Reconstruction of OD demand using mobile phone data: A regression model with spatially varying coefficients

The spatial estimation of origin–destination (OD) flows is essential for understanding urban mobility patterns and supporting transportation planning. In practice, Household Travel Surveys (HTS) are widely regarded as reliable benchmarks of urban travel demand. However, a known limitation of HTS-based OD data is their limited spatial resolution and infrequent updates. In contrast, passive mobility data, such as Call Detail Records (CDR), offer extensive spatial coverage but reflect only an anonymized subset of the population. To address this gap, this study proposes a graph-regularized Spatially Varying Coefficient (SVC) linear regression framework, termed LR-SVC, to reconstruct HTS-based OD demand from passive CDR-based mobility flows by incorporating zonal geospatial attributes, including socio-demographic and land-use characteristics. The model incorporates spatial regularization to capture spatially heterogeneous relationships by ensuring that coefficient estimates vary smoothly across neighboring zones. The proposed model is then applied to two Canadian cities - Montreal and Ottawa. The results reveal several key findings: (1) Integrating CDR-based flows with geospatial features within a SVC framework enables zone-specific calibration effects, allowing the relative influence of passive mobility signals and geospatial attributes to vary across space. (2) The proposed LR-SVC model consistently outperforms traditional linear regression and machine learning models, achieving up to 9%–17% higher reconstruction accuracy while maintaining structural interpretability. (3) The estimated coefficients reveal a compensatory calibration mechanism, whereby geospatial effects intensify in zones where passive CDR-based mobility flows are relatively weak, highlighting how the framework structurally balances heterogeneous data reliability across space.