1 PRE-PRINT for "Mathematical Methods on Optimization in Transportation Systems" – Kluwer Academic Publisher EFFECTS OF DATA ACCURACY IN AGGREGATE TRAVEL DEMAND MODELS CALIBRATION WITH TRAFFIC COUNTS Michele Ottomanelli Dept. of Highways and Transportation - Polytechnic of Bari (Italy - EU) tel: +39 080 5963380, fax: +39 080 5963329 m.ottomanelli@poliba.it Abstract This paper concerns with aggregate calibration of urban travel demand model parameters from traffic counts. A bi- level sequential Non-linear Generalised Least Square Estimator (NGLS) has been proposed to calibrate a travel demand model. The first aim was to find out the effects of the required input data accuracy assumptions on the model calibration. The second aim was to show the possibility to improve model link flows estimation performance even if the starting demand model was properly calibrated by using expensive disaggregate data. An experimental analysis was carried out on a real middle-sized town: the model was calibrated and validated under different “a priori” assumptions on data accuracy level of the starting data. The employed data were a traffic counts set and a maximum likelihood starting estimate of the travel demand model parameters. Key words: NGLS estimator, travel demand models, calibration, traffic counts Introduction In urban transportation planning activities it is often necessary to determine the link traffic flows on the road network. The reliability of the link flows estimate depends on the reliability of the Origin-Destination travel demand matrix (O-D matrix) estimate, as well as the supply and traffic assignment models. The O-D matrices in a given area could be estimated by using two different methodological approaches: the direct estimation method or the mathematical models estimation (indirect estimation). In both approaches very expensive surveys must be conducted in order to collect the necessary input data. Sometimes (e.g. in small urban areas, or when few resources are available), the great cost of the survey might compel the practitioner to conduct travel demand estimation using the resources and the poor available information by applying “very pragmatic” methods, which lack in theoretical consistence. In recent years researchers have made many efforts in order to propose effective methodologies which provide “better and better” travel demand estimates by using information cheap, easy and immediate to collect.