Multi-level Semantic Retrieval Method for Landslide Disaster Data

How to quickly and accurately find the superior information to meet the needs of disaster assessment tasks in massive spatio-temporal big data of landslide hazards is the key basis for comprehensive disaster reduction and disaster relief. The traditional disaster data retrieval is mainly based on the passive retrieval method of “artificial experience + keywords”, which makes it difficult to balance the accuracy and timeliness of tasks. This paper proposes a multi-level semantic retrieval method of spatio-temporal data for disaster assessment tasks. By establishing an explicit semantic description of data feature requirements and high-level semantic mapping between task requirements and data features, a multi-level semantic matching data retrieval algorithm is designed to realize superior data aggregation for disaster assessment tasks. Application of the proposed method to the landslide hazard assessment of Maoxian County in Sichuan demonstrates its high query efficiency such as a seconds-level retrieval efficiency in dealing with disaster data in a 900 km² and 90 day range. The accuracy of the recommended dominant data set is also significant, and the average closeness of the recommended results under the 60-day time gap threshold is over 90%. The results show that the method can quickly and automatically acquire disaster data according to the mission requirements, thus significantly improving the disaster mitigation emergency response capability.