Where did South Africa burn during 2025, and how did the geography of burning change throughout the year?
For Day 3 of the 30 Day Map Challenge, the theme is Polygons. After exploring population and supermarket locations for Day 1's Points map, and South Africa's electricity network for Day 2's Lines map, today is investigating something that changes geographically throughout the year.
Fire seemed an interesting subject, particularly given South Africa's contrasting landscapes and seasonal weather patterns.
Instead of mapping individual fire detections as points, I wanted to show the actual areas of land identified by satellites as having burned. That meant starting with satellite raster data, converting it into polygons, and making the resulting features interactive in Microsoft Power BI.
The final report uses Icon Map Pro, with burned-area polygons stored in PMTiles and linked to analytical data in Power BI. An animated month selector reveals how the mapped pattern changes throughout 2025.
Explore the interactive report
Use the playback controls to move through the months of 2025, or select a month directly from the chart. The map updates to show the burned areas detected during that period.
You can also select a province to explore its burned areas and compare its contribution to the national totals.
Interactive report: A Year of Fire. Use the playback controls to see how satellite-detected burned areas change month by month.
An important distinction: This map represents satellite-detected burned land, not individual wildfire incidents. The data can include uncontrolled wildfires, prescribed burning and agricultural fires.
Starting with NASA's burned-area data
NASA's Fire Information for Resource Management System (FIRMS) provides access to satellite-derived information about fire activity worldwide.
There are two related but distinct types of information: active fire detections, which identify locations where satellites detect thermal anomalies, and burned-area products, which estimate land that has been affected by burning.
For this challenge, the second was more appropriate.
MODIS MCD64A1 Version 6.1, the monthly burned-area product derived from observations by NASA's Terra and Aqua satellites was used for this report.
The product has a nominal spatial resolution of approximately 500 metres and includes a Burn Date layer identifying the estimated day of the year when burning occurred within each mapped pixel.
Downloading a year of data
I used NASA's free AppEEARS service to extract data for the South African mainland.
AppEEARS returned 36 GeoTIFF files, covering the three requested layers across 12 months for South Africa.
The total extracted raster data amounted to approximately 9.45 MB. This was a useful reminder that a relatively small satellite dataset can still contain considerable geographical and temporal information.
From raster pixels to polygons
The MODIS product is raster data, but the Day 3 challenge is about polygons.
To transform the data, I processed each month's Burn Date raster and identified pixels containing valid burned-area detections. These pixels were converted into polygon geometries representing contiguous patches of burned land.
Although the original satellite data is comparatively coarse at approximately 500 m resolution, the resulting polygons preserve much more spatial information than simply representing each burned area with a point or a provincial total. They show the distribution and extent of satellite-detected burning, including the fragmented patterns found across grassland, savanna, agricultural and other landscapes.
Dealing with polygon geometry
Raster-to-polygon conversion introduces some interesting geometric issues. Pixels can meet along an edge or touch only at a corner. Combining these incorrectly can produce invalid polygon geometries, including shapes that touch themselves.
The polygon-generation method was adjusted to avoid joining patches solely through corner contact, and the resulting geometries were validated before continuing. This produced 9,439 valid burned-area polygons across the 12 months. Rather than dissolving every burned patch into one enormous geometry, I retained the individual features. That allows the map to reveal the geographical distribution of burning when users zoom in.
Calculating burned area
The source data was exported in geographic coordinates, EPSG:4326. Because coordinates in degrees cannot be used directly as square metres, the processing included appropriate geographic area calculations to estimate the area of the burned polygons in square kilometres. The original processed dataset represented approximately 26,097 km² of satellite-detected burned area during 2025.
These are estimates derived from the MODIS product and subsequent processing, rather than exact surveyed fire boundaries.
Adding a provincial dimension
A national map provides a useful overview, but I also wanted to understand how burning varied across South Africa's nine provinces. The burned-area polygons were spatially intersected with the nine provincial geometries. Where a burned polygon crossed a provincial boundary, it was split into separate pieces and its area recalculated.
The final province-intersected dataset contained:
- 9,621 polygon parts
- 26,054 km² of province-assigned burned area
Keeping polygon geometry out of Power BI
One of the useful lessons from Day 2 was that detailed spatial geometry does not necessarily need to be stored directly in a Power BI semantic model.
The final dataset was divided into two components.
1. Spatial geometry PMTiles provides a compact, portable way to store and access vector tiles, making it suitable for displaying detailed geographical features at multiple zoom levels.
2. Analytical attributes A separate CSV contained one row for each polygon part, including its identifier, province, month and burned area, which was imported into Power BI. A unique identifier was used to link the data rows to the corresponding polygon geometries in the PMTiles file. As with Day 2, I used the newer Shapes from Files functionality in Icon Map Pro to import the PMTiles archive directly into the Power BI report.
