30 Day Map Challenge 2026 - Day 19 (Icon Map Slicer) - Smell

How many buildings stand close to places that may smell?

Day 19's theme is Smell, which is a hard thing to map. You can't see it, it moves with the wind, and two people standing in the same street can disagree about whether it's there. What you can map is where smells could come from, and who lives and works nearby.

So the report, What's That Smell? Living Near Potential Odour Sources in Secunda, asks a question that planners, property buyers and lenders ask all the time: which buildings stand close to facilities whose processes may produce odours, so we know where to look closer before approving, buying or lending?

It's a proximity screening map. It measures distance, not smell, emissions or exposure, and I've tried to keep it honest about that throughout.

Explore the interactive report

The map shows Secunda and eMbalenhle in Mpumalanga, the five potential odour sources around them, and every building within 5 km, coloured by its distance from the nearest source.

  • Zoom in: the purple outlines are the five sources, each with a label. Around them are five distance bands: 0-250 m, 250-500 m, 500 m-1 km, 1-2 km and 2-5 km, from dark red to pale yellow. Every building is a band-coloured dot until zoom 13, when the footprints take over, in the same colours.
  • Pan or zoom: the three cards and the column chart count the buildings in the current map view. Zoom into a neighbourhood and you see how many of its buildings are within 250 m and 1 km of a source, and how they split across the bands.
  • Hover a source: the tooltip says what the site does and why it could smell, not that it does.
  • The card on the left: "Potential odour sources mapped" stays at 5. It describes the whole analysis, not the view.

Interactive report: Day 19, Smell

Five potential odour sources

I started from the OpenStreetMap extract for South Africa and built a hand-checked list of facilities, not a list of every industrial tag. Each one needed a plausible, named process that can produce odours:

Source What it does Why it could smell
Sasol Secunda complex Coal-to-liquids fuels and chemicals: the synfuels plant, two power stations, polymers, the Sasol Nitro fertiliser plant and the effluent treatment ponds Processes that can release sulphur compounds and ammonia
eMbalenhle WWTW Municipal wastewater treatment works Sewage; documented operational and sewage-management problems (2024-2026)
Evander WWTW Municipal wastewater treatment works Sewage
Trichardt WWTW Wastewater treatment works beside the Trichardtspruit Sewage
Evander industrial estate Includes FFS Refiners (tar-derived fuels and creosote) and a waste-management operation Tar, creosote and hydrocarbons

Three of the five are sewage works. Wastewater treatment produces hydrogen sulphide (H₂S), the "rotten egg" gas, so a rotten-egg smell isn't unique to heavy industry. That's one reason the map says potential odour sources throughout.

Not every facility made the cut. OpenStreetMap's landfill polygons around Secunda turned out to be mostly mine waste: ash dumps, tailings and coal discard. They're dusty but not a classic odour source, so I left them off. Other industrial sites only count as sources when a named odour-prone operation is identified inside them, so a nearby explosives plant isn't one, and its buildings are banded like any other's.

Some details still need care. The Sasol outline is high confidence, but the other four are medium: Evander WWTW's outline in OpenStreetMap is a rough hexagon that may miss its southern ponds, Trichardt's name is inferred from its location, and the Evander estate's outline is the whole estate, not FFS's own fence. Three operating statuses couldn't be confirmed from a published source, so the tooltips don't show them.

Measuring distance to every building

Buildings

The buildings are Overture Maps footprints (release 2026-10-21.0), the same national extract I used for Day 16. For the analysis I used each building's centroid. Footprints would be slightly more precise, but measuring to a centroid only overstates the distance by about half a building's width, a few metres for a house, which is small against 250 m bands.

