Planning and policy initiatives are dependent on travel demand models, such as activity-based models (ABMs). Characterized by its activity-based platform and disaggregate micro-simulation modeling techniques, ABMs tend to have greater spatial, temporal, and demographic details compared with the 4-step model. Increased spatial, temporal and demographic details are believed to improve behavioral realism and model accuracy. However, these details also increase model complexity, longer run time, larger data storage, and increased hardware and software requirements. This paper focuses on the spatial dimension, and attempts to provide insights to the question: how much more spatial detail is enough?
First, the paper describes the multiple layers of geographies used in the SANDAG ABM, including a 4996 TAZ system for representation of highway skims, and 23002 micro-zones for representation of trip demands, transit and non-motorized accessibilities, and land-use data.
Second, this paper describes the construction of two spatial resolution scenarios. In the first scenario, we keep the micro-zone as is at 23002, while in the second scenario we combine some micro-zones to reduce the number of micro-zones by half. Then this paper describes the reconstruction of land use, household, and population data as they are all micro-zone specific. The last part of this section describes network changes, such as the reconstruction of zone connectors. We keep all other model inputs and parameters constant to isolate the impact of spatial resolution changes.
Third, this paper describes and compares the results from the two scenarios. The first section describes the statistical analysis on accessibilities, mode shares, trip/tour length distributions, walk access/egress to transit trip length distributions, vehicle miles traveled (VMT), regional greenhouse gas (GHG) emission estimates, and physical activities spent on walk, bike and walk to transit trips. The second section describes model run time, data storage, and performance impact caused by spatial resolution change.
Finally, this paper discusses the findings of the impact of spatial resolutions on model results. The authors also attempt to answer the question: how much more spatial detail is enough, by recommending a balance between spatial details and increased model complexity without jeopardizing model quality.