Designing the map to show seasonality
My first map displayed all the burned-area polygons in a single colour. This made the overall distribution easy to see, but it hid one of the most interesting aspects of the dataset: when burning occurred so changed the styling so that each month had its own colour.
Rather than using 12 unrelated colours, I chose a seasonal progression. The earlier months use greens and yellows, while June through October move into orange, burnt orange and dark red tones. November and December use purple and dark grey to distinguish them from the main winter burning period.
These colours are deliberately categorical. They represent the month of detection, not fire temperature, intensity, severity or vegetation damage.
Using the monthly chart as a legend
A monthly burned-area chart was placed alongside the map and assigned each column the same colour as its corresponding polygons. This gives the chart two purposes.
First, it explains the colour scheme used on the map. Second, it shows the total area burned in each month, helping the reader understand the relative scale of burning throughout the year.
The chart is also interactive: selecting a month filters the map and other report visuals. A separate provincial bar chart uses a neutral grey to avoid competing with the month colours.
Bringing the year to life with animation
The final addition was a Play Axis custom visual. The visual steps through the months of 2025, filtering the report as the animation progresses.
As each month changes, the map redraws the corresponding burned-area polygons, the national burned-area KPI updates, and the provincial chart shows the distribution for that period. This turns the report from a static map of annual burning into a geographical timeline.
The effect is particularly interesting when moving from the relatively limited burning early in the year into the much larger areas identified during the winter months. It also reveals that South Africa does not have one uniform fire season.
Different regions exhibit different seasonal patterns, which would be much harder to see in a single annual summary.
What does the map reveal?
The processed 2025 MODIS data shows a pronounced seasonal concentration of burned area.
August was the peak month, with approximately 7,586 km² of province-assigned burned area. July and September were also substantial, with approximately 4,542 km² and 4,697 km² respectively. Together, these three months accounted for nearly 65% of the year's province-assigned burned area. This pattern is especially visible across the eastern and north-eastern parts of South Africa.
In contrast, much of the Western Cape follows a different seasonal pattern, reflecting the geographical and climatic variation across the country. The changing colours make these differences visible without requiring viewers to interpret a large table.
Which provinces had the most burned area?
Mpumalanga records the largest total mapped burned area in this dataset, followed by KwaZulu-Natal and Limpopo. However, these are absolute area totals, not the proportion of each province that burned. A province with a larger geographic area or different vegetation cover should not automatically be interpreted as having a greater fire risk based on these figures alone.
The map is descriptive rather than a fire-risk assessment.
What I learned from Day 3
This was a different kind of mapping challenge from the first two days.
1. Raster data can become useful interactive vector data. The original MODIS burned-area data was a series of monthly rasters. Converting it into polygons made it possible to explore individual burned patches, apply provincial filters and connect the results to Power BI measures.
2. Spatial intersections add analytical value. Splitting burned-area polygons at provincial boundaries made it possible to report provincial totals without assigning entire cross-border polygons to one province.
3. Geometry validation matters. Satellite-derived raster polygons can contain awkward shapes and corner-touching pixels. Producing valid polygons before converting them into vector tiles avoids problems later in the workflow.
4. Keep geometry and attributes separate. Storing detailed shapes in PMTiles while keeping their analytical attributes in Power BI provides a practical approach for reports containing thousands of geographical features.
5. Animation should communicate change. The Play Axis adds value because the timing and geographical distribution of burning are central to the story. It is not simply movement for visual effect.
Final thoughts
A polygon is more than an area on a map. It can represent a changing landscape, a measured extent or a boundary used to connect geographical information to other data. For Day 3, the polygons started as satellite raster pixels and became an interactive record of estimated burned areas across South Africa during 2025.
Combining NASA's burned-area data with provincial boundaries, PMTiles and Icon Map Pro allowed me to build a Power BI report that explores not only where land burned, but when. The result demonstrates how satellite observations can be turned into accessible, interactive geographical analysis without storing detailed polygon geometries directly in Power BI.
Sources and attribution
- Burned-area data: Giglio, L., Justice, C., Boschetti, L., & Roy, D. (2021). MODIS/Terra+Aqua Burned Area Monthly L3 Global 500m SIN Grid V061. NASA Land Processes Distributed Active Archive Center (LP DAAC). https://doi.org/10.5067/MODIS/MCD64A1.061. Accessed 2 November 2026.
- Data extraction: NASA AppEEARS, Version 3.134. https://appeears.earthdatacloud.nasa.gov/.
- Provincial boundaries: © OpenStreetMap contributors. Licensed under the Open Database Licence (ODbL).
Methodology note: The mapped areas represent MODIS satellite-detected burned land at approximately 500 m resolution, including potentially prescribed and agricultural burning. They are not confirmed individual wildfire footprints. Burned-area estimates are derived from processed satellite observations and provincial polygon intersections. The figures should not be interpreted as official fire incident counts, fire severity measurements or comprehensive assessments of fire damage.