Construction and Application of Geographic Knowledge Graph for Landslide Disaster Emergency

Abstract Landslide is a common geological disaster worldwide, causing significant harm to human lives, socio-economic development, and the ecological environment. Effective emergency response to landslide requires coordination among multiple departments, while the complex terrain and dam aged roads at disaster sites increase the difficulty of rescue operations. Geographic knowledge graph technology can represent multiple disaster response data in graph form to highlight the complete relationships between the integrated entities. It enhances systematic processing of disaster information to provide scientific decision support for emergency responses. The previous studies on knowledge graph for landslide disaster emergency focused on cause analysis and spatio temporal evolution by using textual data. However, they lacked descriptions of spatial relationships and systematic analysis, which limited their practical application. To solve this problem, a geographic knowledge graph was proposed for landslide emergency response and was implemented in Zhouqu County in China, a region prone to landslides. Firstly, a unified ontology framework was defined, including three ontology concepts named as static base map layer, real-time data layer, and domain information layer, along with spatial, hierarchical, and semantic relationships. Then, the entities, relationships among them, and their attributes were extracted from the data of road networks, emergency response points of interest, and real-time data to form two types of triples including entity-relationship-entity and entity-attribute-value. Next, the ontology alignment methods were applied to eliminate redundant information. The triples were mapped into the unified ontology framework and stored in the Neo4j graph database, forming the geographic knowledge graph for landslide emergency response. Finally, the geographic knowledge graph was used for landslide emergency task assignment and route planning. The experimental results showed that the geographic knowledge graph for landslide emergency response can be effectively extracted and constructed based on the proposed method by integrating multiple spatial data, and it can perform well for decision support.