GeoAI research in transition: thematic structures, temporal evolution and method–domain linkages

Geospatial artificial intelligence (GeoAI) has emerged at the intersection of artificial intelligence and geospatial science, advancing applications in environmental monitoring, urban analytics and methodological innovation. Existing reviews are limited: bibliometric analyses rely on metadata such as keywords and citations, while narrative syntheses focus on narrow subdomains. These approaches provide little semantic depth, capture limited temporal dynamics and seldom link computational methods to application domains. This study applies full-text structural topic modeling (STM) to over 100 peer-reviewed publications, enabling semantic-level theme extraction, temporal trend analysis and systematic mapping of methods to domains. Findings reveal three main insights. First, GeoAI research clusters into three strands: environmental applications (soil, groundwater, floods, air quality), urban and human-centered analytics (vegetation, walkability, safety) and methodological frontiers (foundation models, generative AI, graph neural networks, explainable AI). Second, temporal analysis shows a transition from early emphases on remote sensing and infrastructures toward socially embedded, sustainability-oriented themes, including climate resilience and public health. Third, method–domain mappings indicate both specialization (e.g. CNNs in imagery, ensembles in environmental modeling) and diffusion across domains (e.g. transformers, explainable AI). Overall, this study delivers the first full-text STM-based synthesis of GeoAI, offering a roadmap for aligning AI innovation with sustainability and urban governance challenges.