What we’ve fixed since launch, and why

On 12 July 2026, we published the Australian Zoning Atlas on data version v1.3.0.

Since then we’ve shipped a number of further data releases, each with its own version number—most recently v1.9.2. These new data versions carry a few minor but nonetheless substantive corrections, a routine data-currency refresh across every city we cover, and—occasionally—a real change in the law.

The corrections don’t change the headline story: every data-quality fix below is a fraction of a percentage point, even added together. Real policy change is different. When a council or state government actually rezones land, the number is supposed to move, and it does—Melbourne’s Amendment GC285 (v1.9.0) is the clearest example yet, easing that city’s highly-restricted share by 1.4 points in a single release.

What this post is really about is how we got there: a public dataset making a specific, checkable claim about zoning attracts scrutiny, and scrutiny is how you find out you were wrong about something. Every fix below started with someone looking hard at the map and asking a question we couldn’t immediately answer. Here’s what we found, in order.

Dataset changes

v1.3.1—Ensuring that Sydney’s Height-of-Building layer uses the more permissive overlapping control

Sydney’s LEP Height-of-Building layer occasionally has two height polygons stacked on the same block—usually a precinct amendment sitting on top of the base LEP map. We’d been resolving that overlap by taking the lower of the two heights, rather than the higher. This aligns with our principle of systemically underestimating restrictiveness.

The fix touches around 1.85 km² across Sydney, so the city-wide restrictiveness number ticks by less than a hundredth of a percentage point. Inside the CBD, where precinct stacking is common, some individual rings moved by up to 0.2 percentage points.

Thank you to Dave Black who highlighted this issue over email immediately as we launched.

v1.4.0—Correctly classify R2 land inside Transport Oriented Development precincts

The TOD program lets some R2 (low-density) blocks near stations build up to 22 metres. Our pipeline was still counting those parcels as “low-density” for the headline restriction number even after applying the height increase.

We carved out the roughly 3,900 R2 parcels that sit inside an operative TOD site (1.68 km² of Sydney’s 728 km² of low-density land) from the low-density flag.

This is a small slice of the map. Sydney’s restriction-union headline eased by 0.4 percentage points. The result of this change is that Sydney’s headline figure is now 81.0% exactly; previously, we were rounding down.

Thank you to Eamon Waterford and others for highlighting the need for this correction.

v1.5.0 and v1.6.0—Correctly classifying Brisbane’s Priority Development Area land

Brisbane’s Priority Development Areas—state-declared urban renewal zones like Northshore Hamilton and Bowen Hills—override the normal zoning.

We had classified PDA land as permitting residential development, regardless of the underlying zoning. But this meant that we had tagged some open space and special-purpose land as permissive of residential development, such as the Royal Queensland Golf Club.

We checked how much of Brisbane’s PDA land had this problem. Underneath the PDA overlay, about a quarter of the area has non-residential base zoning—mostly open space, sports grounds, and community facilities. We clipped the PDA re-add against those base zones in two steps: recreation and open space first (v1.5.0), and then community facilities and special-purpose land (v1.6.0).

Land with a non-residential base zone is now correctly excluded from the Atlas’s metrics. As a result, Brisbane’s residential PDA footprint shrank from 7.3 km² to 4.67 km².

Because removing wrongly-unrestricted land pushes the total restricted share up, Brisbane’s highly-restricted figure increased by about 0.2 percentage points across the two releases.

An unfixed data gap: Local Plans in Brisbane’s Statistical Urban Area, but outside the 20 km headline radius

An additional issue identified with the Atlas: North Lakes, a master-planned community in Moreton Bay, is almost entirely missing from the dataset: 95% of its ~8,700 residential lots are unclassified. This is because the area is governed by a Local Plan document that is not in any shapefile we have access to.

Because the area sits outside of the 20 km radius we use to calculate our headline density, we did not manually code the data. We are unable to efficiently code every plan held in a local council PDF, and so this area is excluded from our dataset. This is a limitation in how planning and spatial data is communicated with and made accessible to the public.

Thank you to Geoff Anstey for highlighting the need for these corrections.

v1.7.0—adding Floor Space Ratio for Sydney

Rather than fixing a measure, this release adds a new one: Floor Space Ratio (FSR), which is used across Sydney to restrict the density permitted to be built on a given site. Built-form controls such as height are independent of FSR: a given lot can have a generous height cap but still be highly restricted if the FSR is low enough.

Because no cities other than Sydney use FSRs broadly, we have not at this stage displayed this metric on the site, as inter-jurisdictional comparison is not possible. However, it is worth noting that including FSRs increases Sydney’s highly restrictive headline figure regardless of the threshold considered ‘highly restrictive’:

Metric	FSR ≤ 0.5	FSR ≤ 0.75	FSR ≤ 1.0	FSR ≤ 1.25	FSR ≤ 1.5
Highly restricted by FSR	45.9%	65.5%	71.0%	72.4%	74.3%
Highly restricted (current w/ FSR)	82.2%	85.2%	89.3%	90.2%	91.6%
Increase over current (81.0%)	+1.2 pp	+4.2 pp	+8.3 pp	+9.2 pp	+10.6 pp

These increases reflect the fact that FSR only binds if it is more restrictive than the built form controls it accompanies. Otherwise, it does not show up in the data. This is representative of the general problem with FSRs, which is that they only regulate externalities indirectly, and when combined with built form controls can therefore only be one of two things: restrictive and binding, or permissive and irrelevant.

