If you find that ARCADY or PICADY predicts queues that are significantly different to what you have observed, there are several things to consider:
How reliable are the observed queue measurements?
If queues have been observed on one day only, they may be unreliable because queue lengths have a large daily variability even with the same levels of traffic demand. The queues shown in ARCADY/PICADY are what you would expect to see if you averaged observations from many days. So ideally you should measure queues on several ‘typical’ days, and average the results. Otherwise, you need to be as sure as you can that the measured queues are a good representation of typical behaviour at the site. (If possible, visit the site to check that the level of queueing roughly corresponds with the queue survey data.)
Has demand been measured correctly?
ARCADY/PICADY need to know the volume of traffic that wants to use the junction – i.e. the demand. This should be measured upstream of any queueing. If, instead, you count vehicles crossing the give-way line, you have measured the throughput instead of the demand. If you enter this as the demand then the predicted queues will be very small, because you will just be telling ARCADY/PICADY that the amount of traffic wanting to use the junction is the same as the amount that you have observed flowing through it.
The most common demand profile type in ARCADY/PICADY is “ONE HOUR”. This takes an origin-destination matrix and then assumes that the traffic rises and falls in a specific way over a 90-minute period, to represent a typical peak period. This often gives reasonable results, but relies on a number of assumptions that may not be true at your junction. Alternatively, you can directly enter the demand for each time segment.
Are geometries correct?
Check that the geometries have been measured and entered correctly. If there is unequal lane usage, for example if traffic on one or more arms consistently uses one lane more than another, then consider using Lane Simulation mode.
Check units and other options
Check that the correct units are being used (e.g. PCU/hr versus PCU/time segment) and that you don’t have any scaling factors or other options accidentally switched on.
Consider applying calibration factors (intercept adjustments)
If all else fails then you can apply factors to calibrate the model. Usually this is via intercept adjustments applied to one or more arms. These adjust the capacity predicted by the model up or down by an amount you specify – e.g. -75 PCU/hr to reduce the predicated capacity by 75 PCU/hr. If you have measurements of the throughput on the arm, under saturated conditions (i.e. whilst there is queueing) then you can use these to directly calculate a correction, using the Calibration screen. This is considered the best method you can employ.
Alternatively you can find intercept corrections by a process of trial and error. Corrections are intended to account for factors at the junction which make the junction different to the ‘average’ junction with the same geometries, such as poor visibility, gradient, driver hesitation, unusual layout, and so on. Usually these factors apply at all times of day and in current and future years. If you find that you need to apply very large adjustments to reproduce the observed queues, this suggests that there is something wrong with the model data and it’s worth checking the points above again.