Abstract: Big data has provided new opportunities for landslide research, but in-depth mining of landslide data is still limited due to complex data types, diverse semantic relationships, and unclear sharing mechanisms. And it is difficult to give full play to the advantages of big data in landslide research. In this paper, a landslide knowledge graph construction method for the field of engineering geology is proposed, which extracts, fuses and structures multi-source heterogeneous landslide knowledge to realize the inquiry, correlation and reasoning of landslide knowledge big data. The combination of top-down and bottom-up methods was used to divide the concept and ontology of landslides. A knowledge graph model layer was formed based on 10 categories of knowledge, including landslide field investigation, landslide evaluation, landslide type, landslide geomorphological characteristics, landslide morphological characteristics, landslide disaster information, states of activity of landslides, landslide genesis mechanism, landslide stability analysis method, and landslide prevention and control measures. The knowledge graph model layer includes concept layer, attribute layer, relationship layer, rule layer, and instance layer, and it is constructed by extracting landslide knowledge information from broad data sources, establishing semantic networks and integrating redundant knowledge. The Neo4j platform is used to store landslide knowledge, realize knowledge visualization and retrieval, and provide new ideas and methods for landslide mechanism research and disaster prevention and mitigation. The proposed construction method of landslide knowledge graph can be extended to the research of other types of disaster knowledge graph, and is linked with other disciplines to promote the deep intersection and integration of disciplines.