Which Cape Town buildings can actually see the mountain?
For Day 5 of the 30 Day Map Challenge, the theme is Senses: Sight. After Day 4's fuel station clusters, I wanted to map something you can only see from where you stand. In Cape Town, the obvious thing to see is Table Mountain.
A mountain view is part of how property in the City Bowl and on the Atlantic Seaboard is described and valued. But "mountain views" covers a lot. Which landmark? From the street, or from the roof? And does the tower across the road get in the way?
So the question for a developer, valuer or estate agent is: from which buildings can you see which landmarks, and how many of them? I picked five landmarks around the City Bowl:
- Maclear's Beacon, the highest point of Table Mountain
- Lion's Head
- Devil's Peak
- the Upper Cable Station
- Signal Hill
I then modelled whether each roof can see each one.
Explore the interactive report
The map opens in 3D, looking south from above the harbour towards the mountain. Every building is raised to its modelled height. By default, each roof is coloured by how many of the five landmarks it can see: grey for none, then pale pink for one through to deep magenta for all five.
Use View towards to pick a single landmark. Buildings that can see it take on that landmark's colour, the cards update, and the landmark's label on the map changes to the same colour. Hover over a camera marker to see a photo of the mountain from that spot. Click a column in the chart to keep only the buildings with that count. You can also tilt and rotate the map to look along the sightlines.
Interactive report: Day 5, What Can See the Mountain?
The map shows a model of line of sight, not a guarantee of a view. Trees, walls, balconies and window positions aren't modelled. A building marked as seeing a landmark means that someone on its roof probably could.
Starting with Copernicus terrain
The terrain comes from the Copernicus DEM GLO-30, a global elevation model at about 30 m resolution.
GLO-30 is a surface model, so it includes some of what stands on the ground. That turned out to be useful later. It also has a weakness on narrow summits. The 30 m cells average out Lion's Head's spire, which the DEM puts at about 600 m against its true 669 m. For Lion's Head and the Upper Cable Station, I placed the observer at the known height rather than the DEM height.
Building footprints from Overture Maps
The buildings come from Overture Maps, release 2026-10-21.0, queried by bounding box with the Overture Python tool. This brought back 70,112 buildings from three sources:
- OpenStreetMap, about 60%
- Microsoft's machine-learning footprints, 29%
- Google Open Buildings, 11%
Checking coverage before trusting it
Before using the footprints, I compared counts with the City of Cape Town's 2023 roof outlines. I didn't use the City's layer itself: its terms are non-commercial and all rights reserved, so I only used it for a count comparison.
The comparison changed the shape of the project:
- Atlantic Seaboard and City Bowl: complete. Overture has more buildings than the City here, because the City's layer merges terraced houses into single roofs.
- Southern suburbs: only 25–53% captured. For example, Kenilworth is at 34% and Constantia at 25%. The tree canopy hides roofs from the machine-learning models.
A view score is only useful where the buildings are there to score. So I limited the report to the area where Overture is complete: the Atlantic Seaboard, the City Bowl and Woodstock to Observatory. That leaves 29,592 buildings, about 91% of them from OpenStreetMap.
From terrain line of sight to rooftop line of sight
A first pass with terrain only
The first version followed the usual recipe. For each landmark, it calculated a viewshed: the area of ground from which the landmark is visible. Line of sight works both ways, so five viewsheds cover every building.
I wrote the sweep in Python with NumPy. It traces rays out from each landmark and allows for the earth's curvature and atmospheric refraction. It runs on the DEM resampled to about 10 m for smoother edges, which is cosmetic and adds no new terrain detail.
Why terrain alone wasn't enough
The terrain-only result looked fine in 2D. Then I switched on 3D buildings. Rows of houses behind CBD towers were coloured as having a clear view, with a 100 m tower standing directly between them and the mountain.
So the second version adds the buildings themselves to the surface:
- A building surface. All 70,112 Overture footprints, including those outside the focus area, are drawn onto the 10 m terrain at their height.
- Building heights, in this order:
- Overture's measured height (159 buildings)
- else the number of floors × 3 m (9,016)
- else, for large, low-lying blocks, an estimate from the Copernicus surface model (162 in the focus area). This is where GLO-30 being a surface model helps: big buildings show up in it.
- else a default of 6 m, roughly two storeys
- Rooftop observers. Each roof is tested at its height plus 1 m, roughly someone standing on the roof or at a top-floor window. A building can see a landmark if any of its roof cells can.
The result is a score from 0 to 5 for every building, plus a yes or no for each landmark.
The Maclear's Beacon surprise
You'd expect the highest point on the mountain to be the most visible. It isn't. Maclear's Beacon can be seen from only 667 of the 29,592 buildings, about 2%. It sits back from the front edge of the plateau, so the mountain's own flat top hides it from almost the entire City Bowl.
3D buildings from PMTiles in Icon Map Pro
The footprints are stored as PMTiles, a single-file vector tile format. The file is loaded into Icon Map Pro's Shapes from Files layer, and each tile shape is matched to its row in Power BI on the building ID.
This keeps the heavy geometry out of the data model. Power BI only holds the building ID, its score, the five yes-or-no flags and its height. A measure returns each building's colour, and a second measure, Building height, drives the layer's extrusion. That makes every building stand at the same height the line-of-sight model used, so what you see on the map is what was calculated.
