The performance of current models with respect to forecasting freight transport demand is reviewed. The original motivation was to ask how freight modelling deals with discrepancies between observed model outcomes as model assumptions and forecasting results. In this context a lot of energy seems to be invested into the calibration process. The conventional approach for improving models is updating prior estimates through new information, thus an update in the level of the same kind of information. In the centre of the discussion the author places the idea of including into the modelling process an expression for an a priori unknown kind of information. The knowledge about both, the information of a new level and of a new kind, is acquired through the modelling process and then used in the calibration process. Consequently, the definition of the model assumptions must not only include a specification of everything that is true for the model (necessary assumptions) but also an exclusion of everything that is untrue (sufficient assumption). In current freight modelling scope for improvement has been identified in the area of strengthening the link to policy analysis. Freight modelling as a decision-making tool is lacking mechanisms to monitor the effect of policy measures. So far, the possibility of using freight demand models for policy simulation has not been sufficiently exploited. This is especially true for environmental targets such as stated in the Kyoto Protocol which hardly find any recognition in national freight models. In contrast to previous modelling approaches the proposed model includes a feedback mechanism linking emission estimates which are derived from freight transport forecasts back to transport policy design. The model consists of three modules, which are run under various policy scenarios, a freight demand module (LEFT), a supply chain design module (CAST) and an emissions module: The LEeds Freight Transport Model (LEFT) uses logit models to provide estimates of the effect of macroeconomically neutral scenarios on modal split (road, trainload and wagonload), average length of haul and total market size. The LEFT model is complemented with the strategic network optimization model CAST which is applied by shippers in the private sector for strategic design of their supply chains. In the context of this research CAST is used to model a national distribution network for the UK based on one-year data. The first two modules are used to deliver estimates on freight traffic activity on the basis of which emissions from a set of six air pollutants and greenhouse gases are calculated for the base year.In a next step, the estimated emissions are compared to the environmental target set by the government. Hence, the first iteration of the model attempts to predict the present. This means that the observed data and the model results refer to the same time span. The gap between the observed model outcomes in terms of actual emissions from freight transport can be closed by applying an appropriate set of transport policy measures, such as fuel tax. The objective of the first iteration is to determine the level of these measures. The obtained information is then used to update the original model assumptions. In the second iteration the model is then run under the new assumptions. In summary, the paper would like to contribute to a generic discussion on testing assumptions in freight demand modelling and increasing the accuracy of model forecasts. The underlying question concerns the gap between model observations and the model results. To reconcile the discrepancies a modelling framework is suggested which contains a feedback mechanism for updating information which may include information of a new kind and level. The example of using freight demand modelling for transport policy and the environment gives a flavour of a possible application in freight transport practice. For the covering abstract see ITRD E137145.
Abstract