30 Day Map Challenge 2026 - Day 9 (Icon Map Slicer) - Urban-rural

What does urban or rural look like when every place gets the same amount of space?

For Day 9 of the 30 Day Map Challenge, the theme is Urban-rural. After Day 8's search for a personal utopia, I wanted to look more closely at one of the choices behind it: city, town or countryside.

Those labels suggest different ways of living, but a population total cannot show the difference on its own. A town can concentrate people in a small footprint. A rural landscape can contain thousands of people distributed across villages and scattered homes. The same average density can hide quite different patterns.

So I chose three places in KwaZulu-Natal and gave each the same 6 × 6 km window:

  • KwaMashu: metropolitan concentration.
  • Greytown: a compact town and its surroundings.
  • Keate's Drift: dispersed settlement around a rural service node.

The question is useful beyond the challenge. For someone deciding where to put a service, open a shop or investigate a development site, how many people live nearby is only the start. Are they concentrated in one place, or spread across the landscape? This report explores that pattern. It does not calculate access, demand or travel times.

Explore the interactive report

The report opens with Settlement class selected. The three maps sit side by side, with equal study areas and one shared legend.

The Map colour dropdown switches all three maps between:

  1. Settlement class: the GHSL classification, from very low-density rural areas through rural clusters and peri-urban areas to urban centres.
  2. People per 100 m cell: modelled population, using the same colour thresholds across the three places.
  3. Building footprint (%): the share of each cell covered by buildings.

Hover over cells for values shown as tooltips. The table beneath the maps compares population, the share of populated cells, average density and building coverage.

Interactive report: Day 9, City, Town, Countryside

Each map is a window around a reference point, not an outline of the named settlement. In particular, the KwaMashu window includes neighbouring areas. The figures describe the land inside the windows, not the populations of the towns or their municipalities.

Why equal windows?

One early idea was to start at each place and expand a circle until it contained the same number of people. The resulting circles would show how much space that population occupied.

But that raised a problem. A circle can expand across commercial farms, conservation land or other areas where settlement is limited. Its size would reflect the surrounding land as well as the concentration of people. Around a small town, adding countryside could make the settlement appear much more dispersed than its residential streets actually are.

That is meaningful geography, but it answers a different question.

The final comparison uses fixed windows instead: 36 km² for every place. This holds the area constant and lets the different patterns appear within it.

It does not make boundaries irrelevant. A window still includes surrounding land, and moving its centre can change the result. It simply makes the comparison explicit: the same amount of space, examined in the same way.

Three datasets, one year

All three layers come from the European Commission Joint Research Centre's Global Human Settlement Layer, using the 2020 epoch of the R2023A release.

Dataset What it supplies Resolution used
GHS-POP Modelled residential population 100 m
GHS-BUILT-S Built-up surface in square metres 100 m
GHS-SMOD Settlement class, using the Degree of Urbanisation method 1 km

Population is an estimate for a grid cell

GHS-POP distributes population from census and administrative units across a grid, informed by GHSL's built-up information. A cell's value is therefore a modelled estimate, not a household count made at that location.

For this report, each population cell covers 100 × 100 m, or one hectare. The map colours show the number of modelled people inside that cell. Calling the option People per 100 m cell makes the units clear and distinguishes it from the table's average density in people per km².

Building footprint is not the whole developed area

GHS-BUILT-S gives the built-up surface within each cell. I divided that by the cell's 10,000 m² area to get a percentage.

The report calls this Building footprint. A value of 20% means that buildings cover about a fifth of the cell. It does not mean that the other four-fifths are undeveloped or available to build on. Roads, gardens, open spaces and other uses are not the same thing as building coverage.

Settlement class describes a broader pattern

GHS-SMOD uses population and built-up information to classify the settlement pattern. Its seven land classes run from very low-density rural areas to urban centres; water is a separate category.

These are GHSL's classifications, not Stats SA's census geotypes. The report uses them as a consistent way to compare the three landscapes, rather than assigning a single official urban or rural label to each place.

There is also an important scale difference: the settlement classes have 1 km source detail. Repeating that class across the smaller population cells does not create a 100 m classification. That is why the opening view has larger blocks than the population view.

From raster windows to Power BI polygons

The processing is in Python, using Rasterio, NumPy and PyProj.

  1. Locate the windows. Transform each reference point to World Mollweide, an equal-area projection, and snap it to the nearest 100 m grid intersection.
  2. Read the source pixels. Extract the raster windows around the three locations, retaining the original population and built-up cells.
  3. Keep an equal area. Take a 6 km square around each snapped centre. Each contains 60 × 60 cells, or 3,600 rows.
  4. Attach the settlement class. Give each 100 m cell the class of its parent 1 km cell, without interpolating between classes.
  5. Create the geometry. Transform each cell's corners to longitude and latitude and write its polygon as WKT.
  6. Export the data. Save the cell ID, location, population, building coverage, settlement class and WKT to CSV.

That produces 10,800 map rows across the three locations. The areas and percentages are calculated on the original equal-area grid; the longitude and latitude geometry is for display.

Power Query imports the CSV files and sets their types. The spatial preparation is already done, so the Power BI model stays relatively small and easy to inspect.

Bringing it together in Icon Map Slicer

Three maps from one table

All three maps use the same cell table. Each has a filter for its study area, so the geometry and measure logic are shared rather than copied into three separate datasets.

The WKT supplies each polygon's location and shape, and a unique ID keeps the cells separate. A colour measure reads the selected map metric and returns the appropriate colour for each cell.

For population, the bands are fixed: fewer than one person, 1–<5, 5–<10, 10–<25, 25–<50, 50–<100, 100–<200 and 200 or more. A colour means the same thing on every map. Giving each location its own automatically scaled ramp would make the pictures harder to compare.

