Data architecture

# Habitability Data Architecture v1

Status: Draft for methodology validation — September 2026

## 1. Principle

Habitability must be defined by the constructs we want to measure, not by the indicators available in any single source.

The source strategy is therefore:

1. define the construct;
2. define the ideal observable or raw variable;
3. select the best source for that variable;
4. harmonise geography and time;
5. preserve provenance and quality metadata;
6. derive Habitability metrics;
7. normalize only after the measurement layer is credible.

Eurostat Urban Audit / City Statistics is retained as one source and benchmark layer, but not as the unique empirical backbone.

## 2. Target pipeline

Raw / administrative / spatial data
→ source-specific ingestion
→ common geography and time model
→ provenance + quality layer
→ derived Habitability metrics
→ normalization / trajectories
→ H-D-A-R interpretation

## 3. Evidence classes

### L1 — Observed / administrative
Directly measured or recorded data: population registers, transactions, deaths, monitoring stations, weather observations, networks, land cover, buildings.

### L2 — Habitability-derived
Metrics calculated reproducibly from L1 inputs: housing-cost burden, heat stress days, green space per capita, PM2.5 exposure, doctor accessibility, flood-exposed population.

### L3 — Modelled / estimated / survey
Modelled surfaces, estimates, surveys, interpolations, small-area estimates. Useful, but must remain visibly distinct from L1/L2.

Every metric must retain an evidence_class field.

## 4. Source hierarchy by domain

| Domain | Preferred source family | Role |
|---|---|---|
| Population / demography | national statistical institutes; OECD; Eurostat | official population structure and change |
| Labour / income | national statistical institutes; OECD; Eurostat | employment, income, inequality |
| Housing / affordability | official transaction registers; cadastral/open property data; national statistics | prices, rents, tenure, housing burden |
| Climate | Copernicus / ERA5 / CEMS | temperature, precipitation, drought, flood, fire, projections |
| Air / environmental quality | EEA; Copernicus; national monitoring networks | PM2.5, NO2, ozone, pollution exposure |
| Built form / land use | GHSL; Copernicus Land Monitoring; cadastre; OSM | density, imperviousness, green/blue space, urban form |
| Accessibility / amenities | OSM; GTFS; official facility directories | healthcare, education, shops, transit, travel time |
| Mobility / transport | GTFS; road-network data; national transport statistics | access, service frequency, congestion, safety |
| Institutional / fiscal | national/local public accounts; OECD / Eurostat | municipal capacity, public services, resilience |
| Hazard / resilience | Copernicus EMS; national hazard maps; EEA | flood, heat, wildfire, drought and infrastructure exposure |

## 5. Geography strategy

Habitability must not be tied to a single administrative geography.

Canonical model:
- base spatial layer: grid / polygon-capable;
- FUA: harmonised metropolitan comparison layer;
- municipality / LAU: administrative implementation layer;
- region: macro-context layer;
- custom polygon: future product capability.

FUA remains useful because OECD/EU define it around an urban centre and commuting zone. But every source must be tagged with native_geography and harmonisation_method.

## 6. Time strategy

Every observation must carry:
- reference_period;
- frequency;
- publication_date when available;
- source_vintage;
- whether value is observed, revised, estimated or modelled.

No silent interpolation across missing years.

## 7. Minimum provenance fields

Every ingested observation or derived metric should be traceable through:
- source_provider;
- source_dataset;
- source_url or API identifier;
- source_variable;
- source_geography;
- canonical_geography;
- reference_period;
- retrieved_at;
- evidence_class;
- quality_status;
- transformation_recipe;
- methodology_version.

## 8. Quality statuses

Proposed controlled vocabulary:
- VALIDATED
- PLAUSIBLE
- REVIEW
- EXCLUDED
- PROXY
- MODELLED
- MISSING

A value cannot enter normalization unless its quality_status is VALIDATED or PLAUSIBLE under a documented rule.

## 9. Minimum Viable Habitability Dataset — design target

The v1 objective is not maximum indicator count. It is approximately 15–20 strong metrics spanning the major constructs needed to test H-D-A-R credibly.

Candidate metric families:

### H — Habitability
- unemployment / employment opportunity
- disposable income or purchasing power
- premature mortality / life expectancy proxy
- healthcare accessibility
- housing space / overcrowding
- air pollution exposure
- heat stress exposure
- everyday service accessibility
- transport safety
- personal safety

### D — Desirability
- net migration / population attraction
- residential demand or revealed preference
- amenity / green-space accessibility

### A — Affordability
- housing cost / disposable income
- rent burden
- energy burden
- transport burden
- residual purchasing power after essentials

### R — Resilience
- heat risk
- flood exposure
- drought / water stress
- wildfire exposure where relevant
- demographic vulnerability
- infrastructure / service redundancy where measurable

The exact inclusion set remains subject to data feasibility and non-redundancy testing.

## 10. Data-source decision rule

For each construct, select the preferred source using this order:

1. concept fit;
2. measurement validity;
3. geography compatibility;
4. historical continuity;
5. cross-territory comparability;
6. update frequency;
7. machine accessibility;
8. licensing / reproducibility.

A harmonised but weak statistic should not outrank a stronger source merely because it is already available in the database.

## 11. Eurostat policy

Eurostat is retained for:
- harmonised FUA / city statistics when the underlying series is credible;
- benchmarking against other official datasets;
- gap filling when methodology is documented and quality acceptable;
- metadata and territorial correspondence.

Eurostat is not allowed to define the conceptual scope of Habitability.

## 12. Immediate research gate before further normalization

Before STEP 09 normalization resumes:

1. build a construct-to-source matrix for 30–40 desired raw measures;
2. classify each as A: automatable now, B: engineering required, C: country-specific, D: proxy/unavailable;
3. identify the 15–20 metric Minimum Viable Habitability Dataset;
4. choose 3–5 core source families for v1 ingestion;
5. only then resume cross-city normalization and validation.

## 13. Strategic thesis

The long-term asset is not a collection of Eurostat indicators.

The asset is a reproducible territorial data engine that converts the best available raw spatial, administrative and environmental data into comparable measures of habitability over time.