Where does a retail centre or high street begin and end? It sounds like a simple question, but ask three different people — a planner, a retailer or a researcher — and you will get three different answers.
That ambiguity has quietly undermined town centre policy for years. Without an agreed, consistent definition of where a retail centre is, you cannot reliably compare one centre with another, track how a high street changes over time, or join up the many datasets such as footfall, business rates, vacancy rates or events that describe what is actually happening on the ground.
Today, we are releasing an updated, fully open dataset that answers this question for the whole of the United Kingdom.
The Retail Centre Boundaries dataset maps the boundaries of every town centre, high street, district centre, market town and retail park we can identify, using only openly licensed data. This post walks through how we built it, and the journey from raw points of interest to polygons; how we checked the results against the real world, and how it improves on our earlier attempts.
Explore our Retail Centre Boundaries on GeoDS Mapmaker
The headlines
- 9,477 retail centres mapped across the UK
- +47.5% more centres than our 2022 release (which found 6,423)
- Built from a starting pool of 4.2 million points of interest
- 1,172 retail parks identified directly; up from 497
- 530 km² of retail space delineated (up from 356 km²)
- Coverage that roughly doubles in Scotland and triples in Northern Ireland
- 100% open data — reproducible from end to end
Why consistent boundaries matter
Town centres and high streets are anchors of local economic resilience and civic identity, not just places to shop. Yet national policy leans on phrases like “town centre first” and “high street recovery” while the underlying evidence base struggles with a basic spatial problem: there is no standard, openly available definition of where a centre actually sits. Local authorities draw their own boundaries for planning, but these vary in method, currency and precision, which makes national comparison and longitudinal monitoring difficult. Commercial products exist, but they are not open, so they cannot underpin reproducible research or be freely used by every council in the country.
A consistent, open boundary set fixes the denominator. Once everyone is measuring the same places the same way, footfall counts, event listings, vacancy surveys and opening hours can finally be compared on a like-for-like basis.
From four million points to a map
This is the central challenge of our work, and is conceptual before it is computational: retail centres are not tidy objects with self-evident edges. They are fuzzy, contested and constantly shifting. The task is to take millions of individual shops, cafés and services and decide, defensibly, which of them belong together as a ‘centre’, and where that centre stops.
Starting with the points
We begin with Foursquare Places, a global and openly licensed point-of-interest dataset. Each record is a business with a category, a name, coordinates and, crucially, temporal metadata such as closure dates. Foursquare gives us two things our previous retail centre classifictaion inputs could not: coordinates that are already geocoded (no error-prone address matching), and consistent coverage across all four nations. Our starting pool is roughly 4.2 million UK points of interest.
Cleaning the data
Raw point-of-interest data are, however, messy. So, before any spatial analysis, we ran a rigorous set of quality filters. This removes businesses with recorded closure dates and those matching a curated list of 157 defunct UK retailer brands (e.g. Woolworths, Debenhams, Comet etc), so the map reflects today’s retail landscape rather than a historical accumulation. We then validate every coordinate against the ONS Postcode Directory (dropping records that sit more than a kilometre from where their postcode says they should be), strip out non-UK addresses, and detect duplicates using a combination of spatial proximity, category and fuzzy name matching. A final step keeps only the most recent business where premises have changed hands.
In total, around 7.2% of ‘open’ points fail at least one quality check and are removed, with the closure-date and postcode checks doing most of the heavy lifting.
We then narrow to genuinely retail-relevant categories of shops, food and drink, arts and entertainment, plus a few key services like banks and post offices, which retain roughly a third of the cleaned points, leaving about 1.3 million to work with. The rest of Foursquare’s taxonomy (healthcare, education, transport) is not relevant to retail centres and is set aside.
Cleaned points then feed two parallel pathways, because traditional high streets and car-oriented retail parks look completely different in the data and need different treatment. This split is one of the headline improvements over our earlier work, and the retail-park pathway in particular is a genuinely new contribution.
We then set about removing false positives. Some clusters may look like retail centres, but are better identified as business parks, trading estates and logistics depots, and so are less relevant to retail shopping. We spot these by overlaying OSM commercial and industrial land-use polygons and removing candidates that sit inside them, which strips out 604 misidentified centres, mostly in peri-urban and suburban locations. A complementary step reconnects centres that a park, civic building or other gap has artificially split, merging 986 small fragments back into the high streets and parades they belong to.
Checking it against the real world
We paired a range of automated checks with expert consultation. As with the previous release, we built an interactive web map for ground-truthing, then ran workshops with academic researchers, local authority planners and commercial-sector representatives who reviewed the centres in areas they know intimately and flagged boundaries that needed adjustment.
Practitioners largely confirmed that the Leiden-derived subcentre splits matched their operational understanding of how their cities are structured, but they also identified cases, typically in transitional zones between centres, where the algorithm and local knowledge diverged, and we adjusted thresholds in response. Boundary delineation involves judgements that benefit from local expertise, and building that feedback loop into the process is important.
Help us get the classification and the names right
Boundaries tell you where a centre is, but not what role it plays. So every centre is also assigned to a place in the retail hierarchy and given a name. These were derived algorithmically, and both are exactly where we would value the community’s eyes.
The classification sorts centres into seven tiers — regional centre, major town centre, town centre, market town, district centre, local centre and small local centre — with retail parks and shopping centres handled separately.
A centre’s tier is decided by a deterministic rule set that combines its retail unit count with how it ranks within its region, its local authority and its built-up area, plus a few sensible adjustments (for instance, district centres with several major anchor stores are promoted to town centres). It is a principled set of rules, but it is still a set of rules, and there will be places where the label does not match how a centre actually functions locally.
The names are assembled automatically too, drawing on OS Open Names settlements, OSM neighbourhood and quarter labels, and the dominant street names from the points within each centre, with retail park names taken from OSM land-use polygons where available. Automated naming gets most centres right and a meaningful minority subtly wrong — a centre named after the wrong adjacent neighbourhood, or a high street labelled by a side street.
This is where local knowledge is irreplaceable. We have built a feedback mechanism into the interactive map: if you know a centre that is misclassified or misnamed, you can flag it directly by clicking on a centre to add your suggestions, and we will explore these corrections for future releases. If you work in planning, retail or local government — or you simply know your own high street — please tell us where we have got it wrong.
What’s next?
We have a lot planned for these retail centre definitions over the next 12 months, so expect lots of interesting indicators that will enable you to characterise these places in greater depth, explore their economic resilience and function, and examine the characteristics of likely catchment areas.
This dataset was produced by the Geographic Data Service. It builds on earlier open retail centre boundary work. Read the draft academic paper supporting this project here.