Thank you to Reg Wade for highlighting the importance of considering FSR controls at various thresholds.

v1.8.0—a routine data-currency refresh across every city

The corrections above all started the same way: someone looked hard at the map and found something we’d got wrong. This release started differently—with an actual change in the law.

South Australia’s Planning & Design Code was amended in late June, and our local copy of Adelaide’s zoning had gone stale. Rather than refresh Adelaide alone, we used the occasion to check every city’s zoning, height, and heritage sources against the state and territory governments’ live services—a housekeeping pass we intend to repeat periodically. Of the six cities that moved as a result, only two—Adelaide and Melbourne—reflect a real change in the law; the rest (Canberra, Hobart, Perth, Sydney) moved because our own data had fallen behind an already-existing reality, the same kind of correction as everything above. Parcel and cadastre boundaries were left alone; this is a currency check on planning controls, not a re-survey of land.

Every city’s highly-restricted share within 20 km moved by less than six-tenths of a percentage point, and most moved by far less:

• Adelaide eases from 91.95% to 91.84%. About 40 zone polygons had shifted since our last snapshot, spread across nine separate Code Amendments since October 2025—ordinary ongoing policy changes, not one event.
• Brisbane is essentially unchanged.
• Canberra shifted slightly—from 74.42% to 74.36%—on a refresh of Territory Plan zones, road reserves, and block geometry.
• Darwin is unchanged.
• Hobart eases from 97.29% to 97.01%. We added Hobart’s Commercial Zone Specific Area Plan, which permits residential use up to 15 metres on Commercial-zoned land within its boundary; that land had previously been excluded from total residential land.
• Melbourne eases from 44.76% to 44.41%. The refresh surfaced a zone type new to our dataset, the Precinct Zone, used across the Suburban Rail Loop and encompassing parts of Whitehorse, Monash, Kingston and Bayside (about 15 km²). It now counts toward Melbourne’s residential land total, bucketed as mixed-use with no published height limit—which brings Notting Hill, Victoria back into our per-suburb tables after it had been silently dropped.
• Perth eases from 87.06% to 86.47%, the largest move this release—driven mainly by refreshed density-zoning and height data now reading more land as permitting more than two storeys than our previous copy did. Heritage listings grew too (Heritage Areas nearly doubled in feature count, 282 to 467, as the state’s listing programme continues), but that’s a side note, not the driver: Perth’s heritage share barely moved either way.
• Sydney eases from 81.00% to 80.89%. Zoning, height, floor space ratio, and both heritage layers all showed only minor drift—under 0.3%—since our last fetch.

v1.9.0—a real Melbourne upzoning, not a correction

Every release above is us fixing something. This one is the Victorian Government changing something.

On the day this data was published, the Victorian Government gazetted Amendment GC285—Stage 2 of the Train and Tram Zone Activity Centres Program. It rezones nine activity-centre clusters across six councils (Boroondara, Glen Eira, Greater Dandenong, Kingston, Stonnington and Whitehorse), including the Chapel Street precinct around Prahran and South Yarra, into a housing-choice-and-transport zone with a materially more generous height allowance than whatever applied in each catchment before.

We picked this up the same day it went live. Melbourne’s highly-restricted share within 20km eases from 44.41% to 42.99%—a 1.4 percentage-point drop in one release, the largest single-city move in this post. Land inside these nine clusters that used to be capped at two storeys or zoned low-density-only now correctly reads as open under the new zone. Melbourne’s total residential footprint is essentially unchanged (the rezoning doesn’t add or remove land from our count, it changes what’s permitted on land already counted); nothing else about the city’s data changed, and no other city moved at all.

This is what the Atlas is supposed to do when a government actually loosens zoning: the number moves, visibly, on the day it happens. It’s the same story as the ACT’s Missing Middle reform—real change reads as a real change in the data, not something we need to go looking for.

v1.9.1–v1.9.2—fixing Melbourne’s detailed zoning map

Separate from the rezoning above, we found and fixed two problems with Melbourne’s detailed zoning map—the interactive map beneath the city’s overview, not the headline number.

The map had gone stale: it was still being built from an old, disconnected copy of Melbourne’s zoning boundaries, so recent rezonings—including GC285 above—weren’t visible on the map even though they were already counted correctly in the underlying statistics. We repointed it at the same current zoning data the statistics use, which took it from 12 coarse zone groups to 176 real zone polygons across 105 distinct zone codes.

That repointing surfaced a second, previously-hidden gap: we didn’t yet have a density classification for most of those 105 codes, so the map rendered as mostly flat grey instead of the zone-by-zone colour ladder every other city gets. We’ve now classified all of them, so Melbourne’s map reads the same way as the rest of the Atlas. Like the classification fix itself, this is a map-colour completion only—it doesn’t touch the highly-restricted/open split behind the headline numbers, so no city’s figures move.

Changes overview

Every correction above, at a glance—each release, the city it touched, how much it moved the 20 km headline, and why:

The Atlas will continue to improve over time

Spatial data is messy, complicated, and inconsistent. Managing these datasets across eight different cities has been a mammoth undertaking by a small team of three. We have been pleased to see that the errors shipped at launch—and now corrected—were overall minor, and did not move any of our numbers meaningfully. This is thanks to an embedded conservative approach: our methodology assumes planning controls are permissive until proven restrictive.

Thank you again to all those who gave feedback. Part of the reason for providing the maps is to make it easier to tell if part of our model is getting something wrong. If you spot something that looks wrong on the map, email us. The fixes above exist because people did exactly that.