30 Day Map Challenge 2026 - Day 4 (Icon Map Pro) - Clusters

Which fuel brand dominates which part of South Africa, and where do the forecourt networks cluster?

For Day 4 of the 30 Day Map Challenge, the theme is Clusters. After points, lines and polygons, I wanted a dataset with thousands of locations that only make sense when grouped, and with a category inside each group worth comparing.

South Africa's fuel stations fit well. There are several thousand forecourts, six big brands and a long tail of independents, and the market is changing fast:

  • Shell is leaving. In July 2026, ADNOC Distribution agreed to buy Shell's South African downstream business, including almost 600 service stations, in a deal of about R16 billion. The deal is still subject to regulatory approval and is expected to close in 2027.
  • Caltex has become Astron Energy. Astron passed its 700th rebranded site in April 2026.

The question for a fuel retailer, a franchisee or a buyer is simple: where does each brand lead, and where is there white space?

The report uses Icon Map Pro's clusters, split by a Cluster Group field. Each cluster is a small donut chart of the brands inside it. As you zoom in, the clusters break into smaller clusters and then into individual forecourts.

Explore the interactive report

Each cluster shows the number of stations in the middle, with a segment per brand. Click a cluster to zoom to its stations, which also selects them and filters the cards and chart. Use the brand buttons under the map to show one or more brands. Click a province or brand in the chart to filter the map.

Interactive report: Day 4, Fill Her Up: Who Owns South Africa's Forecourts?

The map shows fuel stations recorded in OpenStreetMap. It is not an official forecourt register, and brands change faster than the map.

Two ideas that didn't make it

Clusters was the day I changed my mind twice.

My first idea was short-term rentals in Cape Town using Inside Airbnb. The data is open and licensed under CC BY 4.0. However, the project's data policy also asks users not to republish the data, and this report is published to the web. So that idea was dropped.

My second idea was mobile network coverage from OpenCelliD (CC BY-SA 4.0). The South African file has 88,917 cells, but 88% are tagged as Vodacom and only 7% as MTN, even though their real networks are comparable in size. About 60% of the cells are in and around Gauteng. The data shows where people contributed measurements more than where the networks are, so comparing operators would have been misleading.

Both datasets were useful reminders that open does not always mean publishable, and crowd-sourced does not always mean representative.

Starting with OpenStreetMap's fuel stations

The fuel stations come from the Geofabrik South Africa extract of OpenStreetMap, downloaded on 30 October 2026. Features tagged amenity=fuel were extracted: nodes (single points) and ways (forecourts mapped as outlines, reduced to their centre point).

Reading the extract without GIS software

I usually reach for Osmium Tool or GDAL for this, but wanted to try something different so I wrote a small Python reader for the .osm.pbf format, using only NumPy. It makes two passes over the file:

  1. Find the matching features. It skips any data block whose string table doesn't contain the wanted tag value. That leaves about 1,250 of 11,139 blocks to read in full, where it picks out the matching nodes and ways.
  2. Locate the ways. It looks up the node coordinates needed to place each matching way.

The 423 MB extract was processed in about 100 seconds. The result was 4,736 fuel stations: 4,107 nodes and 629 ways. The reader is generic, so it will be reused for other OSM days in the challenge.

Grouping the brands

OSM records brands in the brand tag, which most stations have. I grouped them into eight categories:

  • Engen
  • Astron Energy, including 130 stations still tagged Caltex
  • Shell
  • TotalEnergies, including stations tagged Total
  • BP
  • Sasol
  • Independent, any other brand or a recognisable independent name such as Puma, Global Oil, Quest or Exel
  • Unknown, no brand and no recognisable name

Each station also got its province (a point-in-polygon test against the nine OSM provincial boundaries also used on Days 2 and 3), its town, and flags for diesel, LPG, shop, 24/7 opening and toilets where these are tagged.

How complete is OSM for fuel?

This series keeps returning to the question of OSM completeness, and this time the answer is encouraging. Comparing the OSM counts with the networks published by the brands (BusinessTech, May 2025):

Brand OSM stations Published network
Engen 1,083 Over 1,000
Astron Energy (incl. Caltex) 837 About 850 once rebranded
TotalEnergies 608 550
Shell 599 Almost 600
BP 506 500
Sasol 402

For the major brands, OSM is within a few percent of the published figures. Brand-mapping projects, which add brand tags consistently, deserve much of the credit. The gaps are in the details: 449 stations have no name, and 424 have no brand and no recognisable name, mostly in Limpopo, North West and the Northern Cape.

