Objective Massive multi-source remote-sensing data are a critical resource for delivering high-precision services for typical tree species in subtropical regions. Yet the great diversity of species and the cloudy, rainy, mountainous environment characteristics of these regions complicate multi-source remote-sensing features, hindering clear spatio-temporal expression of both basic and composite (state and process) tree-species characteristics. Inspired by geographic knowledge mapping, this study integrated tree-species fundamentals, geomorphological patterns, and multi-source remote-sensing feature knowledge. It proposes a framework for constructing a spatio-temporal remote-sensing knowledge graph for typical subtropical tree species, and by incorporating the DeepSeek large language model (LLM), develops an intelligent Q&A system organizing and expressing tree-species remote-sensing knowledge more effectively.

Method To achieve the above objectives, the study first organized and integrated the basic concepts of tree species, geographic-law knowledge, and remote-sensing interpretation information at three levels: data knowledge, conceptual knowledge, and rule knowledge to build a comprehensive system describing the spatial distribution, temporal dynamics, and geographic context of subtropical tree species. Taking forest class as the interpretation unit, ontology modelling was used to build tree-species ontology knowledge and remote-sensing ontology knowledge for individual temporal states. Temporal and spatial correlations reflecting tree-growth evolution were then introduced to form a spatio-temporal sequence ontology model and to extract relationships among core elements of remote-sensing knowledge for typical subtropical tree species. In addition, forest-inventory data, Baidu Encyclopaedia entries, and remote-sensing pre-interpretation results were fused to enrich the data layer with ontology and attribute information. Finally, entity relationships were stored in a Neo4j graph database to construct the spatio-temporal remote-sensing knowledge graph, while the DeepSeek LLM deployed a local knowledge base to enable knowledge-graph-driven interactive Q&A services.

Result The study successfully constructed a spatio-temporal remote-sensing knowledge graph for typical subtropical tree species, and through deep integration with DeepSeek, developed an intelligent Q&A system supporting natural-language interaction. The system enables associated queries and map-based visualisation of tree-species knowledge, geomorphological knowledge, and remote-sensing knowledge across different states.

Conclusion The proposed approach realises the transformation from “data to knowledge” for subtropical tree species and through the synergistic computing of DeepSeek and the knowledge graph, offers flexible, convenient remote-sensing knowledge queries and services. It supplies data support for knowledge-guided, intelligent remote-sensing interpretation of typical subtropical tree species and provides a demonstrative application for remote-sensing knowledge mapping in other domains. Future work will expand the scale of the knowledge graph to meet the demand for high-quality knowledge services in high-precision forestry remote-sensing monitoring.