An Application of Backward Forecasting for the Validation of an Urban Travel Model System

An Application of Backward Forecasting for the Validation of an Urban Travel Model System


J Kurri, Helsinki University of Technology, FI


The paper presents the methods used, and the results of an experimental external validation of the travel demand model system for the Helsinki metropolitan area. The validation is based on backward forecasting from 2000 to 1988.


Long-term forecasts are routinely used in strategic decision-making about urban transport policies and infrastructure. The plausibility of travel or traffic and patronage forecasts requires that the forecasts are validated. The paper presents the methods used, and the results of an experimental external validation based on backward forecasting. Data from two independent travel surveys in the Helsinki metropolitan area were utilised. The models were estimated with the sample data from a base year of 2000 in which a travel survey with more than 8,000 respondents was carried out. A similar telephone interview survey was conducted twelve years earlier in the autumn of 1988. The travel demand models estimated with the 2000 data were applied to the 1988 forecasted data, and the results were then compared with the 1988 travel survey data. Reported trips and trips predicted by the model system were compared according to a number of criteria including the distribution of trip distances, modal split by trip distance, and trip distribution with four aggregated zones. The forecasts were produced with a sample enumeration technique.

The paper is based on a research study carried out in 2003 for the Helsinki Metropolitan Area Council (YTV). That study had two main aims: a) to further develop the travel demand model system for the metropolitan area towards fully use of disaggregated data; b) to assess the validity of the travel forecasts developed in the study. The paper includes a short description of the model system as well as further development of the system. The modelling and forecasting system is based on a trip-based four-stage model approach with simultaneously estimated nested logit models for trip distribution and modal split. The maximum likelihood estimates of the alternative specific constants of the mode choice models are corrected for differences in the number of predicted and observed trips in the base year. This procedure is necessary for a fair comparison between the numbers of trips forecasted by the model system, and those calculated on the basis of the trips made by the persons sampled in the travel survey. Trip generation is based on average trip production rates calculated separately for some individual-level subgroups (age class, employment status, availability of car).

The study showed that there are much to be done in the utilisation of validation techniques in the development of travel demand models. From a methodological point of view, there were some difficulties in proper generation of population for a forecast year, as well as in maximum likelihood estimation of combined mode and destination choice models aimed at accurate forecasts both in terms of modal split and trip distribution. The sample representing the population of the forecast year of 1988 was generated by copying the socio-economic characteristics of the persons interviewed in the travel survey in 2000, and by calculating new weighting factors based on census data and population forecasts. In addition, the numbers of cars in the households were corrected so that the total number of cars coincides with vehicle ownership forecasts.


Association for European Transport