Scaling network-level transit OD matrices in sparse data: comparative analysis and proposed approach

Automated Fare Collection (AFC) systems provide detailed passenger origin–destination (OD) information; however, low usage rates necessitate scaling to represent the full transit population. Conventional approaches have notable limitations: Iterative Proportional Fitting (IPF) cannot exploit latent transfer information, while Itinerary Scaling Factors (ISF) produce unreliable OD estimates under low AFC penetration. This study proposes an Iterative Proportional Fitting with Transfer Ratios (IPF-TR) method, integrating route-level OD matrices with transfer information to generate robust network-level OD estimates under data sparsity. Two seed matrix types are examined: null seeds assigning uniform weights to feasible OD pairs, and AFC-derived seeds under varying penetration rates. Experiments on the Sioux Falls network show IPF-TR with null seeds is the most robust method under complete APC data, sustaining high accuracy at low penetration across travel-pattern and transfer-rate variations, whereas ISF becomes competitive only at high penetration or when APC data are incomplete. APC completeness and linked-trip representation emerge as key determinants of scaling accuracy.