Activity-based models (ABM) are designed to produce accurate results for project analyses. ABMs can provide improved sensitivity to transportation system changes, policy testing, and traveler characteristics over that of a conventional model since they model interactions among a wide range of variables and traveler characteristics. An ABM can be used to conduct the same level of analysis as a trip-based model, but can also provide more information about the effects of more sophisticated policies such as reversible HOT lanes, environmental justice and Title 6 analysis, land use impacts and sustainable livable centers, as well as transportation policy questions, on a variety of population segments.
While the provision of forecasting capabilities such as those outlined above is, in itself, a worthwhile goal, it is critically important to demonstrate that the added model sensitivities are, in fact, reasonable and valid. Moreover, ABMs consist of a number of interrelated, discrete choice modeling components. That is, the upstream model components (long-term choice models) provide inputs to the person-level models (daily activity, intra-household interactions) which, in turn, provide inputs to tour-level models (destination, mode, time of day) and, finally, trip-level models. Due to the number of model components, the complexity of the various models, and the amount of information passed between models, there is a high chance of error propagation if all of the components are not properly calibrated and validated. A rigorous calibration process is necessary for various levels – geographic (regional, sub-regional, county, or district), market segments (income, trip purpose), and household structures and person-types (gender, full time worker, etc).
This paper describes the calibration/validation process used to explicitly test the reasonableness of the added model sensitivities and calibrate/validate all of the specific model components to minimize compensating modeling errors and model adjustments due to cascading impacts of model specification or calibration errors. This paper focuses on a comprehensive methodology that was laid out for the Houston-Galveston Area Council (H-GAC) ABM development effort. This paper also presents example calibration results from the H-GAC ABM validation effort for several ABM components.