Every modelling choice, with its direction of bias

The Australian Zoning Atlas is underpinned by dozens of assumptions and interpretive choices.

This includes things as simple as where each city ends, how a height in metres becomes a count of storeys, which zones count as rural, and when a given bus stop counts as “rapid”. This page lists each of these choices in plain language, grouped by theme.

The Atlas’s restrictiveness estimates are conservative

The whole Australian Zoning Atlas pipeline is biased toward overstating permissiveness. Every ambiguous call is resolved so that land looks less restricted.

Three choices carry most of this:

• Unknown height limits never count as restricted. Land with no published height limit stays in the denominator (it is residential) but is never added to the 2-storey numerator. Zones with no fixed cap are treated as high-density. We never claim a limit we cannot see.
• The representative-point overlay test is strict. A parcel is flagged as being under a control such as a heritage overlay only when its interior point falls inside a heritage polygon, rather than when the lot merely touches an overlay edge. This means that we likely underestimate heritage on the margins.
• Rural and fringe land is excluded. Rural-zoned land, even where it permits residential use and is within a city’s Significant Urban Area, is excluded from the Atlas. This systematically reduces the proportion of highly restricted residential land reported in our final measures.

The residential-land definition

What goes into the denominator—“residential land”—is itself a modelling choice.

• Residential land is a per-state allowlist of zones that permit housing. Everything not on the allowlist (industrial, commercial, special-use, and so on) is dropped from the denominator.
• The residential base is a zoning concept, not “land where people live”. It is the set of low-, medium- and high-density plus mixed-use-permitting residential zones—so vacant or non-housing parcels inside a residential zone are still in the base.
• Rural-zoned land is excluded from every output, numerator and denominator alike.
• Some genuinely-residential land sits outside the standard zone list and is added back by hand—Hobart’s Particular Purpose Zones, Brisbane’s Emerging Community precincts, Perth’s R-Coded special-use and development land. These re-inclusions are documented per city on Data inputs.

The restriction definition and thresholds

The headline “highly restricted” metric is the heart of the Australian Zoning Atlas. In this context, a “highly restricted” piece of land is defined as being one or more of the following:

1. it is limited to no more than 2 storeys, or
2. it permits detached housing only, or
3. it sits in a “low-density” zone, or
4. it is heritage-listed (state or local).

The choices baked into each trigger:

Height assumptions

Storeys are sometimes converted from metres

• A height in metres is converted to storeys as 3.1 m per storey plus a 1 m roof/ground allowance. Formally, storeys = floor((height in metres − 1) ÷ 3.1). The practical consequence: the 2-storey threshold binds at 4.1 m—a scheme allowing 4.1 m or less is counted as no more than 2 storeys. (bias: neutral—a midpoint convention; tested bidirectionally in the robustness sweep.)

Land with no published height limit is assumed not to be highly restrictive

• Land with no published height limit is assumed to not be highly restrictive. Residential land with no published height limit is never counted as no more than 2 storeys, because the project will not claim a limit it cannot see. This is the single largest source of deliberate conservatism in the restriction metric.
• Unlimited-height and uncoded zones are flagged high-density. Zones with no fixed cap (e.g. a Capital City Zone) are recorded as having no height cap, rather than having a cap of zero storeys.

Discretionary and mandatory height controls are counted the same

• Mandatory and discretionary height controls are treated identically. A hard cap and a discretionary guideline both count the same way—the model reads the number, not its legal force.
• Site-specific height modifiers are not modelled. Topography and other local conditions that might adjust a height limit on a parcel-by-parcel basis are out of scope.
• Some height-control overlays are not measured at all in Victoria. Design and Development Overlays and Built Form Overlays are not read. Instead, the zone-and-schedule height is assumed correct where mapped.

Heritage assumptions

• All heritage overlay gradings are flattened to “local”. Significant, contributory, representative—every local grading is treated as one class, because the gradings are not comparable across states.
• Where state and local heritage overlap, state heritage wins. A parcel under both registers is classed as state heritage.
• All Indigenous-related heritage items identified as such in the data are excluded. They are not counted as a restriction trigger.
• Heritage is excluded from the reform-capacity headline. Counting heritage land in uplift calculations would assume that upzoning extinguishes heritage overlays, which it does not. The capacity calibration runs on residential land excluding heritage.

