Review and evaluation of models that produce trip tables from ground counts
This research effort was motivated by the desires of planning agencies to seek alternative methods of deriving current or base year Origin-Destination (O-D) trip tables without adopting conventional O-D surveys that are expensive, time consuming and labor intensive. This study had two objectives: (t) to conduct a review of existing approaches and models that produce trip tables from ground counts, and to select a few models for testing and evaluation, and (2) to perform a detailed testing of selected models based on application to both hypothetical and real networks, and to conduct performance evaluation and sensitivity analyses of these models. Two models, namely, The Highway Emulator (THE), and the Linear Programming (LP) model developed at Virginia Tech, were chosen for comprehensive testing and evaluation. For test purposes, these two models were applied to the following three case studies: (1) Sample Network, (2) Purdue University Network, and (3) Pulaski Town Network. While the first network was a hypothetical one, the other two were real networks. Different cases of targetable information and combinations of percentage available target cells and link volume information were used in the tests. These tests enabled a comprehensive evaluation of the performance and sensitivity analyses of the models. The test results were judged by two criteria: (1) the closeness of the model output tables to the "correct" or "surveyed" tables, and (2) the replication of observed link volumes by the models. The test results led to the following key conclusions: In general, the LP model results have proven to be superior, both in terms of closeness of modeled trip tables to the "correct"/"surveyed" tables, and in terms of replicating observed link volumes, for all the case studies, The exception to this is the structural target case, when THE produced better results, in terms of closeness of output tables to the "correct "l" surveyed" tables. This is based on the assumption that the "correct"/"surveyed" trip tables used for the case studies were in fact "correct"/"true". 2. THE model performed superior to the LP model for the structural target case (almost all the cases), where the target contains 1/0 cell values, 1 for those cells which represent O-D Interchanges that are feasible, and 0 for those that are not. This has practical implications in that if a region does not have a prior table available as target, then a structural target could be used.A word of caution must be noted with regard to conclusion # 2 above. While one would be tempted to use THE with a structural target for applications where a prior table is not available, it must be noted that the modeled results of both THE and LP turned out to be poor when compared with the "correct"/"surveyed" tables for all the cases, even though THE results were better than those of LP. However, these conclusions are based on tests on specific and limited number of networks, and under the assumption that the data used in testing and evaluation were accurate enough. The adoptability of these models and the use of one model versus the other must be decided based on the above facts, and in the context of error rates reported in this study, However, this study has highlighted the value of using such theoretical models for trip table estimation without performing conventional surveys.