The normal course of model building is to use land use, and trip generation rates to determine peak hour origins and destinations and then to distribute and assign them. However, after all efforts to calibrate some models using that method, it will be seen that either 1) the land use - trip generation rate combinations were incorrect, 2) the gravity model was unable to predict local driver behaviors or 3) the counts available were not adequate, either in their directional split or in the collection time with respect to the arbitrary date and hour chosen as the model's "snapshot in time."

In an effort to overcome this inadequacy, a method may be employed to build a "mask" which then acts as a corrective factor for each interaction between all possible places--that is, for each cell in the trip table.

In order too build this mask, the modeler should take whatever steps are necessary to arrive at a better trip table than the one created through the normal distribution process. This may be accomplished by applying "Willumsen's Method," a mathematical tool which establishes a probable trip table for all zones based on existing traffic counts. This method is available in TMODEL2 as well as in the THE (The Highway Emulator) program, which is nationally available.

While using Willumsen's Method does not typically provide a proper method for growing land uses for future scenarios, it can be useful for diagnosing problem zonal data and for creating a mask or table of adjustment factors. This table of adjustment factors should also be examined for its justifiability. That is, when it says the original trip table should be corrected by a factor, does that seem justified by local knowledge or lack of original data?

In building a mask, there are two "usual" methods. One is to build a mask of the differences between the D&A trip table and the "better" trip table and then apply those differences to any future scenarios. Another is to build a mask of factors between the D&A trip table and the "better" trip table and then apply those factors to future scenarios. However, establishing factors for trip table cells containing small numbers of trips could easily skew future scenarios in which those small number cells grow significantly. Therefore, one could consider factoring cells with more than some fixed number of trips in the original D&A trip table and adding differences to cells with less than that number of trips.

The result is that the final calibration is accomplished by doing a normal D&A run, then multiplying by the factors mask, and then adding the difference mask. All future scenarios are treated in the same manner unless there is justification for altering the masks or the process.