Building Base Matrices from Synthetic and Observed Data: a New Model Structure

Building Base Matrices from Synthetic and Observed Data: a New Model Structure


G Terzis, S Marsh, Jacobs Consultancy, UK; J Bates, John Bates Services, UK; M Logie, Minnerva, UK; H Neffendorf, Katalysis, UK


Building base matrices by statisically combining synthetic and observed data and maintaining consistency between P/As and O-Ds


Incremental demand models pivot off base matrices of travel: they are widely used, and they are generally recommended in the UK Department for Transport?s [DfT] guidance (WebTAG). However, there is little guidance as to how these matrices can be constructed. This paper focuses upon our work for building base matrices using a combination of synthetic and observed data (e.g.: roadside, public transport and household interviews as well as counts). The work has been carried out to support the applications by the Tyne and Wear (T&W) Authorities and, separately, Durham County Council, for Transport Innovation Fund (TIF) investment ? a DfT procedure.

The construction of base matrices using a variety of data sources is a significant problem. There are a number of packages of proprietary software, which allow the estimation of ?matrices from counts?, but these are largely highway-oriented and generally only apply to the construction of Origin-Destination (O-D) matrices. These are suitable for highway assignment purposes, but they are not suitable for demand modelling purposes, where it is essential that the matrices are held on a Production / Attraction (P/A) basis. It is the integration of methods for producing P/A and O-D matrices that are self-consistent and which draw on synthesized and observed sources of information, which provides the distinctive aspect of the work that is described.

While a base year matrix is produced more or less as a default when a conventional ?four-stage? model is implemented, it may be questioned to what extent it is scrutinised in terms of its general acceptability for describing the base situation. The historical lack of emphasis on demand modelling as opposed to supply-side modelling (networks) has led to practical attempts to build matrices being largely data driven.

However, since trip end models are one of the most stable elements in the modelling field, it is vital that they be used. By contrast, observed data may condition more strongly the distribution of trips, which is synthesized less accurately. The resulting base year matrix can therefore be shown to match with the best sources of available information, whether synthesized (typically based on large national data samples) or local surveys.

Given the difficulties of the base year fit for the P/A approach, the most common method is to use the P/A demand model merely to predict the change in demand (this is a type of ?incremental? method). After translating this to an O-D basis, it is then used to adjust a validated O-D matrix which is then assigned to the network. Effectively, this accepts an inconsistency between the base P/A matrix and the validated O-D matrix, resulting in a major weakness in its use for forecasting.

Our challenge was therefore to develop a practical methodology that:

? statistically combines synthetic and observed data to produce P/A matrices;
? allows the adjustments suggested by the count data to be conveyed not only to the O-D matrix used for assignment, but to the underlying P/A matrix.

This removes the inconsistency between P/A and O-D base matrices. If the base P/A matrix translates into an O-D matrix which validates appropriately at the network level, then any changes predicted at a P/A level by the demand model can be directly translated into O-D form for assignment purposes and properly reflected in forecasts.

(At the time of submitting this abstract we had results appropriately checked from our T&W study and we expected results from our Durham study by March 2007.)


Association for European Transport