The Meliferea chatbot: Harnessing large language models to interact with geoanalytical services using natural human language

Geoanalytics provide valuable insights about the earth, its evolution and processes. Currently, the analysis and interpretation of geospatial data requires knowledge of complex software and methods, limiting the potential insights from geospatial science to only a small group of knowledgeable individuals. This paper explores the use of large language models (LLM), together with geoanalytical services, to enable natural language-based question answering related to geography. The paper examines the possibilities and limitations of LLMs to address human questions of a varying complexity. The results from a user-based evaluation indicate good performance in terms of completely correct answers, with accuracy exceeding 82%. However, as the complexity of questions increases, involving multiple geoanalytical services and locations together, the system struggles to provide accurate responses and the accuracy rates drop significantly. This research highlights the potential of LLMs to transform the usability of geoanalytical tools, lowering the barrier of using GIS software by beginners or everyday users.