Development of Methods To Estimate Unobserved Link Flows and Their Application to the Amsterdam Orbital Motorway

To monitor highway network performance, environmental impacts, and congestion levels, highway authorities maintain statistics on traffic volumes. A typical aggregation level of such statistics is 1 h, and a distinction is made between month and type of day. Traffic volumes usually are observed directly for only part of the network. For the remaining part of the network, these time-dependent traffic volumes need to be estimated using time-dependent traffic data and additional static information, such as trip tables. Six estimation methods are discussed. These methods are characterized by the selection of data used, the assumptions on interdependencies between origin-destination (O-D) demands they represent, and their estimation mechanisms. A methodological novelty is in the modeling of the correlation between errors contained in the O-D cells of the prior matrix in case of common departure period, origin, or destination. All methods were verified and analyzed using synthetic data. Experiments with empirical data from the Amsterdam Orbital Motorway show that, for all methods, bias is the main source of error. This problem can be remedied by collecting and using appropriately aggregated data. Experiments also show improved estimation results for methods that take into account a correlated error structure for the prior O-D matrix.