One selector, one matching legend

The colour selector changes the maps and the shared legend together. Population uses a pale-yellow-to-dark-red ramp, building coverage uses blues, and settlement class uses distinct category colours.

The legend title changes too. Readers should not have to work out whether a number means people, percentage coverage or a category.

The table explains the pictures

The comparison table uses four measures:

  • Modelled population: the sum of the population estimates.
  • Populated cells (%): the proportion of 100 m cells containing at least one modelled resident.
  • People per km²: population divided by the area being summarised.
  • Building footprint (%): total building surface divided by that area.

“Populated” is an exploratory threshold chosen for this comparison. It is not an official residential land-use boundary. A cell just below one person is not necessarily empty, and a cell above one person is not wholly residential.

What the map reveals

Study window, 36 km² each Modelled population Populated cells People per km² Building footprint
KwaMashu 281,742 72.2% 7,826 26.9%
Greytown 21,806 22.7% 606 5.1%
Keate's Drift 8,801 32.1% 244 4.1%

KwaMashu: concentration within a wider urban landscape

The entire KwaMashu window is classified as Urban centre. That gives the opening map a continuous purple block.

Switching to population reveals the variation within that classification: dense concentrations, less populated gaps and the shape of the wider metropolitan settlement. The classification describes the broader urban area; it does not say that every piece of land inside it has the same density or use.

With about 282,000 modelled people in the window and buildings covering 26.9% of its area, KwaMashu is clearly different from the other two examples.

Greytown: a concentration with an edge

Greytown's opening view shows a Dense urban cluster, peri-urban cells and surrounding rural classes. Switching to population makes the concentration more apparent, with much less populated land around it.

The window contains about 21,800 people, but only 22.7% of its 100 m cells meet the populated-cell threshold. Its average density combines the concentrated settlement and its surroundings.

Keate's Drift: fewer people, distributed more widely

The Keate's Drift window mixes rural classes with cells classified as suburban or peri-urban. The population view shows several concentrations rather than one dominant compact footprint.

The interesting comparison is with Greytown. Keate's Drift has fewer than half as many people, but populated cells cover a larger share of its window: 32.1% against 22.7%. That is approximately 11.6 km² of qualifying cells, compared with 8.2 km² around Greytown.

It does not have more building surface: its building-footprint share is slightly lower. The finding is about how the cells containing people are distributed, not which area has more buildings.

That is the part an average density alone would miss. Greytown has more people in a more concentrated pattern; Keate's Drift has fewer people spread across more of the window.

The boundary changes the number

During processing, I also tested 4, 6 and 10 km square windows around the same centres.

Greytown shows the effect most clearly. Its average density falls from about 1,290 people per km² in the 4 km square to 236 in the 10 km square. Yet the enclosed population rises only from about 20,600 to 23,600.

The town has not become less dense. The larger window has brought much more surrounding land into the denominator.

I also shifted each 6 km window by up to 500 m on both axes, testing nine placements per location. The population ranking stayed the same, although the totals changed. That is a check on the choice of centre, not a confidence interval for the population model.

Equal windows make the comparison easier to understand. They do not remove the need to explain what the boundaries include.

Design choices

Open with the theme. The first view was population, which showed the settlement patterns well but risked looking like another population map. Opening on Settlement class makes the Urban-rural theme explicit. Population and building coverage then reveal the finer detail behind it.

Show the places together. A single map with a location selector makes readers remember one place while looking at the next. Three maps keep the comparison in front of them.

Keep the geography comparable. The maps have the same study area and similar framing. The surrounding basemap gives context, while the square of data shows exactly which land contributes to the figures.

What I learned from Day 9

1. A boundary is part of the measure. A density figure is only meaningful when you know what land went into it. Around Greytown, changing the window changed the answer substantially.

2. Population and distribution are different questions. Greytown has more people than the Keate's Drift window, but those people are concentrated into fewer qualifying cells.

3. Keep the source resolution visible. A 1 km settlement class remains a 1 km classification when drawn on 100 m polygons. Smaller shapes do not mean more detailed evidence.

4. Start with the view that explains the theme. Settlement class provides the opening explanation. Population and building footprint add the detail, and the shared legend keeps the switches understandable.

5. Say what the units actually are. “People per 100 m cell” is clearer than “Population density” when that is what the colour bands show. The table can then use people per km² without ambiguity.

Final thoughts

City, town and countryside are useful starting points, but they leave a lot unsaid. Within the same amount of space, people can form a dense metropolitan landscape, a compact town or a network of smaller settlements.

The three views let readers examine those differences in turn: classification, population and building coverage. For planning or site-selection work, that is a starting point for better questions about where people are and how a service might reach them. Roads, travel times, facilities and local knowledge would be the next layers to add.

Explore the report by changing the map colour, and compare how much of each pattern survives the switch.

Sources and attribution

Methodology note: Each study window is a 6 km square in World Mollweide (ESRI:54009), centred on a reference point snapped to the nearest 100 m grid intersection. GHS-POP and GHS-BUILT-S values retain their native 100 m support. Each 100 m cell inherits the class of its parent 1 km GHS-SMOD cell, without adding classification detail. Population is summed; density divides population by window area; populated-cell share counts cells with at least one modelled resident; building-footprint share divides built-up square metres by total cell area. Results describe fixed windows, not town populations or official settlement boundaries. GHSL classification and population are methodologically related, so their agreement is not independent validation. These are three selected examples, not a representative sample of South Africa. Modelled data can contain anomalies; the results have been checked to the best of my knowledge.