Generative AI-Driven Spatial Data Extraction in OpenStreetMap using Natural Language

Abstract

With the rising availability and support of geospatial data and tools, geospatial data analysis is increasing rapidly. However, geospatial data is challenging to extract and understand by individuals with limited or no prior knowledge of handling such data. This study presents a novel platform that integrates generative artificial intelligence (Gen. AI) and prompt engineering techniques for geospatial data retrieval and analysis. This is achieved by firing natural language queries and integrating a generative pre-trained transformer GPT-3.5 for data retrieval and analysis. The platform translates unstructured natural language inputs into structured Overpass API queries, retrieving detailed geospatial data from OpenStreetMap (OSM). The system streamlines the process, from query to visualization, enabling users without technical geospatial expertise to access spatial information seamlessly. It supports geospatial data retrieval tasks such as Point of Interest (PoI) extraction, proximity queries, and attribute-based retrieval. The experimental results show that the proposed approach outperforms existing tools such as Google Earth Engine (GEE), GeoGPT, GeoInsight, MapQA, OSM-GPT with an average query-execution time of 17.3 seconds and an average accuracy of 95%. It shows a significant improvement in usability over manual Overpass query construction. The proposed framework achieves higher performance while maintaining a lightweight design that does not require model fine-tuning or external training data. Unlike existing tools that heavily rely on fine-tuned transformers with tightly coupled components, the proposed framework is modular, prompt-driven, and API-based, which enables its rapid deployment and minimal resource usage. This lightweight architecture helps to improve system maintainability, scalability, and makes it easily accessible for real-time applications and end-users with limited technical infrastructure. Overall, the framework offers a scalable, accessible, and extensible solution for spatial data querying in open-source GIS workflows. This study can transform conventional geospatial data analysis practices into a more inclusive and user-friendly approach that features a geointelligent environment.