A traffic loop is a coil of wire in the road that measures how long metal is sitting over it, reported as occupancy, a percentage of time. Lane three at one site reports four percent when the flow through it is zero vehicles an hour.
I am a traffic data analyst for a regional network. Fourteen years.
What a loop actually measures, because the second number is the one nobody outside this job has heard of.
An inductive loop is wire in a saw cut in the surface, with a current in it. Metal passing over changes its inductance and the detector reports a vehicle. Two loops a few metres apart give you a speed as well.
Flow is vehicles per hour. Occupancy is the percentage of time the loop is covered by metal, and it is the better measure of congestion: a lane at eight percent occupancy is flowing, a lane at thirty is a queue.
Every site reports flow, speed and occupancy per lane, per minute, to the control system, and it is that data the automatic queue protection uses to drop speed limits.
At one site, lane three, between about half three and half five in the morning, occupancy reads four percent with flow at zero.
Four percent means the loop is covered for two and a half seconds of every minute.
Flow at zero means no vehicle passed.
Both are the same loop, in the same minute, in the same record.
What that looks like in the raw data, which most people never see.
The per minute summary is built from individual detections. Underneath it there is a stream of events: loop covered at this millisecond, uncovered at this millisecond.
I pulled it for eleven of these nights.
The events are there. Nine to fourteen of them a minute, each between a hundred and fifty and four hundred milliseconds long, spread irregularly.
A vehicle at seventy miles an hour covers a loop for about eighty milliseconds. A lorry covers it for four hundred. So each of these events looks like something the size of a lorry sitting over the loop for a fifth of a second and then not.
They do not produce a flow count because flow requires the pair of loops to be triggered in sequence, and only one of the pair ever sees anything.
Which one, and this is why I have spent four years on this.
It alternates.
Sometimes it is the upstream loop and sometimes the downstream one, and never both in the same event, and there is no pattern to which.
What has been checked.
The loops. Both replaced when the section was resurfaced in 2021. New saw cuts, new wire, new feeders back to the cabinet, and the site recommissioned.
The detector card. Replaced twice, and the sensitivity setting reduced by a step, which took the count down for a week and then it came back.
Crosstalk from the loops in lanes two and four, which is a real and well known fault when loops are close and detectors are not tuned. Lanes two and four report nothing at all in those minutes and their own occupancy is zero.
The cabinet, the earthing and the power supply, all checked by the maintenance contractor in 2022 after I escalated it.
Their report says the site is functioning correctly and that the detections appear to be genuine detections.
What it does to the network, which is the only reason anybody senior has ever listened.
Nothing at four in the morning. Four percent occupancy with no flow does not trigger queue protection, because the algorithm wants both.
I have checked that carefully, because a system that dropped a fifty on an empty motorway at four in the morning would be a real problem and I would have had it fixed by now.
What I do.
I produce the reports and that site's lane three carries a footnote, which I wrote, and which says that overnight occupancy at this site is not reliable and should not be used.
The one thing I have done that is not analysis is that I went and looked.
The site is on a rural section with a lay-by about three hundred metres upstream, and in October 2022 I drove out at four in the morning and parked in it, which is a thing a data analyst has no business doing and which I did because four years of a spreadsheet is not the same as seeing a road.
Lane three is the offside lane and from a lay-by you are looking across two lanes of empty motorway at it.
The data for that night, which I pulled the next morning, has eleven events a minute for the hour I sat there.
The surface of lane three at four in the morning is wet and it reflects the lighting. Eleven times a minute, for a fifth of a second, in a place about the length of a lorry, that reflection went out.