GeoGraphRAG: A graph-based retrieval-augmented generation approach for empowering large language models in automated geospatial modeling
Geospatial modeling aims to integrate multiple geospatial data sources and geoprocessing functions based on user application demands to address complex geospatial challenges. While large language models (LLMs) have shown remarkable capabilities in semantic understanding and task planning, their limi...
Saved in:
| Main Authors: | , , , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Elsevier
2025-08-01
|
| Series: | International Journal of Applied Earth Observations and Geoinformation |
| Subjects: | |
| Online Access: | http://www.sciencedirect.com/science/article/pii/S1569843225003590 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Geospatial modeling aims to integrate multiple geospatial data sources and geoprocessing functions based on user application demands to address complex geospatial challenges. While large language models (LLMs) have shown remarkable capabilities in semantic understanding and task planning, their limited geospatial domain knowledge still hinders their efficiency in geospatial modeling. Therefore, there is an urgent need to integrate geospatial knowledge with the inherent semantic understanding capabilities of LLMs, enabling their effective application in specialized geospatial domains. In this study, we introduce GeoGraphRAG, a graph-based retrieval-augmented generation (RAG) approach tailored for geospatial modeling. Specifically, GeoGraphRAG utilizes LLMs as intelligent agents to identify user demands, retrieve relevant subgraphs from external graph knowledge base, and injects both structural and semantic information into the LLM. Through the graph-driven solution planning phase, GeoGraphRAG automatically generates modeling solutions and geospatial code in response to specific user demands. The proposed method is evaluated against different baseline approaches across various LLMs based on the benchmark dataset designed for geospatial modeling, and a representative application scenario is presented to illustrate its practical effectiveness. Our experimental results demonstrate that the injection of graph-structured knowledge improves both the efficiency and interpretability of LLM-based geospatial modeling, thereby confirming the effectiveness of our proposed approach. GeoGraphRAG represents the first attempt to introduce graph-based RAG into geospatial modeling and provides novel insights for future research and practical applications in this field. |
|---|---|
| ISSN: | 1569-8432 |