This release has a zoning system on "Gemeindeverband"-level, counting 4602 zones. Transport networks are refined accordingly. The road network now also includes tertiary roads. Unless there is 4 TB RAM available, the resulting size of the OD table requires OD sampling. Each set of scenarios has a consistent OD set (inherited through the scenario hierarchy) that is sampled among major population centres, random zones, and a set of scenario-relevant zones specified in parameters.xls. Zone clustering and NUTS3-level modelling is still possible, but discuraged due to lower accuracy.
There are also two more demand segments than before, as the purpose "buy/execute" (formerly included for compatibility with "Verkehrsverflechtungsprognose 2030") is now split into "shopping" and "errands". Travel purposes now fully align with "Mobilität in Deutschland". All demand models have been re-estimated and refined accordingly. Compulsory trip purposes are distributed with a doubly constrained assignment, including inner/inter-zonal distinction. It performs well based on employment data and OSM-sourced education data. The destination choice model for non-compulsory trips performs slightly worse, but is required to connect volumes to POI data from OSM. Though, it is not possible to estimate a valid inner/inter-zonal choice model, even after many trials (it would also lack logic). The generation choice model, too, is logically not fully sound, which is why exogenous trip generation from MiD data is recommended (though it does not depict elasticity of demand).
Additionally, except some bugfixes and documentation clarifications, the following changes were applied:
quetzal_germanynow uses the libraryquetzal-lite, which is easy to install and compatible with new versions of pandas and other dependencies- Generation volumes and inner/inter-zonal shares are exogenously computed from MiD on RegioStaR7-level
- Volumes for compulsory and non-compulsory trips are saved in different files, making the model structure more modular
- Distances in choice models are now euclidian distances rather than road network distances, making the model less error-prone
- The composite cost file is now in the
input_staticfolder