How Bad Were the 2026 Western Cape Wildfires? The View From Space

The 2025/26 fire season was, by any measure, a severe one for the Western Cape: major fires at Franschhoek and along the Overberg coast in January, hundreds of smaller fires across the province, and months of sustained pressure on firefighting services. It also left an accounting puzzle. In January 2026 the provincial government reported 132,000 hectares burned. In April, a parliamentary reply on agricultural losses listed district totals that add up to 168,798 hectares, which is 28 percent more than the whole-province figure, from the same government, for the same season.

Numbers like these can differ for good reasons, and part of the purpose of this article is to explain those reasons. But the differences also raise a fair question: how much of the province actually burned? To answer it independently of any official count, we measured the season directly from satellite data, downloaded from NASA and processed by us. Every map and chart below is generated from that data, and the maps are interactive, so you can check the results against places you know.

The season from orbit

The map below shows every patch of the Western Cape that NASA's burned-area product detected as burned during the season, drawn in red over a road map so that the location and true size of each fire can be judged against familiar towns and distances. Use the tabs to compare seasons, drag to pan, and use the + / − buttons (or pinch on mobile) to zoom in. The basemap switch in the corner also offers a detailed street map and satellite imagery.

Flip between the tabs and the geography stays remarkably consistent. The two largest seasons, 2016/17, the drought-year record, and 2025/26, this season, both light up the same fire-prone belt: the winter-rainfall west and the southern coast, where fynbos and renosterveld carry fire readily in a dry summer. Switch to 2022/23, the smallest season on record at roughly a tenth of this season's area, and that same belt is nearly empty. The arid Karoo interior in the north-east barely burns in any year. Zoom into the Boland mountains behind Franschhoek, or the Overberg coast near Pearly Beach, and you can trace individual fire scars against the towns and roads.

For scale: a hectare is a square of 100 by 100 metres, roughly a rugby field. 312,929 of them is a little over 3,100 square kilometres, an area about one and a quarter times the size of greater Cape Town, burned in a single summer. It amounts to 2.4 percent of the province's total land area.

How a satellite counts burned land

A quick explainer, because the whole article rests on it. NASA operates two satellites, Terra and Aqua, that photograph the entire Earth every day with an instrument called MODIS. Scientists at the University of Maryland turn those daily images into a monthly map called MCD64A1: for every 500 by 500 metre square of the planet, it records whether that square burned that month, and on which day. It does this by looking for the distinctive way burned ground reflects light, confirmed against the thermal glow of the fire itself. It is the standard global burned-area record used in fire science, and it goes back to November 2000.

Our measurement is then simple to describe. We took every monthly map from November 2025 to April 2026, clipped it to the Western Cape's official boundary, and counted every pixel that burned at least once during the season. Each pixel is 21.5 hectares. Count the pixels, multiply, and you have the season's burned area: no self-reporting, no arguments about what counts as a "wildfire", just scorched earth as seen from space. The same recipe, applied to every season since 2001/02, gives a like-for-like history to compare against.

Official figures vs the satellite

Here is every official figure we could source for the season, alongside the satellite measurement. They are not all measuring the same thing, and that is precisely the point: the ranges tell you what each count includes and leaves out.

Read together, the numbers tell a consistent story. The 132,000 ha headline was announced in late January, with two and a half months of the season still to run, so it could never have been a final figure. The district sum of 168,798 ha covers the full season but is an agricultural loss record: it reflects what farmers reported losing, not what burned on conservation land, municipal land, or anywhere no claim was filed. The satellite measurement includes all land regardless of ownership or reporting, which is why it is larger than both. On this evidence, the two official figures each describe a genuine subset of the season, and the full extent, about 313,000 hectares, is roughly double the figure most widely quoted.

When it burned: month by month

The season had a clear shape. It started early: November fires already accounted for more than 51,000 hectares, and December added a further 73,000. The peak came in January 2026, when 127,572 hectares burned in a single month. That is roughly 41 percent of the entire season, and nearly as much as the government's whole-season headline figure. The Franschhoek mountains, the Overberg coast at Pearly Beach and Stanford, and dozens of smaller fires were burning in the same weeks, which explains the exceptional simultaneous demand on firefighting resources that month.

2025/26 against 25 years of fires

One season in isolation tells you little; the value of a satellite record is that it applies the same method to every season back to 2001/02. On that consistent measure, 2025/26 produced the second-largest burned area in a quarter of a century. Only 2016/17, at the height of the multi-year drought, burned more, at 339,204 hectares. This season's total was 78 percent above the 24-season average, and a third higher than last season, which was itself well above average. You can compare the actual burn patterns of these years directly on the interactive map above.

Two details in that chart are worth noting. The last three seasons are all above average, and two of the record's five largest seasons have occurred in the past three years. A 25-season record is too short to establish a trend on its own, but the pattern is consistent with projections that Western Cape fire seasons will lengthen and intensify as the climate warms, and it is worth watching in the seasons ahead.

