiency and weak capacity of conventional evaluation frameworks to extract multi-level spatial grid rules, this paper takes Nanning City, the capital and largest city of the Guangxi Zhuang Autonomous Region in southern China, as the research object. Ten types of terrain and geological control factors combined with historical landslide inventory records are adopted to build a two-stage coupled evaluation framework integrating the information value method (IVM), a Bidirectional Temporal Convolutional Network (BiTCN) and Transformer, named IVM-BiTCN–Transformer. The hierarchical framework first adopts IVM to finish preliminary hazard grading and calculate factor contribution weights, then inputs classified grid samples into the BiTCN-Transformer module to realize local terrain feature and global factor fusion, which significantly lifts the overall identification precision. Ten widely adopted landslide evaluation algorithms are selected for contrast simulation, with multiple quantitative metrics adopted to judge model reliability. Experimental outcomes prove that the presented IVM-BiTCN–Transformer framework obtains superior hazard discrimination capacity, which can raise the precision and stability of landslide zoning and offer reliable technical support for targeted regional geological disaster prevention.