TRANSFER LEARNING CONTRIBUTES TO THE SEISMIC LANDSLIDE HAZARD ASSESSMENT IN CENTRAL ASIA

Seismic landslide hazard assessment is of significant importance for earthquake hazard prevention and mitigation in Central Asia. However, current research lacks support for seismic landslide area assessment and fails to fully utilize the rich knowledge from multi-domain seismic landslide data to enhance seismic landslide hazard assessment of target areas. To bridge this gap, this study focuses on seismic landslide hazard assessment in Central Asia using transfer learning. First, seismic landslide inventories and relevant geological and geographical data are extensively collected to establish a seismic landslide database based on slope units. Subsequently, by exploiting the correlation between seismic landslide location assessment and area assessment, a multi-task learning model is developed to achieve synchronous evaluation of seismic landslide location and area. Furthermore, leveraging the similarity between the source domain and the target domain, a multi-source domain transfer learning model is established. By incorporating various historical seismic landslide knowledge, the transfer learning model greatly improves the accuracy of seismic landslide hazard assessment results.