Modeling Remote Sensing Data Relationships with Spatiotemporal Knowledge Graphs

Understanding spatiotemporal relationships in environmental monitoring is crucial for effective decision making and analysis. In this work, we leverage knowledge graphs (KGs) to model such relationships using measurements collected from field sensors. Specifically, we present datasets from a prescribed fire at Sedgwick Reserve, Santa Barbara, and introduce an ontology that defines key relationships between entities in these datasets. To facilitate data integration and analysis, we developed an ingestion pipeline that structures the information into a KG and facilitates querying along spatial and temporal edges. We evaluate our implementation with two tasks, the ability for spatial relationships to improve vegetation metric prediction, and the use of the inferred relationships between carbon emission data and vegetation metrics to estimate prescribed fire outcomes. Our approach highlights the potential of knowledge graphs in environmental data management, offering a scalable and interactive framework for analysis and decision support.