FREIGHT DEMAND MODEL ESTIMATION FROM TRAFFIC COUNTS

Abstract

Abstract: The estimation of models of freight demand is an expensive and time consuming undertaking. This is particularly difficult for road haulage where shippers and transport firms are widely dispersed thus making data collection difficult even for very modest sampling ratios. A number of model forms have been suggested in the past to represent demand for road haulage; these range from simple aggregate models to more disaggregate approaches. Although the latter offer the promise of closer representation of the underlying factors in road haulage decisions, aggregate models are often preferred for forecasting purposes. The authors have developed a family of aggregate models of freight movements which can be calibrated from traffic counts and other low cost data. Three model types are examined, a gravity (GR), an opportunity (OP), and a gravity-opportunity (GO) model. There are reasons for this variety as each may be more appropriate for particular conditions. Three different methods were developed to calibrate these models from low cost data: non-linear-least-squares (NLLS), weighted-non-linear-least-squares (WNLLS), and maximum-likelihood (ML1, ML2) methods. The models and these calibration methods have been implemented in a micro-computer package capable of dealing with up to ten commodities simultaneously. The approach has been tested using the 1982 Freight Movement Survey in Bali (Indonesia). The models were found to provide a reasonably good fit when dealing with five commodity types. General conclusions regarding the applicability of the approach to other environments and its potential for transport demand forecasting and planning are given at the end of the paper.