Traffic and speeds
Every street on the network map has a daily flow, a lorry share and speeds. Here is where each figure comes from, and how much weight it will bear.
The best figure first
Each stretch of road takes the best figure it has. Your own survey comes first, unless a DfT count point on that stretch has a figure for a later year. Then DfT's figure for the stretch, or one from further along the same road. Our model covers the rest, which is most minor roads. The street card names the source.
The range
Every modelled flow comes with a low and a high figure. Each run checks the range on counts the model has not seen, and sets it so about 8 in 10 of them fall inside. It is narrower where the model is surer. A wide range on a street that matters to you is where your next survey should go.
How accurate it is
Every run splits the counted minor roads into five groups and predicts each group with a model that never saw it. Read a modelled figure on a quiet street as a guide to its scale: tens, hundreds or thousands a day. It is not a survey.
Speed limit
Where OpenStreetMap records the limit, we use it. Where it does not, the limit the police recorded in a recent collision there. Elsewhere it is the default for the type of road, built-up or not: 30 mph in England and Scotland, and 20 mph in Wales since September 2023. Near the Welsh border, a street can take the other side's default.
85th percentile
The 85th percentile is the speed 85% of drivers keep to or under. Your own survey gives it where you have one. On some motorways and main A roads it comes from satellite pictures, a late-morning sample. Everywhere else it is our estimate. The street card flags it over the limit only where it is measured: by your survey, or by satellite where at least ten vehicles were seen. A satellite flag shows where a speed survey is worth doing. The survey is the evidence you take to the police and a speed limit review.
Rush hour and bus delays
Bus positions, logged a minute apart, give each road's bus speed hour by hour across England. The slowest weekday hour against the fastest is the slowdown. Added up over every bus, it ranks the roads where a bus lane or signal priority would save the most.
Your surveys
Send us your automatic traffic counts, radar speed surveys and manual counts as a CSV. We place each site on its road, and the model re-runs overnight with it. Your speeds, and your counts below A roads, also train the model, which sharpens the figures for the streets around them. You get more from the surveys you already pay for.
Our modelled flows, estimated speeds and bus speeds come with an authority or police account. Anyone can see DfT's counts on the open map. Each source and its licence is on the data and licences page.
Each road link (the stretch between two junctions) takes the best figure it has, in this order. The street card says which one it used.
DfT publishes a figure every year for every motorway and A road link between major junctions. It counts most links only every few years, and estimates the years between. We take DfT's latest figure, so on a main road it is usually an estimate. DfT's road traffic statistics site shows which years each count point was counted.
DfT has counted traffic at around fifteen thousand points on minor roads since 2010. The model learns from those what a road's traffic tends to be, from things we know about every road. Technically it is a gradient-boosted tree model: many small decision trees, each one correcting the errors of the ones before.
DfT counts a minor road on a single day, and has counted many of them in several years. We use every year it counted: each count is moved to today's level by how much traffic rose or fell across Britain that year, and the recent ones count most. 2020 and 2021 are used only where there is nothing else. So one odd day weighs less.
What it looks at for each road:
Along a stretch of one road (by its number, or a street's name), between junctions with other numbered roads, the links the model estimates share one figure: traffic changes where another main road joins, not at every side street, so the map's flow does not jump at a junction that cannot explain it. Each run checks this on counts the model has not seen before using it.
Where one of these is not loaded yet, the model does without it. Each run also trains the model on only what it looked at before October 2026, and keeps whichever predicts the counts it has not seen better.
A second model learns how far off the estimate tends to be on a road like this one, from how far the main model missed each count it had not seen. The range reaches that far either side of the estimate, scaled for each kind of road (B, C or unclassified, built-up or not) so that about 8 in 10 of those unseen counts fall inside. So 'likely 400 to 4,300 a day' means there is roughly an 80% chance the real flow is in that band. A wide range means the model is unsure about that road. A narrow one means similar roads nearby have been counted and agree.
Before October 2026 the range came from two more versions of the model, one aiming low and one high, and was not checked this way. Tested on counts it had not seen, it held the real count about 3 times in 4, not 4 in 5. With the new inputs and the second model, the honest range is as narrow as the old one was on the roads DfT has counted (the high figure about 6 times the low). It can be wider on very quiet roads, which DfT hardly ever counts.
Every run tests itself. The counted minor roads are split into five groups, and each group is predicted by a model that has never seen it. The street card shows the latest result. On the first run the typical miss was about 45%, and just over half the estimates were within 50% of the real count. Since the October 2026 re-run with the new inputs, the typical miss is about 42%, and almost 6 in 10 are within 50%.
So read a modelled figure on a quiet street as a guide to its scale: tens, hundreds or thousands a day. It is not a survey.
The lorry share comes with the flow figure, where your survey or DfT's figure has one. Everywhere else a second model estimates it, trained on the counted minor roads in the same way.
We take OpenStreetMap's recorded limit where there is one, on around four in ten road links. Where it has none, we take the latest limit the police recorded in an injury collision on that stretch or street in the last three years (STATS19). At the exact spot the police's and ours agree about three times in four, so the street card says where its limit came from, and asks you to check where the police recorded a different one. Everywhere else we assume a limit from the type of road:
We judge whether a road is built-up by how much road there is around it. Near the Welsh border, a street can take the other side's default.
The satellite data covers only Europe's E-road routes, which in Britain are some of the motorways and main A roads. Until you add speed surveys, speeds on town streets are the least certain figures on the card.
The speed that 85% of drivers keep to or under. Officers use it alongside the mean speed to review limits and choose where speed checks are worth doing. In order:
The street card flags an 85th percentile above the limit only where it is measured: by your survey, or by satellite where at least ten vehicles were seen. A satellite flag shows where a speed survey is worth doing. It is not evidence on its own. Your survey is.
The typical speed, slowed by as much as buses slow down on that road at rush hour. Rush hour is 7 to 9am and 4 to 6pm on weekdays, set against 10am to 3pm. Where too few buses have been logged, it takes the slowdown on similar roads in similar areas.
Bus positions come from the Bus Open Data Service, which covers England. They are logged a minute apart, and how far a bus moved since the minute before gives its speed. That speed is added to the road it was on, for that hour of the week. So bus delays are for England only, on roads where buses have been logged.
For each road, the busiest weekday hour is set against the quietest. The busiest is the slowest hour between 7am and 8pm. The quietest is the fastest hour, often late evening or early morning.
Bus speeds include time at stops. Comparing a road with itself at other hours cancels most of that out, so what is left is mostly traffic.
Send us your automatic traffic counts (ATCs), radar speed surveys and manual classified counts as a CSV. We place each site on its road and add it to that street's card. The model re-runs overnight with it.
Your count becomes the figure for its link, unless DfT has a later one there (see above). Your speed survey becomes the link's speed. Your speeds, and your counts below A roads, also train the model for the streets around them. Ask us to withdraw a survey and the next night's run leaves it out.
The CSV needs one row per site. Give its position as latitude and longitude, or easting and northing. Add at least one of a daily flow (AADT or ADT), a mean speed or an 85th percentile speed. Add an HGV share if you have it. Speeds are read as mph unless the heading or a units column says km/h. A site reference helps. So does the survey's end date: it decides whether your count or DfT's comes first.