Constructing a Stratigraphic Knowledge Graph (StraKG) with Multi-source Data to Better Understand the Earth’s Rock Layers

The strata formed in different geological time reflect diverse elements of the geosphere, biosphere, and atmosphere in the Earth’s history of evolution. Stratigraphy, thus, is able to provide a comprehensive understanding of the Earth's rock formations from many aspects, such as litho-stratigraphy, biostratigraphy, event-stratigraphy, and chrono-stratigraphy. In the past decades, massive geological survey data have been made open, which present invaluable resources for studying the stratigraphy in different regions and at different scales. However, many of the open datasets are recorded in the literal form with heterogeneous terminology and there is a shortage of efficient approaches to explore and analyze them. In this research, we designed and constructed a Stratigraphic Knowledge Graph (StraKG) to help address that challenge. StraKG has two layers: schema layer and instance layer. In the schema layer, we used community-recognized terminology from geological dictionaries. In the instance layer, we used natural language processing techniques to analyze open data and obtained relationships between strata and entities such as rocks and spatial locations. The records in the two layers were associated to establish a consistent structure of stratigraphic records in StraKG. To verify the functionality of the resulting knowledge graph, we applied it to the Baidu Encyclopedia, which the largest online Chinese encyclopedia and has heterogeneous labels of stratigraphic sections and massive records of stratigraphic descriptions. A list of three experiments were implemented on the topics of stratigraphic correlation, spatial distribution of a certain stratum, and spatio-temporal distribution of open stratum data. The results show that StraKG is able to provide a powerful reference for the quantitative study of stratigraphy. Used together with data exploration and data mining methods, StraKG illustrates a new approach to analyze the open and big literature data in geoscience.