抄録
This study addresses a fundamental question in smart city geography: how can multisource spatiotemporal data and artificial intelligence (AI) be transformed from tools for spatial representation and prediction into a coherent framework for dynamic spatial cognition and mechanism-oriented inference? Through a systematic synthesis of the field’s evolution from statistical analysis, spatial econometrics, and GIS-based decision-making to GeoAI-enabled spatial intelligence, the study organizes major advances in multisource sensing, deep learning, spatiotemporal knowledge graphs, causal inference, complex networks, reinforcement learning, and digital twins along a continuous “sensing–representation–inference–decision” pathway. A Cyber–Physical–Social Systems perspective is further used to conceptualize the coupling of physical, informational, and social spaces, from which a four-layer framework is developed encompassing multisource urban sensing, spatiotemporal relational cognition, intelligent scenario simulation, and collaborative governance. The synthesis identifies cross-scale effects and the Modifiable Areal Unit Problem, spatial confounding and causal identification, geographic heterogeneity and transferability, model interpretability, and spatial equity as major methodological constraints on the transition from predictive performance to decision credibility. Emerging priorities include mechanism-constrained GeoAI, cross-scale causal inference, digital-twin-based scenario analysis, human–AI collaborative planning, and trustworthy spatial intelligence. The study’s principal contribution is to recast smart city geography as an integrated spatial-intelligence framework that couples multisource observation with geographic knowledge, intelligent inference, and accountable governance, providing a theoretical and methodological basis for advancing urban analysis from static description and prediction toward dynamic, interpretable, and mechanism-oriented understanding.