Housing typology assumptions

Land that permits only detached dwellings is part of the “highly restricted” definition.

• It is a manual binary, coded per planning scheme at the zone level. For each zone the project records: do typologies beyond a detached house count as a permitted use?
• “Permitted” is read generously. Anything from a by-right use to a performance-based or discretionary pathway counts. If it is possible to build something denser than a detached house in the zone, then it counts.
• Granny flats and secondary dwellings do not count as allowing more than a detached house.
• Where typology data is absent the share is reported as not-classified, never as zero. In cities where detached-only is largely unclassified (e.g. Melbourne) the value is published as “—“ (null), so a missing classification is never silently counted as 0% restricted.

Rapid-transit rules

Transit proximity drives the “mid-rise near transit” reform and the rapid-stop maps. Whether a stop is “rapid” is a frequency test, not a route-type test, computed from each agency’s GTFS feed.

• A stop is rapid if its mean headway is ≤15 minutes over the weekday all-day window (a representative Tuesday, 07:00–19:00). Mean headway is the window length divided by the number of arrivals—the standard “effective wait” convention.
• Aggregation differs by mode. For bus, tram and ferry the test is per-route: at least one passenger-facing route at the stop must hit ≤15 min on its own (a rider cannot board “whichever shows up”). For train and metro the test is trunk frequency: all rail arrivals at a station count as one combined service, because cross-city through-running means no single line hits 15 min even where a train arrives every few minutes.
• Coverage is uneven as a result. Rail-heavy cities (Melbourne, Sydney, Brisbane) capture more of their network than bus-heavy ones (Canberra), where the per-route filter excludes most services.
• Walkable catchment is an 800 m straight-line (“crow-flies”) buffer from each rapid stop—a standard ~10-minute walk radius, not a street-network walk distance.
• Feeds are clipped to the SUA. Brisbane currently uses a pre-strike GTFS snapshot; see the limitation note on Data inputs.

Reform-uplift parameters

The reform page of the Atlas tests the potential of nation-wide zoning changes.

In the initial release, four reforms are tested:

• Townhouses everywhere,
• Mid-rise near transit,
• Mid-rise everywhere, and
• High-rise near the city centre.

For each reform, the model cycles through every residential parcel and asks: would this rule actually let you build more homes than the current rules allow?

The number is conservative by construction, for four distinct reasons:

• It counts only the extra homes. Not the total a lot could hold—only the additional dwellings the reform adds on top of what is already permitted. A lot that already allows an eight-storey building gets zero uplift from an “eight storeys near transit” reform, even if the lot is currently not fully built out. The model measures changes in permissibility, rather than the total.
• Each lot is stepped down to the form its size and frontage can actually hold. A reform might say “eight storeys allowed here,” but it is often not possible to fit an apartment block on a tiny lot. Before a lot is credited with uplift, it is tested against a size-and-frontage gate and stepped down to the tallest form it can genuinely take:

• Each lot’s uplift is equivalent to the highest form it qualifies for. Under an eight-storey reform, a lot that only clears the townhouse gate is credited with a three-storey townhouse, not eight storeys. The eligibility check can only ever remove or shrink a lot’s contribution, never inflate it.
• Each lot is only counted once. When reforms overlap—a lot near the CBD covered by both “high-rise near the centre” and “mid-rise everywhere”—the two are not added together. The model takes the highest density a given lot would be able to build and counts it once.
• Unconfirmable lots count for nothing. Lots that already allow tall buildings count for nothing; lots whose eligibility we cannot confirm count for nothing.

The modelling constants

Beneath the prose sit roughly forty scalar constants—dwelling yields, calibration dials, absorption rates, denominator definitions—centralised upstream into a single source-of-truth file. They are grouped below by what they govern.

Dwelling yields (homes per hectare)

How a permitted building form becomes a dwelling count.

(bias: the apartment-yield re-alignment raised modelled capacity; the constants themselves are calibrated to a real-world build model, not chosen to maximise the number.)

Per-city baseline zoned capacities

Calibration only runs where a city has a published state-government zoned-capacity baseline. Five capitals are anchored to a published figure, two draw from the work of the Grattan Institute. Darwin is the lone derived proxy.