Files in each delivery
- File
- Extracted road layer
- Format
- Vector, common GIS formats
- What's in it
- Road centrelines for your area, one line per segment the imagery shows. This is the full extraction, so it is your baseline on the first pass and a fresh read of the whole area on every pass after. One delivery per quarter.
- File
- Change diff
- Format
- Vector, common GIS formats
- What's in it
- Only what moved since the last pass: segments added, removed or modified, each carrying its change type. Hand it to the editors as a priority queue, or run it into the graph after review. From the second pass on.
- File
- Conflation-ready export
- Format
- Vector, common GIS formats
- What's in it
- The diff laid out for merging into an existing graph, so your conflation scripts have new geometry to match against your segment layer. You choose the format your pipeline reads.
- File
- Low-confidence flags
- Format
- Attribute on the vector layers
- What's in it
- Stretches the imagery could not show clearly, such as canopy, deep shadow or cloud, are marked as such. They stay on your editors' list. They are never reported as removed roads.
How it works
01in
Define the area
Send a boundary or a tile set, and say which vector format your pipeline reads. If you have a segment layer you want the diff matched against, send that too.
02run
Quarterly pass
High-resolution RGB satellite imagery is collected over the area, once a quarter.
03run
Extract and vectorise
Roads are segmented in the pixels and traced to centrelines.
04out
Diff
Each pass is compared with the previous one. The first pass becomes the baseline; later passes return the diff alongside the full layer.
Limits
- Condition
- Resolution floor
- Effect on the output
- At 0.5 m a pixel is smaller than a lane, so a new road reads clearly. At the coarse end of the range, 2 m, a pixel is wider than a footpath or a farm track, and narrow roads come out as thin, uncertain lines. Junction geometry is the first thing to suffer. Expect review there.
- Condition
- Canopy and shadow
- Effect on the output
- Roads under trees, in building shadow or in a deep cut are partly or wholly hidden from above. They are flagged low-confidence, and a hidden road is not a removed road.
- Condition
- Unpaved and narrow tracks
- Effect on the output
- Tracks are extracted only where they are visually distinguishable at the stated resolution. Many will be missed.
- Condition
- Clouds and clear-sky windows
- Effect on the output
- Optical imagery needs a clear day. A cloudy quarter can push a pass later or leave gaps in the area.
- Condition
- Quarterly, not continuous
- Effect on the output
- The graph is as current as the last pass over each tile. A road built the week after a pass appears in the next one.
- Condition
- Not survey grade
- Effect on the output
- Positions are derived from imagery and need review before anyone treats them as ground truth. The layer does not carry road class, speed limits, turn restrictions, one-way flags or names.
Why this exists
New roads show up in imagery well before they show up in the graph. Editors work the queue in arrival order, whatever it contains, and a survey cycle may not reach every region before the next map release. Often nobody holds a count of the roads the graph has never heard of, so catch-up on a fast-growing region is guesswork about where to look.
Who it's for
For
- Basemap production leads and map data teams who keep a road layer or routing graph current
- Teams with a conflation and QA pipeline that can take an incoming vector layer
- Regions where new construction outpaces the edit queue
Not for
- Survey-grade centrelines or cadastral road boundaries
- Real-time or continuous graph updates
- Road attributes such as class, speed, restrictions and names
- Teams that only need a one-off image
Questions from the production team
Does this replace our survey or our edit queue?
No. It adds an imagery-derived source and tells your editors where to look first. Anything that gets treated as ground truth still goes through your own review.
What formats do the layers come in?
Common vector formats. Tell us what your pipeline reads and we will confirm it before the first delivery.
How do we check it against our own graph?
Pick a tile you already trust and compare the extraction with your own layer. Count what it found that you lacked and what it missed, then judge whether it earns a place in the pipeline.
Will it match our segment IDs?
The extraction comes with its own geometry. Matching it to your IDs is conflation work in your pipeline, and the conflation-ready export is laid out to help with that.
How current is current?
As current as the last quarterly pass over that tile.
Can you go faster than quarterly in a fast-growth region?
Ask. Quarterly is the cadence on offer. Anything faster would be scoped with you separately.
Does it get unpaved tracks?
Only where the imagery can show them. Narrow or covered tracks are often missed.
Who owns the output and the imagery?
Ownership and licence terms for the deliverables and the underlying imagery go into the project agreement before anything is ordered. Ask for them early if your legal team needs to read them.
How is a project priced?
It depends on the size of the area and the imagery source, and we price it for your area. We will not quote a figure on this page.
Send us the region, get the diff.
Tell us the area and the vector format your pipeline reads. We will say whether a quarterly pass fits your release calendar.
Start a project