Does no bird record mean no birds, or no birders?
Day 18's theme is NULL: the value that isn't there. Most maps quietly hide it. If you draw a map from records, a place with no records simply has no row, so nothing is drawn. The blank looks like an absence of the thing you're mapping, when it's really an absence of evidence.
Birds are a good way to show this. South Africa has one of the best citizen-science bird atlases in the world, the Southern African Bird Atlas Project 2 (SABAP2). Thousands of volunteer atlasers have submitted hundreds of thousands of checklists since 2007. Yet there are still squares nobody has surveyed. On a species map those squares look empty, and an empty square is easy to read as "no birds here".
So the report, No Records ≠ No Birds, asks a question any field-survey business will recognise: where is our evidence too thin to support confident conclusions, and where should the next survey effort go?
Explore the interactive report
The map shows every SABAP2 pentad in South Africa, coloured by how much survey evidence exists. Pentads are the atlas's 5-minute by 5-minute grid squares, about 9 km north to south and 7 km east to west in South Africa.
- Year slider: drag it to add up the cards from 2007 to the chosen year. In 2007 almost the whole country is orange; by 2026 the gaps have shrunk to scattered clusters. The cards and the map follow the slider; the chart below always shows all 20 years.
- Highlight buttons: No survey records (the default), Ad hoc only, Full-protocol coverage and Show all. The chosen status keeps its colour and everything else fades to pale grey, so you never lose the geography.
- Four headline cards: South African pentads, pentads with no full-protocol survey, the share of the country that leaves ("evidence gap remaining"), and how many pentads got their first full-protocol survey in the selected year.
- Hover a pentad: the tooltip mirrors the pop-up on the SABAP2 coverage site: pentad code, status, full-protocol cards in the selected year and in total, ad hoc cards and incidental records.
- Zoom in: from zoom 10 each square shows its pentad code, so you can look it up on the SABAP2 site.
Interactive report: Day 18, NULL
What the map does not show is birds. It doesn't say what lives in an unsurveyed square, whether it's good habitat, or whether it's easy to reach. It maps evidence, not wildlife.
Starting with SABAP2 coverage
SABAP2 is run by the FitzPatrick Institute of African Ornithology at the University of Cape Town, and its data is open through the BirdMap API. Its coverage endpoint returns one feature per surveyed pentad, with:
- Full-protocol cards: at least two hours of dedicated atlasing in the pentad, visiting as many habitats as possible and recording every species in the order it's seen. Each atlaser can submit one per pentad every five days. These are the backbone of the atlas, the cards that track change in bird communities, and the coverage data gives a count per year from 2007 to 2026.
- Ad hoc cards: any card where the atlaser didn't follow all the protocol rules, for example less than two hours, did not cover all habitat types or bad weather. Incidental records are one-off sightings. Both are useful, but they're undated in the coverage data: you get a total, not a year.
- Hours spent on full-protocol cards, and the pentad's quarter-degree code.
I first downloaded South Africa's coverage on 27 October: 15,282 surveyed pentads. When I checked one square against the live site a few weeks later, it showed 150 full-protocol cards to my 149. Atlasers keep submitting, sometimes late. So I scripted the download and refreshed it on 17 November. Three weeks had added 2,242 cards (387,685 in total) and given three ad-hoc-only pentads their first full-protocol card, but not one new square had been surveyed. That turned out to be the story in miniature.
Finding every square, including the empty ones
The coverage data only contains pentads with records. To show the NULLs, I needed the complete grid, then a join that keeps every square whether or not it has a match.
The Pentad Grid
SABAP2's own grid. The coverage page links to SABAP2's own pentad download, which assigns every pentad to exactly one country and province. Every one of the 15,282 surveyed pentads matched a square in the grid, which is the check you want.