What kind of land burned

A fair challenge to any satellite total is: how much of that is just farmers burning stubble? Agricultural burns are routine, intentional and mostly harmless, and if they made up the bulk of the 312,929 hectares, the big number would be misleading. Because the district figures came from an agricultural-losses process, it is tempting to assume farm burning explains the whole gap.

So we tested it, overlaying every burned pixel on the European Space Agency's WorldCover map, which classifies the land surface at 10 metre resolution. The result is unambiguous:

85 percent of everything that burned was natural vegetation. Half the total was grassland, and nearly a third was shrubland, which in the Western Cape means fynbos and renosterveld. Cultivated cropland accounts for just 14 percent. This was a season of veld fire, not field fire.

That finding also explains the paradox in the official numbers. How can "agricultural losses" total 168,798 hectares when only about 44,000 hectares of actual cropland burned? Because most agricultural land in the province is not cropland at all: it is grazing veld. When a fire runs through a sheep farm's rangeland, the farmer reports an agricultural loss, and the satellite sees natural grassland burning. Both are right. The loss ledgers and the land-cover map are describing the same fires from different angles.

Three big fires, up close

The MODIS pixels are 500 metres wide, good for provincial totals but coarse for a single fire. So for three of the season's most talked-about fires we went to the European Sentinel-2 satellites, which see at 10 to 20 metre resolution. The technique, called dNBR (differenced Normalised Burn Ratio), compares how the vegetation reflected infrared light before and after the fire. Each map below shows the precise burn scar over the real terrain, so you can zoom in and see which farms, valleys and coastline actually burned.

The first two were exceptionally large fires by any historical standard: Franschhoek and Pearly Beach-Stanford together account for roughly a fifth of the province's entire season. Zoom into either map and you can follow the burn around specific valleys, farms and stretches of coastline. Mossel Bay is instructive in the opposite direction. It received days of news coverage, yet the measured scar is 670 hectares. Raw satellite hotspot detections, the kind that power live fire maps, suggested a footprint more than twice that size, because heat and smoke register more widely than the flames themselves. Careful before-and-after imaging is what separates perceived extent from burned area.

What the satellite cannot see

Every measurement has edges, and it matters to state them. The burned-area product misses fires smaller than roughly 100 hectares, so hundreds of small veld fires are invisible in our total, meaning the true figure is, if anything, slightly higher still. It also cannot see fires that burn buildings rather than vegetation. The Dunoon informal settlement fire, which destroyed homes and dominated the human cost of this season, does not register a single pixel. Its absence from our totals is a limitation of the instrument, not a statement about its importance. Satellites measure scorched land; they do not measure the loss of a home.

The bottom line

How bad were the 2026 Western Cape wildfires? Measured the same way for every season since 2001, this was the second-largest burned area in twenty-five years: 312,929 hectares, nearly two and a half times the widely quoted headline figure and 78 percent above the long-term average. The burning came from hundreds of separate fires, peaked sharply in January, and consumed mostly natural veld rather than cultivated farmland. The official numbers were not wrong so much as partial: each counted the slice of the season its record-keeping was designed to capture. The satellite record, which counts every hectare the same way, shows a season roughly twice as large as the headline figure suggested, severe by historical standards, though not unprecedented.

Data & methodology

Burned area. NASA MCD64A1 Collection 6.1 monthly burned-area product, 500 m resolution, on the native MODIS sinusoidal grid so every pixel is exactly 21.466 ha. A season is November to April; a pixel counts as burned if it burned in any month, counted once. Historical seasons (2001/02 to 2024/25) were read from the Microsoft Planetary Computer archive; because that archive currently ends at July 2025, the 2025/26 season was computed from the identical product downloaded directly from NASA's LP DAAC. The two sources agree to within 0.2 percent on an overlapping season. The provincial boundary is GADM 4.1; the pixel geometry reproduces the official provincial area to within 0.1 percent. Monthly totals in the timeline chart sum to slightly more than the season union (313,982 vs 312,929 ha) because a small number of pixels burned in two separate months.

Fire events and dates. VIIRS active-fire detections (Suomi-NPP and NOAA-20, standard science quality) from NASA FIRMS, clustered into 673 distinct events with a space-time algorithm (DBSCAN). Event dates for the three case studies come from these clusters.

Case-study severity. Sentinel-2 L2A imagery, cloud-masked median composites before and after each fire, differenced Normalised Burn Ratio (dNBR) at 20 m with standard Key & Benson thresholds; burn perimeters mapped at dNBR ≥ 0.10 with sub-hectare speckle removed. Post-fire windows begin two weeks after each fire's last active detection.

Land cover. ESA WorldCover 2021 (10 m), majority class per burned MODIS pixel. Land cover describes vegetation, not tenure: grazed natural veld on farmland is classed as grassland or shrubland, not cropland.

Interactive maps. Burn scars and fire perimeters are vectorised from the products above and drawn with Leaflet over OpenStreetMap / CARTO and Esri basemaps for geographic reference.