Abstract
Recent advancements in Generative AI offer promising capabilities for spatialanalysis. Despite their potential, the integration of generative AI withestablished GIS platforms remains underexplored. In this study, we propose aframework for integrating LLMs directly into existing GIS platforms, using QGISas an example. Our approach leverages the reasoning and programmingcapabilities of LLMs to autonomously generate spatial analysis workflows andcode through an informed agent that has comprehensive documentation of key GIStools and parameters. The implementation of this framework resulted in thedevelopment of a "GIS Copilot" that allows GIS users to interact with QGISusing natural language commands for spatial analysis. The GIS Copilot wasevaluated based on three complexity levels: basic tasks that require one GIStool and typically involve one data layer to perform simple operations;intermediate tasks involving multi-step processes with multiple tools, guidedby user instructions; and advanced tasks which involve multi-step processesthat require multiple tools but not guided by user instructions, necessitatingthe agent to independently decide on and executes the necessary steps. Theevaluation reveals that the GIS Copilot demonstrates strong potential inautomating foundational GIS operations, with a high success rate in toolselection and code generation for basic and intermediate tasks, whilechallenges remain in achieving full autonomy for more complex tasks. This studycontributes to the emerging vision of Autonomous GIS, providing a pathway fornon-experts to engage with geospatial analysis with minimal prior expertise.While full autonomy is yet to be achieved, the GIS Copilot demonstratessignificant potential for simplifying GIS workflows and enhancingdecision-making processes.