Squares that aren't really South African
SABAP2's South Africa grid has 16,688 pentads. Seventeen of them carry the country name where the province should be: two squares in Mozambique east of Komatipoort, eight just west of the 20th meridian on the Namibian border, and seven coastal squares with only a sliver of land. Fifteen of those were never surveyed and don't appear on SABAP2's own South Africa coverage map, so I left them out (16,673 remain). The other two have ad hoc records from the shore, at Port Nolloth and Plettenberg Bay, so they stay, given their Northern Cape and Western Cape provinces.
Keeping NULL as NULL
The processing is a DuckDB SQL script: start from the grid, LEFT JOIN the coverage, and leave the unmatched squares with NULL card counts, not zeros. That distinction matters. Zero says "we looked and found nothing". NULL says "we never looked". The SABAP2 protocol makes the same point from the other side: a full-protocol card with only a few species is a valid record if the rules were followed, because it suggests the area really is largely devoid of birds. You can only learn that a place has few birds by surveying it. The model keeps them NULL all the way to the tooltip, which reads "No SABAP2 survey records - no data, not no birds".
Undated ad hoc records
The trickiest decision was about time. Full-protocol cards have years, so I can say exactly when a square first got one. Ad hoc records don't. A square with only ad hoc records might have had them in 2009 or last week.
My first version treated "no survey records by 2015" and "ad hoc only (all years)" as separate classes. The fix was one definition for every year: the evidence gap is every pentad with no full-protocol survey by the selected year, whether or not it has ad hoc records. The map and chart still split it into orange (no records at all) and lilac (ad hoc only), and the legend says plainly that ad hoc is undated.
Drawing the gaps in Icon Map Pro
Each pentad is a WKT polygon built from its code, which encodes its top-left corner. Icon Map Pro draws all 16,673 squares as one WKT layer, and the styling comes from measures bound to the layer:
[Effort class]works out each square's class from its cumulative cards up to the slider year.[Fill colour],[Outline colour]and[Outline width]turn that class into the data-gaps palette, and read the highlight buttons. The chosen status keeps its colour; everything else becomes pale grey.[Show on map]in the visual-level filter keeps the layer tidy.
A slider that adds up, not one that filters
A year filter would be wrong here. Pick 2015 and a square surveyed only in 2014 would turn blank again. The year slider is Power BI's less than or equal to slider over a small disconnected years table, and the measures read the highest year left. So 2015 means "all cards from 2007 to 2015", and a square never goes blank once it's been surveyed.
Testing PMTiles against WKT
Sixteen thousand WKT polygons is a lot of text to push through a Power BI model. As an experiment I also wrote the grid as vector tiles: GDAL, inside DuckDB, writes a 1.8 MB PMTiles file with just the pentad code on each square. Icon Map Pro's vector tile layer can join report rows to those tiles and colour them with the same measures. It needs hosting over HTTPS before Power BI can read it, so the speed comparison will have to wait for another day.
Designing for missing data
My first version looked like a survey-coverage dashboard: a mostly blue map, a chart of cards per year, and a few grey squares. It was accurate, but it told the wrong story. The eye went to where surveys had happened.
So I reversed the emphasis:
- Orange is the only strong colour, and it means "no survey records". The four classes of survey effort became a pale blue-grey ramp. You can still see effort, but it sits in the background.
- Lilac marks ad hoc only: observations, but no full-protocol survey.
- The chart tracks the gap, not the activity. A stacked column per year shows pentads with no full-protocol survey, orange for no records and lilac for ad hoc, falling from 15,748 in 2007 to 1,932.
- One new card does a lot of work: "First full-protocol survey in 2023: 92". It measures how far coverage is spreading, not how many checklists were submitted.
What the map reveals
The gaps are concentrated. The Northern Cape holds 1,026 of the 1,391 never-surveyed pentads, 74% of them, and one in five Northern Cape squares has no survey records at all. The Eastern Cape interior around Mthatha is the next cluster. Gauteng, the Free State and Mpumalanga have none.
