With accelerating urbanization, accurately predicting intra-urban population mobility has become a fundamental requirement for urban planning and policy formulation. However, the adaptability and performance of existing mobility models across spatial scales remain unclear, and there is a lack of systematic evaluation frameworks that integrate spatial granularity, travel distance, and population heterogeneity. This study addresses these gaps by proposing a cross-scale comparative framework to evaluate three representative mobility models—the Gravity Model (GM), Radiation Model (RM), and Population-Weighted Opportunities Model (PWO)—under varying urban conditions.
We construct three groups of controlled experiments using high-resolution mobile phone data from Shanghai to assess model performance across spatial (grid size), distance, and population density scales. Furthermore, we apply multivariate analysis of variance (MANOVA) to decompose the relative contributions of different spatial factors to prediction errors.
The results demonstrate distinct scale sensitivities among the models. The GM model, grounded in Newtonian gravitational principles, shows high robustness over longer distances (>5 km), yet suffers from performance degradation under fine spatial granularity due to spatial heterogeneity. Its accuracy improves with population scale but decreases significantly when regional area disparities exceed a threshold—prediction performance drops by over 40% when grid size differences surpass 3 km. The RM model, based on the nearest-best-opportunity assumption, performs well for short-distance, origin-driven flows, such as commuting, but introduces systematic bias in small-scale contexts. Its sensitivity to origin population density makes it more suitable for high-density urban cores. The PWO model enhances RM by incorporating destination population weights, exhibiting superior compatibility with spatial heterogeneity in dense, polycentric cities. It performs best at short distances (<5 km) but loses effectiveness as travel distance increases.
MANOVA results confirm that GM is primarily influenced by population density and area scale, whereas RM and PWO are more sensitive to distance and destination-related factors. Based on these findings, we propose a model selection strategy tailored to mobility drivers: GM is recommended for long-distance prediction in spatially homogeneous regions, while PWO is preferred for short-range flows between small, densely populated areas. RM serves as a complementary model when origin-driven flows dominate.
This study not only clarifies the physical mechanisms underlying scale-dependent model performance but also offers an actionable decision-making framework for selecting appropriate models in different urban mobility scenarios. Future research can further improve predictive accuracy by developing hybrid models that combine strengths of multiple frameworks, integrating multi-source spatial data such as POIs and land use, and coupling traditional models with deep learning approaches to enhance non-linear pattern recognition while preserving interpretability. By uncovering the scale-sensitivity of mobility models, this work lays a theoretical and methodological foundation for multi-scenario mobility forecasting in smart city planning and fine-grained urban governance.