Stepwise Model Calibration/Validation

Stepwise model calibration/validation is a structured process by which travel models are calibrated to match base year conditions in a sequential manner. As each model is calibrated, the results are passed as input to the subsequent step in the modeling process. In this way, the impacts of error propagation are reduced throughout the modeling process, thus reducing the need to make wholesale changes to match, for example, total regionwide vehicle-miles of travel (VMT).

For example, it is quite common for trip-based models newly developed from travel survey data to produce 85-90 percent of the regional VMT estimated from observed traffic volumes when they are first applied. An all to common approach for addressing the shortfall in VMT is to assume that trips have been under-reported in the travel survey and increase trip rates upward to account for the VMT shortfall. Using a stepwise model calibration/validation process, the inputs to the modeling process (e.g. socioeconomic data and interchange travel times) are first validated against observed data. The validated socioeconomic data would then be used as input to the trip generation model. The results of the trip generation model would then be compared to observed data such as expanded household survey data or Census Transportation Planning Package (CTPP) data. The comparisons could be performed for trips made by various socioeconomic groups (e.g. households by household size or by auto ownership level) and by geographic areas as well as for overall regional trips. Once the trip generation results reasonably match the observed data for the various strata, they are used as input to the subsequent modeling step, typically trip distribution.

This straightforward process has been used for many trip based model development projects. The process can, and should be, adapted for use in activity-based models that include numerous model components. The stepwise model calibration/validation process may be more time consuming due to the number of model components included in most activity-based models. In addition, the implementation approach affects the ease with which the stepwise calibration/validation process is implemented. Some activity-based models are "vertically" implemented by applying all model components for a sample household prior to modeling the next household in the region and some are "horizontally" implemented by applying a model component for all households in the region before proceeding to the subsequent model component. The stepwise model calibration/validation process may be more easily and efficiently applied for the horizontally implemented models than for the vertically implemented models.

# References