The Strategic Flemish Freight Model at the Intersection of Policy Issues and the Available Data
Stefan Grebe, Significance, Gerard De Jong, Significance & ITS Leeds, Michiel De Bok, Significance
A strategic freight model forecast the demand of freight transport. It should predict the effects of policy measures affecting transportation costs and times of the modes. We discuss the challenges to build such a model given the available data.
The Flemish authorities use a strategic freight model to forecast the demand for freight transport in the future and to support the decision making process for large infrastructure investments. An important requirement is the ability of the model to forecast effects of policy measures affecting transportation costs and times of the modes.
These policies requirements on the one hand and the available data on the other hand put demands on the model (and the modeler) that may sometimes be hard to reconcile. In the case of the Flemish model the input data comes from multiple sources with varying precision, vehicle type information, commodity classifications and geographic aggregation levels. In the paper we will discuss different strategies to combine and process the input data. We evaluate them by presenting the direct effects and the impact on the model coefficients and elasticities. In addition we compare the used methods to the methods used in the aggregate-disaggregate-aggregate (ADA) freight models of Scandinavia and the Dutch BasGoed model.
The current version of the strategic Flemish freight model is a classical four-step traffic model with several additions such as a time-period choice model and the use of logistical hubs (by mode).
The model starts with the application of production and attraction multipliers on socio-economic data (for future years these are forecasts themselves) for each zone. The model uses 518 zones within Belgium and 96 external zones in Europe. Given the productions and attractions per zone, the distribution is modelled by using a gravity model.
Mode choice and vehicle type choice are integrated in one nested logit model and are estimated simultaneously. The mode and vehicle-type choice part of the model distinguishes commodity groups and considers 3 road, 3 rail and 12 inland waterways vehicle/vessel types as direct and intermodal transport modes. Skims of the road, rail and inland waterway networks have been used to estimate distances and travel times. The cost functions used, include transport time dependent cost, transport distance dependent cost, toll fees, resting periods, as well as costs for loading, unloading and transhipment. The OD matrices for the base year 2010 have been constructed on the basis of data from available transport statistics. The methods used (as described below) and the effect of different methods on the results of the model are the main topic of this paper.
For road transport the transport volumes are available on NUTS 3 level. They have been distributed with a Furness approach to the 614 zones of the model. In addition, the vehicle type for road transport is modelled with a deterministic model using trip frequencies to match the national vehicle type split. We discuss both the effect of the zone sizes and the effect of the trip frequencies on the results.
For rail and inland waterway transport, information on vehicle type is partly available. However, different commodity classifications are used and the data sources are not covering the whole study area. To be able to estimate a model, the conversion to a single commodity classification is necessary. A challenge in this process is to avoid biases on the sensitivity of the model to policy measures. We will discuss different approaches to handle small OD flows including bucket rounding and random draws.
The paper focuses on the model for vehicle type and mode choice and on the data that are used in this part. The key question is how different ways of processing the raw data affect the model estimates and therefore also the model elasticities.
Association for European Transport