Bands from the outline, not the centre

The distance is measured from the facility's outline, not its centre. A 2 km band around the middle of an 8 km² coal-to-liquids plant would sit mostly inside the plant. The processing is a DuckDB spatial script working in UTM metres:

  1. Read the facility outlines from OpenStreetMap with DuckDB's GDAL OSM driver, which builds full outlines from multipolygon relations, as large sites are often mapped that way.
  2. Group the Sasol parts into one complex.
  3. For every building centroid within 5 km, find the nearest source and the distance to its outline.
  4. Assign it one band, and count it once, at its nearest source.

Buildings whose centroid falls inside a source outline are "On site", mostly the plants' own buildings. There are 5,887 of them at Secunda, and they're not counted as neighbours.

Counted once, at the nearest source

Counting each building once is what keeps the map, the chart and the cards consistent. A building 400 m from eMbalenhle WWTW and 900 m from the Sasol complex is in the 250-500 m band, coloured dark orange, and counted once. If every source had its own independent buffers, that building would sit inside two sets of rings, and readers could reasonably expect it to be counted twice.

The map draws the bands as merged rings: each band is "within this distance of any source, less the bands inside it". The odd shapes are real. Where two sources are close, their rings join.

Building the map in Icon Map Slicer

Why the Slicer

Icon Map Slicer can filter the rest of the report by the map view. I bound its slice latitude and longitude to a table of the 104,287 building centroids, so panning or zooming narrows that table and the chart and cards count only the buildings on screen. The map layers don't allow click selection, because a click would push that one row's coordinates onto the buildings filter.

Rings as WKT, buildings as tiles

The source outlines and distance bands are WKT polygons. Some bands are long, so each ring is split into MULTIPOLYGON rows of whole parts, each under Power BI's limit of about 32,000 characters per value. I didn't simplify them, so neighbouring bands share exact edges, with no gaps or slivers between them.

Every one of 104,287 buildings as a data row would be heavy, so the buildings are vector tiles. A Python script writes one PMTiles file (14 MB) with two layers:

  • points (zoom 9-12): one dot per building centroid, de-duplicated to one dot per band per pixel, so a dense township doesn't produce millions of overlapping dots.
  • buildings (zoom 13-15): the Overture footprints, each with its band.

Icon Map Slicer reads both as catalog layers with rule-based colours by band. The dots show up to zoom 13 and the footprints from 13, so you go from a town-wide view to individual roofs without a gap.

Labels without icons

Each source's label sits on an extra image row at a point inside its largest outline, with an empty SVG as the icon, so the row carries only the label. Sasol Nitro sits apart from the main complex, so it has its own label, "Sasol Nitro (Sasol complex)", and doesn't read as an unlabelled sixth source.

What I took out

Most of the work on this report was deciding what to remove.

Source and band slicers. I added dropdowns to pick a source and a band. Pick Sasol, and the other sources' rings disappeared, but their buildings stayed on the map in their band colours, and the chart stopped counting buildings that were still on screen. It was defensible, but it needed explaining. Pan and zoom are now the only controls.

A solid purple fill. At 55% transparency, the Sasol complex was the biggest, darkest shape on the map, and the dark red buildings at its fence disappeared against it. A faint fill with a stronger outline keeps the sources clear as context, and lets the bands and buildings carry the story.

What the map reveals

Most of the closest buildings are beside a sewage works, not the coal-to-liquids plant. Of the 1,492 buildings within 250 m of a mapped source, 903 (61%) are nearest to eMbalenhle's wastewater treatment works. Within 1 km it's 12,876 of 16,566 (78%). The much larger Sasol complex has 554 within 250 m and 1,295 within 1 km.

