AUTHOR=Wang Xianmin , Du Aiheng , Hu Fengchang , Liu Zhiwei , Zhang Xinlong , Wang Lizhe , Guo Haixiang 

TITLE=Landslide susceptibility evaluation based on active deformation and graph convolutional network algorithm

JOURNAL=Frontiers in Earth Science

VOLUME=Volume 11 - 2023

YEAR=2023

URL=https://www.frontiersin.org/journals/earth-science/articles/10.3389/feart.2023.1132722

DOI=10.3389/feart.2023.1132722

ISSN=2296-6463

ABSTRACT=Landslides are a type of disastrous natural hazard and cause enormous losses to human lives, critical infrastructures, and social economy. Moreover, a huge amount of hidden landslides have not been discovered and posed a great threat to the sustainability development of society and economy. However, identification of hidden landslides has always been a worldwide challenge because of good concealment, steep topography, and hard-to-reach positions. Thus, there is an urgent need to solve the bottleneck problem and discover the hidden landslides. In addition, landslide susceptibility evaluation (LSE) can predict the regions where landslides are the most possible to occur. Therefore, hidden landslide discovery and LSE are valid measures for mitigation of high landslide risks. This work suggests a universal rule set of hidden landslide discovery, and these rules can improve the identification accuracy of landslides. In addition, this work evaluates landslide susceptibility considering both known landslides and the discovered hidden landslides to make the LSE map more rational and exact. Moreover, a novel graph convolution network (GCN) is constructed and for the first time employed in  LSE. The universal rules, LSE idea, and new GCN algorithm are applied to Wanzhou County, a famous area for serious landslides, and achieve a relatively high accuracy. Moreover, the reason for frequent landslides is revealed.