Berlin II: CEMDAP-MATSim-Cadyts Scenario

To correctly model initial demand properties not included in MATSim iterations in speci c studies (i.e., activity choice), suitable data are needed. Travel diaries containing departure times, mode choice decisions and activity locations are widely used. However, much of this data source content, particularly location information, is considered sensitive in terms of data privacy legislation and thus increasingly di cult to obtain and process in many areas (e.g., in Germany and the United States; Ziemke et al., 2015). The Berlin II scenario (also referred to as the CEMDAP-MATSim-Cadyts scenario according to the applied models in its setup), is the outcome of an alternative approach relying exclusively on freely available or easy-to-obtain input data. All of these data do not rely on individual trajectories, but instead on “anonymous” data that is aggregated so much that the data providers are no longer concerned about privacy issues. The starting point for this scenario is a publicly available commuting matrix containing homes and workplaces of workers with social security on the municipality level. Based on this information, it is possible to model morning and evening commuting peaks. To obtain a full-population demand representation, two further major modeling steps are required. First, in cases like the Berlin case, see below, where the commuter matrix spatial resolution is quite coarse, higher resolution O-D information is necessary. Second, a procedure is needed to model secondary activities, i.e., all other activities beyond home and work. The importance of the rst step becomes obvious when looking at the German case; here, the whole city of Berlin, with 3.4 million inhabitants, is represented by exactly one zone (Bundesagentur für Arbeit, 2010). In the United States, commuting matrices are typically available only on a county-to-county level. Since such location-aggregation-based matrices may become the rule, rather than the exception, in privacy-sensitive societies, a (generalizable) method to attain O-D information at a higher resolution is needed (Ziemke et al., 2015). The standard solution would be to estimate an activity location choice model. This, however, is di cult if no trip data to estimate