Terrain, exaggerated
The map uses Icon Map Pro's built-in Mapterhorn terrain and hillshade, with the satellite basemap draped over them. I set the terrain exaggeration to 10. That is far more than real life, but at city scale it makes the relationship between the slopes, the buildings and the summits readable at a glance.
Landmarks, viewpoints and photo cards
The five landmarks are drawn as triangles in their own colours, and four photo viewpoints as camera icons. Both are SVG images returned by a DAX measure. Each has a card label whose title is always shown: the landmark's name, or "View from: Sea Point promenade". The landmark labels are black by default. When a landmark is picked under View towards, its label turns that landmark's colour, so the eye goes straight to it.
Hovering over a camera opens a landscape card with a photo of the mountain from that spot: Sea Point promenade, the V&A Waterfront, Kirstenbosch and Camps Bay. Each card has a credit line. All four photos are CC BY or CC0 from Wikimedia Commons.
Design choices
Colour. My first count ramp was a set of teal shades, which almost disappeared against the satellite imagery. A rainbow of distinct colours fixed that but clashed with the five landmark colours. The final ramp runs from grey through pale pink to deep magenta. It uses a hue none of the landmarks use, and each step is clearly lighter or darker than the next.
The chart is the legend. The column chart of buildings by count uses the same colours as the roofs, so there is no separate legend on the map.
What the map reveals
- Almost every building sees something. About 96% of the buildings in the area can see at least one of the five landmarks from the roof. Lion's Head is visible from 83%, the Upper Cable Station from 74%, Devil's Peak from 71% and Signal Hill from 46%.
- Woodstock and Salt River have the best all-round views. Their buildings see an average of 3.6 of the five landmarks, the highest in the study area. They look across at Devil's Peak, the front of the mountain and Signal Hill all at once. Sea Point and Green Point average only 1.3, because Signal Hill and Lion's Head block most of the mountain behind them.
- Camps Bay sees Lion's Head and nothing else close by. Every building there sees Lion's Head, and most see the Upper Cable Station. Not one sees Signal Hill.
- The full set is found in Observatory. All but one of the 656 buildings that can see all five landmarks are in Observatory and Mowbray. It is the only area that sees around the side of the plateau to Maclear's Beacon.
Some caution is needed. These are modelled sightlines over a 30 m surface, with a default 6 m height for about two-thirds of the buildings and no trees. They show where a view is likely, not that a particular flat has one, and they say nothing about how good the view is.
What I learned from Day 5
1. Check coverage before you score. The Overture footprints were excellent in one part of the city and patchy in another. Counting against a reference layer, even one I couldn't use directly, decided the study area.
2. Put the obstacles in the model. A terrain-only viewshed answers "can the ground here see the summit?", not "can this building?". As soon as the map went 3D, the buildings had to block the view too.
3. Make the map show what was calculated. Extruding each building by the same height the model used means the 3D view and the numbers agree.
4. Pick colours against the basemap you'll actually use. A ramp that works on a light map can vanish on satellite imagery, and it shouldn't reuse the colours that already mean something else on the map.
Final thoughts
Sight is a sense that depends on where you stand, and that is what makes it worth mapping. The same building can face a landmark across the bay and still be blocked by the hill behind it or the tower next door.
The approach works well beyond Cape Town: sea views, views of a new development, sightlines to a wind farm or a mast, or which windows look onto a proposed tower. Start with a surface that includes the buildings, put the observers where people actually stand, and let the map show the result in 3D at the same heights the model used.
Sources and attribution
- Terrain: Copernicus DEM GLO-30, tiles S34/S35 E018. © DLR e.V. 2010–2014 and © Airbus Defence and Space GmbH 2014–2018, provided under COPERNICUS by the European Union and ESA. Free to use, including commercially, with attribution. https://dataspace.copernicus.eu/explore-data/data-collections/copernicus-contributing-missions/collections-description/COP-DEM
- Buildings: Overture Maps Foundation, buildings theme, release 2026-10-21.0, extracted November 2026. Overture's buildings theme is released under the Open Database Licence (ODbL). Its sources:
- © OpenStreetMap contributors, available under the ODbL;
- Global ML Building Footprints, licensed by Microsoft under the ODbL;
- Google Open Buildings, available under CC BY 4.0.
- Coverage check only (not mapped): City of Cape Town building roof outlines (2023), used for count comparison.
- Terrain and hillshade in the map: Mapterhorn, via Icon Map Pro. Basemap: Icon Map satellite (Copernicus Sentinel).
- Photos (Wikimedia Commons):
- Sea Point promenade, by David Stanley, CC BY 2.0
- Table Mountain and the V&A Waterfront from Table Bay, by South African Tourism, CC BY 2.0
- Kirstenbosch, view from the Botanical Gardens, by Didier B (Sam67fr), CC BY 2.5
- Cape Town coastal suburbs, by lumoplank, CC0
Methodology note: Visibility is modelled from each landmark to each building roof (+1 m) over the Copernicus GLO-30 surface model, resampled to 10 m. Overture building footprints are drawn onto that surface at their height. Earth curvature and refraction (k = 0.13) are included. Building heights come from Overture where available, else from floor counts × 3 m, else from a surface-model estimate for large blocks, else a default of 6 m. Trees, walls and window positions are not modelled, and some large buildings already in the surface model may be counted twice. Results indicate likely views from the roof, not confirmed views from any particular property, and say nothing about view quality.