The gap is closing, slowly. In 2009 atlasers reached 2,541 pentads for the first time. Since 2023 it has been about 100 a year: 92, 113, 129, and 47 so far in 2026. At that pace the remaining 1,932 squares without a full-protocol survey would take well over a decade.
Effort goes back to places we already know. Around 22,000 full-protocol cards a year are submitted, yet only about 100 squares a year are new. The busiest pentad, on the West Rand in Gauteng, has 4,141 cards. The top 10% of surveyed squares hold 74% of all full-protocol cards, while the median surveyed square has five.
Fewer records may mean we need more observers, not that there are fewer birds. That's the point of the map. The empty Northern Cape is semi-arid, sparsely populated and a long drive from the big cities. Those are reasons for few atlasers, not few birds.
The map shows survey evidence. It doesn't prove anything about bird populations, and the effort classes describe survey effort, not statistical confidence.
What I learned from Day 18
- NULL is not zero. "We looked and found nothing" and "we never looked" need different values, different colours and different words. Keep NULLs as NULL all the way to the tooltip.
- You can't map what isn't there from records alone. Start from the complete grid and join the records onto it, or the gaps simply disappear.
- Check whose grid you're using. A grid that doesn't match the data's own country assignment invents gaps. Using SABAP2's grid made every surveyed square match.
- Undated data needs one honest definition. When part of the data has no date, don't build two time frames that quietly disagree. Pick one definition that holds for every year, and say what it includes.
- Emphasis is a design decision. The same data told a survey-activity story in blue and a missing-data story in orange.
Final thoughts
Every field dataset has its blank squares: inspections never made, customers never visited, clinics that didn't report. On a dashboard they look like zero activity, and decisions get made on that. The pattern from Day 18 works anywhere: build the complete grid of where you could have data, join what you have, keep the gaps as NULL, and give them the strongest colour on the map. Icon Map Pro makes it easy to draw every square, colour it from measures, and let the reader watch the gaps close.
Sources and attribution
- SABAP2 / BirdMap: pentad coverage for South Africa (downloaded 27 October 2026, refreshed 17 November 2026) and the SABAP2 pentad grid for South Africa (downloaded 16 November 2026), from the BirdMap API, https://api.birdmap.africa/sabap2/v2/. FitzPatrick Institute of African Ornithology, University of Cape Town. Licence MIT, per the API metadata. With thanks to the many citizen-science atlasers whose cards make the atlas. Coverage map: https://sabap2.birdmap.africa/coverage
- SABAP2 protocol: Southern African Bird Atlas Project (SABAP): The Protocol Explained, updated 9 March 2021. FitzPatrick Institute of African Ornithology, University of Cape Town. https://sabap2.birdmap.africa/
- Citation: Brooks M, Rose S, Altwegg R, Lee ATK, Nel H, Ottosson U, Retief E, Reynolds C, Ryan PG, Shema S, Tende T, Underhill LG & Thomson RL (2022). The African Bird Atlas Project: a description of the project and BirdMap data-collection protocol. Ostrich. DOI 10.2989/00306525.2022.2125097
Methodology note: The grid is SABAP2's own pentad grid for South Africa (16,688 pentads), less 15 unsurveyed edge squares that SABAP2 gives no province and leaves off its South Africa coverage map (16,673 remain). Port Nolloth (2920_1650) and Plettenberg Bay (3405_2305) were given their provinces by hand. Coverage is joined by pentad code; pentads with no coverage keep NULL card counts. Effort classes use full-protocol cards only: never surveyed, ad hoc only, 1, 2-4, 5-49 and 50+ cards. They describe survey effort, not confidence. The year view adds up full-protocol cards from 2007 to the selected year. Ad hoc and incidental records are undated in the coverage data, so ad-hoc-only pentads are counted in every year, and the evidence gap is every pentad with no full-protocol survey by the selected year. In 13 pentads the yearly card counts don't add up to the total in the source; totals are used for the classes. 2026 is a partial year, to 16 November.