GeoGraphRAG-Soil: An Open-Source Graph-Based Retrieval-Augmented Generation Framework for Automated Geospatial Soil Modelling

Digital Soil Mapping (DSM) has advanced substantially over the past two decades, with predictive models becoming increasingly sophisticated. However, the methods and steps involved in constructing these models are not always fully documented and often rely on expert judgement, which can make modelling decisions difficult to reproduce or interpret. There is therefore a need to integrate DSM with modern technologies such as large language models (LLMs) to generate deeper insights and support more timely analytical outcomes. This paper introduces GeoGraphRAG Soil (GGRS), an open-source conceptual framework that automates and formalises DSM using graph-based retrieval-augmented generation and large language models.GGRS is built on a four-layer architecture: (1) a PostGIS spatial database for managing soil and environmental data; (2) a geospatial knowledge graph in ArangoDB encoding soil-environment relationships and geoprocessing operations; (3) a hybrid retrieval system in Milvus that identifies task-relevant subgraphs; and (4) a LangChain-based orchestration layer that translates natural language queries into executable geospatial workflows using open-source LLMs.GGRS advances DSM toward a more explicit, knowledge-driven approach that prioritises automation, interpretability, and methodological transparency by integrating structured domain knowledge with generative modelling techniques. The framework provides a cost-effective, open foundation to support soil health management in public sector initiatives and enables empirical evaluation in agricultural practice. GGRS can translate user queries into actionable, site-specific recommendations, such as identifying zones with low nitrogen and suggesting appropriate NPK fertilizer rates, predicting micronutrient deficiencies and recommending corrective crops, or assessing terrain and climate constraints to guide safe fertilizer application and crop selection. As an open-source reference architecture for researchers and policymakers, GGRS employs state-of-the-art technologies to support systematic workflow design and lays the foundation for future empirical validation and large-scale deployment by both public and private organisations.