The Swiss federal government has asked the IVT, ETH Zürich in collaboration with the TU Dresden and Emch+Berger, Zürich to estimate origin-destination matrices by mode and purpose for the year 2000. The zoning system employing about 3’000 zones of very uneven size required a solution algorithm which is fast, but also able to model generation, distribution and mode choice simultaneously, while addressing the different data availability for traffic within, destined for and passing through the country. The EVA algorithm developed by Lohse (1997) was adapted for this purpose. The key proper-ties of the algorithm are its disaggregate description of demand, its use of appropriate logit-type models for the demand distribution, while maintaining the known marginal distributions of the matrices generated. This last point is of particular importance in a large scale planning applica-tion such as the one at hand. The algorithm calculates trip production and attractions by zone using activity pairs. The 17 ac-tivity pairs distinguished are the combinations of two activities, such as home-work or work-leisure. The relevant daily rates are derived for each of the 17 activity pairs from the 2000 Swiss National Travel Survey (Bundesamt für Statistik and Bundesamt für Raumentwicklung, 2001). The zonal attractivity is defined separately for each trip purpose. In addition to the common variables, such as employment or population, detailed descriptions of education places, shop-ping or leisure facilities, overnight accommodations, shopping centres etc. are employed (see Tschopp, Keller and Axhausen, 2003 for the data). The combined destination and mode choice models estimated for the different traveller types and activity pairs are based on the Swiss National Travel survey (RP data), but incorporates re-sults from a prior SP study on mode and route choice (Vrtic and Axhausen, 2004). The different zone sizes and the different levels of data available required the formulation of new additional models for the transit traffic passing through Switzerland and the traffic originat-ing outside, respectively leaving the country The matching network models for public transport and road traffic were implemented using VISUM 9.0 of PTV AG, Karlsruhe. The timetable based assignment considers all scheduled train services plus the relevant interurban bus services, in particular in rural areas. The paper has three main parts: the first main part derives and describes for the first time the EVA algorithm in English, including the solution method used. The second part summarizes the results of choice model estimation using the generalised cost elasticities of demand by purpose and traveller type. The third part assesses the quality of the results. These assessments are based on two independently derived matrices, which are available for rail-travel from on board - counts and for commuters from the 2000 national census. In addition, we compare the assign-ment results with the available cross section counts. The conclusions discuss computing times, accuracy and issues for further research.