Nearest source Within 250 m Within 1 km Within 5 km
eMbalenhle WWTW 903 12,876 58,129
Sasol Secunda complex 554 1,295 16,479
Evander WWTW 18 249 3,122
Trichardt WWTW 15 1,486 18,000
Evander industrial estate 2 660 8,557
All five 1,492 16,566 104,287

Buildings thin out right next to the fences. Building density is lowest in the first 250 m (177 per km²), more than doubles at 500m - 2 km (about 460 per km²), where eMbalenhle and Secunda town sit, and falls again beyond 2 km. So I kept the 5 km band: at 2 km the map would have cut the towns in half.

eMbalenhle WWTW has a documented history. Local reports describe the works as only partly working, and a sewage crisis in eMbalenhle from 2025 into 2026. In 2024, an eMbalenhle resident told the Ridge Times it was hard to eat "because of the stench" of sewage spills. That's context, not a measurement: the map can't say what anyone smells.

Checking the sources

I downloaded hydrogen sulphide readings from South Africa's air quality information system (SAAQIS), but it publishes no licence or terms of use, and the data stays with each monitoring network's owner. So no readings appear in the report or this post.

What I learned from Day 19

  • Map what you measured. A regional story was the more dramatic story, but the report could neither prove nor disprove it. The local proximity analysis is less dramatic and much more defensible.
  • Measure from the outline, not the centre. For large sites, distance from the centroid is meaningless: most of the near bands would be inside the plant.
  • Count once, at the nearest source. It keeps the colours, rings, chart and cards telling the same story, and avoids double-counting buildings that are near two sources.
  • Every control needs a reason. The slicers were analytically fine but made the map harder to read. Pan and zoom, with counts that follow the view, were enough.
  • Check every quote against the original. A search summary is not a source.
  • Context shouldn't shout. A strong fill on the sources pulled the eye away from the buildings, which are the point of the map.

Final thoughts

Proximity screening is everyday work in planning, property due diligence and environmental assessment: which sites, which neighbours, how close? The pattern from Day 19 works for any kind of potential nuisance, from odour to noise to dust: a hand-checked list of sources, distance bands from their outlines, every building counted once at its nearest source, and a map whose counts follow the view. Icon Map Slicer makes the last part easy: bind the map view to the buildings table, and every pan becomes a question about the buildings on screen.

Sources and attribution

  • Buildings: Overture Maps Foundation, buildings theme, release 2026-10-21.0 (ODbL). Centroids from the Day 16 national extract; footprints for the tiles from the Overture release on Amazon S3.
  • Facilities: © OpenStreetMap contributors (ODbL), South Africa extract, read 17 November 2026; hand-checked and grouped.
  • Evander industrial estate: FFS Refiners' hydrocarbon-waste licence application, WSP / DFFE ref 12/9/11/L240416122208/6/N (2024).
  • eMbalenhle WWTW and the sewage crisis: Ridge Times (2024-2026), including "eMbalenhle residents are tired of sewage spills", Mphikeleli Masangu, 25 May 2024; DA Mpumalanga, April 2026.
  • Industrial odour: David Hallowes, "Foul air on the Highveld: The sour smell of environmental racism", Daily Maverick op-ed, 22 January 2018; Tulani Ngwenya, "Sasol's carbon colossus is a killer, says report", Oxpeckers, 7 September 2026 (also in Business Day).
  • March 2026 episode (background): DFFE statement via SAnews, 13 March 2026.

Methodology note: The five potential odour sources are a hand-checked list of OpenStreetMap outlines; the Sasol parts (synfuels, power stations West and East, polymers, Sasol Nitro, effluent treatment) are grouped into one complex. Distances run from each source's outline to each building's centroid, in UTM metres, to the nearest source only; each building is counted once, in one band: 0-250 m, 250-500 m, 500 m-1 km, 1-2 km or 2-5 km. Buildings whose centroid is inside a source outline are "On site" (5,887) and not counted as neighbours. Overture buildings are machine-detected and include commercial, industrial and farm buildings, so counts are of buildings, not homes. The bands are proximity zones, not predicted odour areas; the counts don't measure emissions, smell or exposure, which depend on wind, terrain, weather and how each site is run. Operating statuses for Evander WWTW and Trichardt WWTW are unverified, and Trichardt's name is inferred from its location.