Measuring queues seems to be the obvious way of checking a model, because queue lengths are one of the main outputs from the programs. In fact some traffic engineers, and their customers, insist on such checks being carried out. If so, they should understand the implications of what they are doing. Apart from the practical difficulties of measuring mean queues over successive time intervals there are also mathematical problems to consider.
During peak periods (when the flow/capacity ratios are high) there is a large daily variation in queue lengths even if the average flow for each time segment does not vary from day to day. To take a typical example, a mean queue of 26 pcu would be derived from queues which varied between 5 pcu and 50 pcu from day to day. In fact, on 1 day in 20 the queue would be outside even this large envelope of possible values. So you can see that many days of queue measurements would have to be taken to obtain a reliable estimate of mean queues. The junction model predictions are based on an infinite number of days! It is surprising how many people think that one day is enough – yet they wouldn’t dream of predicting the result of the next General Election after canvassing one person chosen at random (OK maybe this is a more extreme case!)
The routine within the junction models which calculates mean queues is more accurate than the capacity-predicting routine, which of necessity gives average results which ignore the effects of site peculiarities and location. The queue calculation is almost certainly more accurate than your best estimate of demand flows. So if you do go to the trouble and expense of measuring queues, and find a difference between the model predictions and your observations, then you might pause to consider whether perhaps the model is correct and your queues are not. Even if your measured discrepancy is genuine, you will have achieved nothing other than to demonstrate that the demand flows are inaccurate or that the model’s capacity prediction is not taking account of all the local circumstances. So time would be better spent tackling these issues.
Consider the accuracy of the demand flow estimates in the model (bear in mind that the HOUR profile type (ODTAB in older versions of ARCADY) is the least accurate method of specifying them but the most frequently used). The accuracy of the model’s capacity prediction can be significantly improved by carrying out a “site specific capacity correction” – refer to the Application Guide for details. Such on-site measurements are much easier and cheaper than trying to measure queues properly!
If the flows and capacity-predictions are correct, the queue predictions will be correct.