Abstract. In the context of rapidly evolving geospatial technologies, this study provides a comprehensive review of GIS model integration and service composition, emphasizing their critical roles in enhancing spatial analysis accuracy, decision-making efficiency, and cross-domain interoperability. Model ensemble techniques, rooted in machine learning and data mining, address limitations of single models by combining predictions from multiple base learners, thereby improving robustness and reducing overfitting. GIS model integration involves combining diverse spatial algorithms—such as buffer analysis, network analysis, spatial regression, and machine learning models—to tackle multifaceted geographic challenges. Key algorithms are systematically integrated to optimize outcomes in urban planning, disaster management, and precision agriculture. For instance, land-use change prediction synthesizes spatial regression, machine learning, and remote sensing, while natural disaster systems merge meteorological models with post-disaster assessments. The fusion of industry-specific models with GIS enhances location-based decision support by embedding spatial variables into domain workflows. Cloud-native architectures and AI-driven automation emerge as pivotal trends, offering scalable, real-time GIS solutions via platforms like serverless computing and SaaS. These innovations promise self-learning agents capable of automated spatial pattern recognition, real-time alerts, and optimized resource allocation. Despite progress, challenges persist in model selection, interpretability, and robustness. Future research directions emphasize large language model (LLM)-powered agents for intelligent geospatial processing, cloud-GIS hybrid platforms for elastic resource management, and industry-tailored SaaS solutions. By bridging traditional GIS tools with intelligent service ecosystems, this evolution aims to drive digital transformation, enhance cross-sector competitiveness, and unlock new potentials in spatial decision-making.
Ding et al. (Tue,) studied this question.