Passively collected data have become a key input for urban travel demand analysis, enabling large-scale observation with high spatiotemporal resolution. In shared mobility systems, however, these data record only realized trips and omit suppressed attempts when vehicles are unavailable, leading to demand censoring and potentially biased inference. Although various methods have been proposed to impute unmet demand, their performance is difficult to validate because the ground-truth unmet demand is typically unobserved, and it remains unclear whether correcting for censoring is substantively necessary. This study addresses this gap by proposing a unified evaluation criterion and an empirical benchmarking framework to assess both the necessity and consequences of unmet-demand imputation. We apply five representative imputation approaches and compare the resulting imputed demand against the observed demand baseline in terms of temporal and spatial demand representation and the inferred effects of explanatory factors. Results show that censoring is non-random and strongly heterogeneous across time and space, with imputed increments concentrating in peak periods and high-demand areas. Moreover, demand imputation can materially alter the estimated significance of key determinants, indicating that relying on raw trip records may lead to systematically distorted conclusions. These findings suggest that unmet-demand imputation is not merely a data-enhancement step. It should be explicitly considered when drawing inference and making operational decisions from shared-mobility data, especially in settings where stockouts are frequent and spatially concentrated.
Feeder bus services are essential for extending the service coverage of urban rail transit systems, yet feeder-route detour and its supply–demand matching remain insufficiently understood at the city scale, particularly in terms of spatial heterogeneity, underlying mechanisms, and implications for feeder service design. This study develops an integrated framework for feeder-route detour identification, typology, and supply–demand diagnosis by incorporating bus network topology, multidimensional route indicators, deep embedded clustering, and spatially explicit mismatch analysis. A case study is conducted in Nanjing, China, using rail and bus network data, land-use data, and smart card data. The results reveal five distinct feeder-route types with different physical meanings. Supply–demand mismatch varies systematically across route types. Relatively direct routes are not always associated with strong demand, while highly detoured routes may reflect either genuine inefficiency or deliberate coverage-oriented service provision. High-demand routes are concentrated in and around the city center, whereas low-demand routes are more widely distributed across intermediate and peripheral areas. Feeder-route detour is not a uniform manifestation of poor design, but a spatially differentiated outcome shaped by urban form, barrier conditions, and service strategy. By moving beyond conventional directness measures toward typology-based and spatially explicit diagnosis, this study provides new evidence for understanding feeder bus detour and offers practical implications for feeder-route optimization, station-area accessibility improvement, and transit equity.
Public transit infrastructure serves as a critical determinant of urban equity, yet evaluations of transit disparity historically rely on macroscopic proximity metrics, overlooking the micro-scale quality of the transit node environment. To address this methodological gap, this study introduces a scalable, automated framework utilizing Vision-Language Models (VLMs) for high-resolution transit stop assessment. We develop the Bus Stop Boarding Environment Index (BSBEI), a multidimensional metric encompassing six environmental domains. Leveraging zero-shot visual instruction tuning, our framework autonomously audited approximately 115,000 multi-perspective street-view panoramas across five diverse metropolitan regions (Seoul, Wellington, Santa Clara, Chicago, and London). Our analysis reveals a pervasive bimodal distribution of infrastructure quality dominated by a low-score peak, suggesting that transit agencies across these distinct contexts default to a minimum viable operational paradigm that restricts premium amenities. Spatial stratification analysis utilizing Lorenz curves and Gini coefficients indicates significant disparities, particularly within the accessibility and information sub-dimensions. To elucidate the geographic and socioeconomic drivers of these inequities, we applied geospatial analysis and econometric models, including Local Indicators of Spatial Association (LISA), quantile regression and group comparison tests (ANOVA and Kruskal–Wallis). Correlating BSBEI scores with covariates, e.g., ridership intensity, latent demand, neighborhood deprivation, and urban typology, we identified a consistent positive association between environmental quality and actual ridership. Based on these drivers, we classify the transit systems into four distinct archetypes: demand-driven (Chicago), development-oriented (Seoul), spatially heterogeneous (Wellington and Santa Clara), and highly standardized (Central London). By successfully integrating VLMs with public transit evaluation, this research validates the feasibility of large-scale, automated micro-environment auditing, providing a robust empirical foundation for targeted policy interventions and the optimization of global transit equity.
As the logistics environment changes dramatically in cities around the world, various freight-related issues have emerged. To address these issues, responses from the public sector are becoming increasingly important, especially at the local government level. However, knowledge and experience in urban freight/city logistics have not been sufficiently accumulated at the local level, making it challenging to address local freight-related problems. In this paper, we present an urban freight typology based on data from 688 local government units across seven metropolitan areas in Japan. Unlike previous typology studies that primarily rely on fine spatial units and estimated freight demand indicators, the proposed framework adopts a municipality-level perspective and incorporates observed freight demand data alongside socio-economic, land-use, and transport system characteristics. Using factor analysis and an agglomerative clustering approach, we identify seven urban freight factors and nine clusters. The results confirm that each cluster effectively captures freight characteristics specific to local governments. Furthermore, to demonstrate the practicability of the proposed typology, we analyze the relationship between clusters and vehicle stop frequencies, and, using the case of the Kanto metropolitan area, identify clusters preferred for locating extremely large logistics facilities. Statistical validation showed that the typology explains substantial variation in vehicle stop density and the spatial distribution of large-scale logistics facilities, supporting its usefulness for freight planning and policy analysis.
