Method · how FRACTURA assesses a place

1 · What is assessed

An address is resolved to a single point — the OpenStreetMap address node — and everything on the sheet is measured or modelled at that point. Two things are assessed against the World Health Organization's health guidelines: road exposure, meaning modelled road-traffic noise and annual nitrogen dioxide, and regional air, meaning annual fine particles (PM2.5). Two planning benchmarks are shown beside them, never graded: public green space within a short walk, and tree canopy around the address. Greenness (NDVI) is reported as a value. Distance to water is shown as context and is not yet assessed. The geometry of the surroundings is measured at five scales and archived, and does not enter any finding.

The sheet says, for each guideline, one of three words — clears, marginal, exceeds — and shows the value, the WHO line and the date of the data behind it. There is no score, no letter and no prediction on the page. Coverage today is Ontario and metro Atlanta; the data stack is global and the archive grows as cities are added.

2 · The rule — two groups, two tiers, a letter kept in the API

Measures with a published WHO health threshold are banded against it. There are three: noise, nitrogen dioxide and PM2.5. Noise and nitrogen dioxide partly measure the same thing — the road outside — so they form one group, *road exposure*, and only the worse of the two counts. PM2.5 comes from a 1 km regional cell and forms the second group, *regional air*. Each group is clears, marginal or exceeds. Nothing is added or averaged across groups; the two findings are shown side by side.

Marginal has a precise meaning: above the WHO guideline but within the last of WHO's own interim targets for that pollutant — interim target 4 for PM2.5 (10 µg/m³), interim target 3 for nitrogen dioxide (20 µg/m³). Noise has no interim ladder and no marginal band: it clears at or below the WHO line for its source, or exceeds it.

Measures with a research-informed target but no health threshold — park access and tree canopy — are shown as met or not met beneath the findings and never change them. Measures with evidence but no threshold at all — NDVI — are shown as a value with their citation. If any group cannot be determined, the sheet says so and gives the reason; an unknown is never shown as a pass.

The combination of the two groups into a single letter, A to E, exists in the API as a fixed, published lookup so that a partner sorting many addresses has one key to sort on. It is deliberately not shown on this site. The rule is versioned; every sheet carries the version it was assessed under, and a change to the rule never silently re-orders addresses already assessed.

3 · Noise — modelled Lden at the address point

The lines are WHO's 2018 Environmental Noise Guidelines for the European Region: road traffic 53 dB Lden (45 dB Lnight), railway 54 (44), aircraft 45 (40). All three Lden lines sit on the same basis — the level at which 10 % of people are highly annoyed, on each source's own curve — and the Lnight lines on sleep disturbance. Road traffic additionally carries a cardiovascular line at 59.3 dB Lden, so 53 is the more protective of its two. WHO's caveat that 40 dB Lnight may not be fully protective for aircraft is carried into the output whenever an aircraft source is present.

The value is modelled, not measured on site. Roads, railways and industrial land from OpenStreetMap are the sources; road class sets the emission level; distance decay and built-form screening (ISO 9613-2 §7.4, using the building footprints between the source and the address) give the level at the address point. The model was checked against government noise maps — the US DOT National Transportation Noise Map for Atlanta and published maps for a UK reference set — and agrees within 1–2 dB; where real traffic counts exist, the road-class defaults sit within ±3.5 dB of them. Counts are used to validate, never as an input, so the method is the same everywhere.

Noise is assessed at the address point as a proxy for the most exposed façade, which is where WHO defines its values. The proxy is FRACTURA's modelling choice and is disclosed on every sheet. It has two consequences worth knowing: a corner lot facing an arterial reads at that arterial, and façade or address-point levels carry some exposure misclassification, which in the health literature tends to bias effect estimates downward rather than upward. Aircraft exposure is not modelled: within an airport's approach corridor the road group is reported as not determined and the sheet says why. The rail emission reference is marked provisional.

4 · Air — PM2.5 and NO₂ from 1 km satellite cells

The lines are WHO's 2021 Global Air Quality Guidelines: PM2.5 annual 5 µg/m³, nitrogen dioxide annual 10 µg/m³, with WHO's own interim-target ladders (35 / 25 / 15 / 10 for PM2.5; 40 / 30 / 20 for NO₂) defining the marginal band.

Both values come from the Washington University Atmospheric Composition Analysis Group's global products at 0.01° (about 1 km): PM2.5 from SatPM V6.GL.03, annual 2024; NO₂ from SatNO₂ V4.GL.01, the mean of the twelve 2024 monthly fields, requiring at least ten of twelve months. NO₂ is published in parts per billion and converted at the WHO factor of 1.88. The value shown is the value of the cell containing the address point — no interpolation, no local adjustment.

The products were compared with 23 station–pollutant pairs from Ontario's provincial monitors and the US EPA around the assessed parcels. PM2.5 reads about 0.7 µg/m³ high against monitors; NO₂ about 0.8 ppb low, with the largest misses in dense downtowns. The bias is reported, not corrected. Near-road monitors read 4–7 ppb above their cells, which is expected — a 1 km cell cannot see a single road — and is the reason NO₂ sits in the road-exposure group beside noise rather than on its own.

