How wide are London's pavements? Measuring 7,908 cross-sections in Whitechapel

No public dataset tells you how wide a pavement is in the UK: not the left side, not the right side, and not the carriageway in between. So I measured all three on the same cross-section for 10 streets in Whitechapel, East London, and ended up with 7,908 cross-sections. Measuring was the easy part. The work was deciding which numbers to trust.

Why does pavement width matter?

Width decides whether a wheelchair, a pushchair or two people walking side by side can get down a street. UK government guidance (Inclusive Mobility, Department for Transport) recommends 2 metres of clear footway and treats 1.5 metres as the minimum acceptable where space is constrained. Councils and planners make decisions against those numbers every day, usually without a measured baseline for the street in front of them.

I picked Whitechapel because the UK’s geospatial data for it is publicly available and the streets are a good stress test: old, narrow, irregular, full of junctions.

What is wrong with the obvious method?

The obvious method fires a line straight out from the road centreline and measures the pavement it crosses. On a straight street that works.

At a junction it fails badly. The line shoots down the side street and counts the whole side street as pavement. My first run told me one pavement was 44 metres wide.

The bigger problem is that this method is right about 95% of the time. The failures don’t move the average, so summary statistics look fine. They hide in the tail, 4.7% of readings, and the tail is where the bad decisions get made: those are the junctions and pinch points that matter most for accessibility.

How did I fix the measurement?

I changed the question each measurement asks:

• Aim at the kerb, not the middle of the road. Each cross-section is anchored to the kerb line, so a junction ends the measurement instead of extending it.
• Measure from surveyed Ordnance Survey geometry. Aerial imagery is used only to check a number, never to set one.
• Let OpenStreetMap fix what OS mislabels. Cycle tracks filed as “path” would otherwise be counted as pavement.
• Reject anything it can’t trust, and record why. Every discarded cross-section carries a reason code. Nothing disappears silently.

How do you know the numbers are right?

I tried to prove myself wrong four ways that share no code with the measurement:

1. Against Ordnance Survey’s own carriageway widths: median difference 0.19 m, correlation r = 0.91.
2. Against the surveyed surface: the median measured pavement strip sits 99.9% on surveyed pavement polygons.
3. Closure against building frontage: where buildings line both sides, left + road + right should add up to the frontage-to-frontage distance. The median gap is 0.00 m.
4. Sensitivity: 12 different parameter settings move the answer by under 10 cm.

None of these proves the data perfect. Together they make it very hard for a systematic error to hide.

What did the data show?

58% of Gunthorpe Street’s pavement is narrower than 1.5 m, below what UK guidance calls the minimum acceptable. Along 41% of one side there is no pavement at all.

That is the kind of finding an average would never surface, and the kind a council can act on.

What I took away from it

• Validate the tail, not the mean. A method that is right 95% of the time can still be useless for the decisions that matter.
• Separate the instrument from the check. Imagery that sets numbers can’t also be the thing that validates them.
• Make rejection a first-class output. “We couldn’t measure this, and here’s why” is data too.

I’m planning to publish the cross-sections and strip maps for all 10 streets. If you work on streets, accessibility or urban data and want them sooner, get in touch.