Enhanced Landslide Susceptibility Assessment in Angren, Uzbekistan Using FR Feature Selection and CNN-GWO Hyperparameter Optimization

초록

This study developed a landslide susceptibility map (LSM) for the Angren area of Uzbekistan by comparing a frequency ratio (FR) model with a convolutional neural network optimized using the grey wolf optimizer (CNN-GWO). The study area features steep mountain slopes characterized by Neogene sedimentary lithology, overlain by Quaternary loess deposits, and is affected by intensive mining activities, making it highly susceptible to seasonal landslides. The susceptibility mapping was performed by integrating landslide inventory with comprehensive topographic parameters, soil characteristics, land-use data, normalized difference vegetation index (NDVI), and geological information. The FR model assessed the statistical correlation between landslide occurrences and individual influencing factors, whereas the CNN-GWO model integrated all raster layers to identify complex spatial patterns in landslide susceptibility. The results show that the susceptibility maps generated by the FR and CNN-GWO models display comparable spatial patterns, with area under the curve values of approximately 78.11% and 80.11%, respectively. These findings provide a useful framework for spatial planning, infrastructure protection, and landslide risk reduction in the mountainous Angren area.

키워드