5 · Planning benchmarks — park access and canopy

Park access: public green space of at least 0.5 hectares within 300 metres. WHO Europe's 2017 brief gives 0.5–1 hectare within 300 metres straight-line; FRACTURA takes the lower end of that range and measures by network walk, following the 3-30-300 measurement consensus (Browning et al. 2024). Both are platform choices and are stated as such. The benchmark records mapped proximity only — whether the entrance is open, the hours, or the condition of the park are not measured, and the sheet says so.

Tree canopy: at least 30 % canopy cover, the 3-30-300 target. The zone is a 500 m radius circle around the address point, the same zone as NDVI; the literature fixes no geometry, so this is FRACTURA's choice. Canopy is read from the Meta/WRI canopy-height map (1.2 m pixels), counting pixels modelled at 3 m or taller; the imagery date is read from the tile and printed on the sheet, because it is a 2018-era snapshot. Beside the local figure, canopy on the 1.6 km road-network buffer used by the Australian cohort studies is shown for comparison only. Neither benchmark enters any finding.

6 · Greenness — NDVI, reported

NDVI within 500 m is computed from Sentinel-2 L2A imagery at 10 m: a per-pixel median of June–August scenes under 30 % cloud, cloud-masked, water excluded so a lakeside address is not read as bare, at least two clear observations per pixel, averaged over the 500 m circle. The season and the number of scenes are printed with the value.

It is reported with its citation (Rojas-Rueda et al. 2019, a meta-analysis of greenness and mortality) and never banded, because no health body has published a threshold for it. Greenness overlaps air quality — greener places tend to be cleaner places, and the association weakens once air pollution is accounted for — so greenness is never counted as additional protection against a road or air finding.

7 · Measured form — five radii, not graded

The surroundings are measured as a box-counting fractal dimension of the mapped geometry — a number between 1.0 and 2.0 describing how the built and natural pattern fills space — at five radii around the address point: 50 m, 250 m, 750 m, 2.5 km and 10 km (façade, block, precinct, neighbourhood, city). Each value carries the regression fit behind it as a confidence and the count of mapped features it was computed from.

No health threshold applies to these measures and they do not enter any finding. They are kept because they are measured from the same map, at the same point, every month, and because change in them over time is part of the archive.

8 · Sources, vintages and the monthly archive

Every value on a sheet carries the product it came from and the date of that product: OpenStreetMap for noise sources, footprints, parks, water and form, with the vintage of the map extract; SatPM and SatNO₂ 2024 for air; the canopy-height map with its imagery date; Sentinel-2 with its season and scene count. Thresholds and citations live in a versioned configuration beside the rule, and the rule version and review date are printed on every sheet.

Addresses in the archive are re-measured monthly, after each map refresh, and every measurement is kept with the versions it was made under, so an address can be compared with itself over time. Where a sheet's second measurement exists, the change is shown.

A result can be questioned from the sheet ("Query this result"). Queries are reviewed against the inputs and the rule; a result found to be wrong is corrected and the correction recorded.

9 · Street pages — the fixed 15 m reference point

City and street pages assess every named street in a city the same way. Each OpenStreetMap street way gets one point, beside its midpoint and 15 m to one side of the centreline. The same 15 m is used in every city. It stands in for the front of a typical first house: it is an assumption about the street, not a measurement of any particular home, and every street result is labelled with the distance it was assessed at. An individual address searched on the site is assessed at its own point, not at 15 m.

The distance matters because road noise falls away from the road. Traffic acts as a line source, so the level drops by about 3 dB for each doubling of distance; moving from the model's 5 m minimum to 15 m lowers it by about 5 dB (10 × log₁₀ 3 = 4.8 dB). A point on the road itself would read louder than any house.

A worked example, Burlington, Ontario: residential rights-of-way there are mostly 14 to 18 m wide (Official Plan 2020, Table 2), which puts the property line 7 to 9 m from the centreline, and minimum front yards are 6 to 11 m depending on the zone (Zoning By-law 2020, Part 2, Table 2.4.1). The nearest house front therefore sits about 13 to 20 m from the centreline, and 15 m falls inside that range. Burlington is an illustration only; the standard is not tied to its by-laws.

Reviewed by Tatiana Pilipenko, PhD, 6 November 2026.

10 · What this is not

It is not a health score, a risk score or a prediction. It is a comparison of modelled and satellite-derived exposure at a point against published guidelines, and a guideline is a population-level statement, not a statement about any one person.

It is not a measurement taken at the house. Noise is modelled; air is a 1 km cell; canopy is a 2018-era map. The sheet says which is which.

It is not a local compliance check. WHO's guideline values are stricter than many national and provincial standards; an address that exceeds a WHO line may still meet the applicable local standard, and the sheet does not say otherwise.

It is not listing data. Nothing on this site comes from an MLS.

It is not fixed. Tiers and thresholds change only when the evidence does, under a new version, and the version is on the sheet.

The evidence behind every line is in the four memos and the review above, by Tatiana Pilipenko, PhD, who reviewed the rule against them and approved it on 16 October 2026.