PM2.5 metric specification

# H05 PM2.5 exposure — metric specification v0

Status: RESEARCH / REVIEW
Date: 2026-09-12

## Objective

Specify the first end-to-end MVHD candidate tightly enough that Data Engineering can test a reproducible source-to-FUA pipeline without silently making methodological choices. This document does **not** validate H05, publish a territorial result, or authorize inclusion in a composite Habitability score.

## Construct and role

- Construct: long-term ambient fine-particulate exposure affecting human health.
- Habitability domain: environmental health / Natural + Human interface.
- Methodological role: **H — core observed condition**.
- Direction: lower long-term population exposure is better, all else equal.
- Unit: annual mean PM2.5 concentration, µg/m³.

PM2.5 is treated as an exposure condition rather than a proxy for adaptive capacity or a revealed territorial outcome.

## Primary metric definition to test

For FUA `f` and calendar year `t`, candidate H05 is the population-weighted annual mean ambient PM2.5 concentration across the FUA:

`H05(f,t) = sum_i(C_i,t * P_i,t) / sum_i(P_i,t)`

where `i` denotes spatial cells intersecting the FUA, `C` is annual mean ambient PM2.5 concentration and `P` is resident population used for weighting.

This formula is a **research specification**, not yet a frozen production transform. The implementation must document grid resolution, FUA boundary vintage, population raster vintage, treatment of partial cells, missing concentration cells, missing population cells and any temporal interpolation.

## Why population weighting

WHO defines SDG indicator 11.6.2 as mean population-weighted annual PM2.5 concentration in urban areas. EEA also uses population-weighted PM2.5 exposure in regional environmental-inequality indicators. This supports population exposure, rather than unweighted territorial surface concentration, as the primary construct to test.

## Absolute reference and relative comparison

The WHO 2021 annual PM2.5 air-quality guideline is **5 µg/m³**. The guideline is a health-based reference and should be shown alongside the observed concentration.

Do **not** convert the WHO guideline into an unsupported binary statement that a territory is simply "habitable" below 5 and "uninhabitable" above 5. PM2.5 health risk and Habitability are not binary at this threshold.

For v0 research output retain at least:

1. observed population-weighted annual concentration in µg/m³;
2. difference from the WHO 5 µg/m³ annual guideline;
3. relative cross-FUA comparison only as a secondary descriptive view;
4. multi-year trajectory where comparable source vintages exist.

No production normalization/scoring function is authorized here.

## Candidate source hierarchy

### Concentration

Preferred European research path: EEA/Copernicus or another authoritative European gridded/modelled PM2.5 product with documented annual coverage and reproducible access. PR #2 already contains an EEA PM2.5 access probe; reuse should be evaluated rather than rebuilding blindly.

EEA reports that its city air-quality ranking now uses modelled data and covers 761 Urban Audit cities, while historical/current official pollutant data can also be downloaded. These city products are useful validation/comparator evidence but must not silently substitute their city geography for the project's FUA geography.

### Population weighting

Use a documented gridded population source compatible with the chosen FUA boundaries. PR #2 contains GHSL and GISCO/OECD geography probes that should be assessed for semantically neutral reuse.

### Geography

Canonical target remains European Functional Urban Areas. Boundary source/vintage must be explicit. City/Urban Audit results may be comparator checks, not automatic FUA observations.

## Source / licence gate

EEA states that its reports, graphs and data can be reused for non-commercial and commercial purposes subject to acknowledgement and its legal-notice conditions. Before a pipeline is promoted, the exact dataset/product used must have its own source-registry entry with provider, product/dataset name, URL/API, retrieval date, licence/terms evidence, permitted reuse, attribution and transformation recipe.

A general EEA reuse statement does not remove the requirement to check product-specific notices or third-party components.

## Data-quality and reproducibility gates

A candidate H05 observation must fail closed unless all of the following are recorded or tested:

- FUA identifier and boundary vintage;
- reference calendar year / averaging window;
- concentration product and version/vintage;
- population product and version/vintage;
- common CRS and spatial resolution or explicit resampling method;
- partial-cell/intersection treatment;
- valid concentration coverage within FUA;
- valid population-weight coverage;
- missing-cell rule (no zero fill);
- population-weighted calculation reproducible from raw/intermediate inputs;
- source retrieval/provenance metadata;
- comparator/sanity check against an independent official city/station/modelled product where feasible.

Initial research gate proposed for testing: report coverage percentages rather than inventing a pass threshold. A numerical coverage threshold should be set only after observing source behaviour across multiple FUAs.

## Temporal semantics

Primary observation is annual. Trajectory must compare like-for-like source methodology. If a concentration model is materially revised across vintages, do not interpret the resulting discontinuity as territorial change without a back-cast or comparability assessment.

The EEA average exposure indicator uses a three-year average for a different regulatory construct. It is useful contextual evidence but should not be silently substituted for the proposed annual FUA metric.

## Validation plan

Phase 1 — reproduce one known FUA from source inputs and record every transformation.

Phase 2 — run a small heterogeneous comparator set (minimum: different countries and pollution regimes) to expose boundary, missingness and population-weighting failures.

Phase 3 — compare direction/order of magnitude against authoritative EEA/WHO city or station evidence where geography is sufficiently comparable; document expected differences rather than force equality.

Phase 4 — rerun from clean inputs and confirm deterministic output/provenance.

Only after these phases should H05 be considered for Research → Validated review.

## Falsification / exclusion criteria

Hold or reject the proposed implementation if:

- FUA coverage is systematically incomplete or spatially biased;
- population weighting cannot be reproduced from documented inputs;
- source licensing/reuse conditions are unresolved;
- geography mismatch dominates the metric;
- model/version changes prevent defensible trajectory interpretation;
- independent official evidence shows material unexplained disagreement;
- the pipeline requires undocumented zero filling, proxy substitution or manual corrections.

## Evidence anchors

- World Health Organization, Air Quality Database / SDG 11.6.2 methodology: population-weighted annual PM2.5 concentration in urban areas.
- WHO 2021 Global Air Quality Guidelines: annual PM2.5 guideline 5 µg/m³.
- European Environment Agency, Average Exposure Indicator / Air Quality Status: long-term urban population exposure and regulatory context.
- European Environment Agency, European city air quality viewer: current modelled city-level comparison and supporting measurement trend context.
- European Environment Agency reuse FAQ / legal notice: reuse permitted subject to attribution and applicable conditions; exact product-specific terms still require registry capture.

## Next Data Engineer mission

Audit the PR #2 EEA PM2.5 probe plus its GISCO/GHSL dependencies against this specification. Extract only the semantically neutral code/provenance pieces onto a clean branch, then run a reproducibility test without promoting any quantitative result to VALIDATED or Published.