GeoScout: A Conversational Agentic GeoAI Framework for Spatial Decision Support in Methane Leak Detection and Repair
Jinya Wang, Mingke Li, Steve Liang
Field-inspection programs are inherently spatial because site locations, spatial arrangement, and crew movement shape what can be inspected and how quickly problems can be addressed. Recent studies increasingly combine large language models (LLMs) with simulation systems to lower the technical barrier to configuring and running complex models through natural-language interaction. However, existing work has largely emphasized scenario generation and execution, while result interpretation is rarely grounded in explicit spatial computation, limiting its use for spatial field operations such as methane Leak Detection and Repair (LDAR). We present GeoScout, a conversational agentic GeoAI framework that extends LLM-assisted simulation into spatial decision support. GeoScout configures two simulation engines with different geographic representations from a single plain-language program specification and compares their outcomes. Holding the program fixed while varying only facility layout increases modeled per-transition crew travel roughly fourfold in the distance-aware engine while leaving the geography-agnostic engine unchanged. GeoScout also grounds post-simulation interpretation by routing spatial questions to deterministic spatial (GIS) tools and retaining tool traces so that spatial claims remain traceable to their computations. We apply GeoScout to methane LDAR equivalency under Alberta’s Directive 060. These capabilities provide an assumption-transparent and auditable framework for agentic geospatial simulation and spatial decision support.