Clusters in Icon Map Pro

Clusters in Icon Map Pro are an option on the Circles layer. Each station is a row with an ID, longitude, latitude and a small fixed circle size (5 pixels). Turning on Cluster circles groups nearby stations, and putting the brand into the Cluster Group field turns each cluster into a donut chart split by brand.

A few settings made a big difference:

  • Cluster radius 60 and minimum 3 points. This gives readable clusters at national scale without hiding small towns.
  • Max cluster zoom 13. At street level, the clusters stop and every forecourt shows as its own dot.
  • Zoom on Click and Select on Click both on. Clicking a cluster zooms in and selects its stations, so the cards and chart describe that cluster.

Keeping the colours consistent

The trickiest part was colour. By default, Icon Map Pro gives each Cluster Group value a colour from the report theme, in the order the values arrive. Filtering out a brand can then shift the remaining colours.

Instead, I set Color Source for Cluster Groups to From a field or rule. The colour then comes from a DAX measure, Station colour, that returns a fixed colour for each brand group. I also set a Custom order (Engen first, Unknown last), so the donut segments always appear in the same order. The same colour drives the station dots, so a forecourt keeps its colour whether it's inside a cluster or not.

The colours are inspired by the brands but are not logos: Engen blue, Astron red, Shell yellow, TotalEnergies orange, BP green, Sasol cyan, independents purple and unknown grey.

A slicer that doubles as the legend

Under the map, a button slicer lists the eight brand groups, each button filled with the same colour as the map. It filters the map, cards and chart, and because the colours match, it also works as the legend. That freed up space on the map itself.

Next to the map, a 100% stacked bar chart shows each brand's share of stations by province, sorted by the number of stations. Selecting a province or brand filters the map.

What the map reveals

Nationally, Engen has the largest mapped network, with 1,083 stations (23%), followed by Astron Energy (837) and TotalEnergies (608).

But the clusters show that the market is regional:

  • Engen leads Gauteng (28% of 1,543 stations), KwaZulu-Natal (24%), the Free State (24%) and North West (20%).
  • Astron Energy leads all three Cape provinces: the Eastern Cape (28%), the Western Cape (25%) and the Northern Cape (24%). This reflects Caltex's long Cape history.
  • TotalEnergies leads Mpumalanga (22%), with Sasol third (15%). Sasol is strongest in Mpumalanga and Gauteng.
  • Shell's network is spread fairly evenly, at 8–17% of stations in every province, with its highest share in KwaZulu-Natal (17%). For a new owner, that means a ready-made national footprint rather than a regional stronghold.
  • BP is concentrated in the metros: 15% in the Western Cape and 13% in Gauteng, but only 4% in Limpopo and Mpumalanga.

These are shares of mapped stations, not of fuel volumes or market share. A large highway site and a small rural forecourt count the same.

What I learned from Day 4

1. Check the data policy, not just the licence. A dataset can be openly licensed and still ask you not to republish it. For a report that will be published to the web, that settles it.

2. Crowd-sourced data shows where the crowd is. The OpenCelliD file was technically fine, but its contributor bias made the most obvious comparison misleading.

3. Sometimes OSM is very good. For branded fuel stations, OSM is close to the brands' own networks. That is worth saying as often as pointing out the gaps.

4. Fix cluster colours with a rule. Driving Cluster Group colours from a measure, with a custom order, keeps the brand colours stable when the report is filtered.

5. Clusters need a zoom plan. The cluster radius, minimum points and maximum cluster zoom together decide whether the national view and the street view both work.

Final thoughts

Clusters are useful for showing that a market which looks national is actually regional. Six brands sell fuel across South Africa, yet each province has its own leader. And the Shell sale means the map is about to change.

The same approach works for any branded network: banks, retailers, pharmacies, franchise restaurants or competitors' depots. Extract the points, group the brands, set fixed colours, and let the clusters show where each network is strong.

Sources and attribution

  • Fuel stations and province boundaries: © OpenStreetMap contributors, licensed under the Open Database Licence (ODbL). Extracted from the Geofabrik South Africa download (30 October 2026), amenity=fuel.
  • Published network sizes: BusinessTech, "The biggest and most popular petrol station brands in South Africa", 13 May 2025.
  • Shell sale: BusinessDay, 7 July 2026; EWN, "Sale of Shell's local petrol station network moves a step closer", 21 August 2026.
  • Astron Energy rebrand: Engineering News, "Astron Energy nears rebrand completion with 700th site milestone", 14 April 2026.

Methodology note: Station counts are mapped OSM records, grouped into brand categories by the author. They are not an official forecourt register and do not reflect fuel volumes, ownership or market share. Ways are represented by the mean of their nodes. Brand changes since the extract date (including the Caltex to Astron Energy rebrand and the Shell sale) may not be reflected in OSM.