Environment

Each H3 cell at resolution 9 can have six environment measures: noise, air quality, flood risk, deprivation, broadband and crime. All six come from government open data. This page explains where each measure comes from, which countries have it, what a missing value means and how to query it.

Sources

Each measure is its own block on the cell. Each block carries a source that names the publisher and the year. The six datasets update on six different cycles, so there is no single environment version.

Every source is coarser than a resolution 9 cell. A 1 km air quality square covers about nine and a half cells. An LSOA covers five, at the median. A cell inherits the value of the source area it sits in. We never interpolate. Neighbouring cells often carry the same reading. A sharp change between two cells is the edge of a source area. The Data API serves environment data at resolution 9 only.

What's available in each country

The Data API has environment data for the United Kingdom, France, Spain, the Netherlands, the United States and Australia. Each measure covers a different set of them. The coverage table on the Data page shows which countries have environment data at all. The sections below say what each measure covers within a country, and what its source values mean. Read a block's source before you compare two cells across a border.

Air quality

Air quality is written only for inhabited cells. An empty stretch of moorland has no air quality block, even where a publisher models it.

• United Kingdom. Defra's PCM background maps, on a 1 km grid, in all four nations.
• France, Spain and the Netherlands. The EEA's 1 km interpolated maps, with source eea_aq_interpolated_2024. This is a different model from the UK's. It stops at the EEA mapping domain. AirQualityBlock in the API reference says what a cell outside it gets.
• United States. source is acag_v6gl03_2024_anenberg_no2_2020. We sample two rasters at each cell's centre: ACAG's PM2.5 for 2024 and Anenberg and Mohegh's NO2 for 2020. air_year names only the PM2.5 year. pm10_ugm3 is always null, because neither raster publishes it. A cell needs a reading from both rasters to get a block. About 100,000 of the roughly 6.38 million populated US cells have none. Most are on the coasts and lake edges of the lower 48 states, in Alaska and in Hawaii, where the NO2 raster has no data under a valid PM2.5 pixel. A small strip of the Aleutians lies west of the PM2.5 grid's -170° origin and has none either.
• Australia. source is acag_v6gl03_gl_2024_anenberg_no2_2020: the same two rasters and rules as the US, with ACAG's global (GL) PM2.5 cut in place of the North America one.

Noise

Noise is written only for inhabited cells, and only for cells a publisher modelled. Noise comes from four UK products, a French one, a Spanish one, a US one and a Dutch one. They do not publish the same sources as each other. Read source and extent before comparing two cells across a border.

• United Kingdom. defra_noise_2022_r4 for England, wg_noise_2022_r4 for Wales, sg_noise_2021_r4 for Scotland and daera_noise_2017_r3 for Northern Ireland. Northern Ireland is Round 3 because DAERA's Round 4 rasters are in a format we cannot read. Each field says what its own nulls mean. England and Scotland write an explicit zero below their published floor. Wales and Northern Ireland flag it as no data, so a quiet Welsh cell has no block rather than a low one.
• France. Cerema's round 4 road and rail bands, cerema_cbs_2022_r4. They cover metropolitan France except Paris (75), Hauts-de-Seine (92), Seine-Saint-Denis (93) and Val-de-Marne (94). They map no airports. Like Wales, they are contours, so a quiet French cell has no block.
• Spain. Spain's is MITECO's round 3 (miteco_mer_r3_2017), which mapped only the 16 regional and provincial road authorities (no state roads), 6 metro and regional rail operators (no ADIF lines), 11 airports and 23 agglomerations. It has no roads at all in Galicia, Aragón, Cantabria, the Balearics, the Canaries, Ceuta or Melilla. We use Spain's contour lines only where they close, so a sample is the level of the innermost closed contour around it: within one contour step of the modelled level, above it where a quieter ring sits inside a louder one, and further under where a louder contour inside it never closed and was dropped. A Spanish cell with no block may still be loud.
• United States. BTS's National Transportation Noise Map 2022, source bts_ntnm_2022_laeq24h. It covers road, rail and air (no rail in Hawaii). BTS publishes no Lden or Lnight: for each *_lden_db level, in the US it is BTS's LAeq24h (see NoiseBlock.source), labelled rather than converted, which reads lower than Lden, and every *_lnight_db is null there. Like Wales, below its 45 dB floor is nodata rather than a written zero, so a quiet US cell has no block at all. over55_area_pct is always null for the US.
• Netherlands. The Netherlands' is RIVM's "Geluid in Nederland" (source rivm_geluid_nl), an indicative model covering all roads including municipal ones, rail and (Lden only) aviation, nationally. RIVM's own wording is "ruwe indicatie", and it is not a statutory END round. RIVM publishes no aviation Lnight. Each of its grids writes an undeclared fill below its floor rather than no data. Values are reported only from 40 dB up, so a Dutch cell with nothing at or above that floor gets no block. The Netherlands is the one country where no block can mean assessed and quiet. RIVM's aviation grid is whole-country too, so it counts toward over55_area_pct alongside road and rail. A Dutch cell's share can be driven entirely by aircraft noise near an airport.

Flood risk

Across the six countries, flood covers England, Wales, Scotland, Spain, metropolitan France's TRI perimeters, the Netherlands, the US counties FEMA's NFHL maps, and Victoria, the ACT and Brisbane. The publishers' zones do not compare like for like across a border. FloodBlock in the API reference says why. Read flood.source. In every country, a cell a band reaches keeps its block whether anyone lives there.

• United Kingdom. SEPA's maps of flooding from rainfall and small watercourses are left out, so zones 2 and 3 are rivers and sea in every nation. Zone 2 is the EA's 1 in 1,000 band.
• Spain. Zone 3 is MITECO's 100-year lámina. Zone 2 is the 500-year one, a narrower outer band than the EA's 1 in 1,000. No Spanish cell carries zone 1. MITECO publishes its assessed areas as river-reach lines, so there is no perimeter to place zone 1 inside. Spain's source is always miteco_snczi_zones_only.
• France. Zone 3 is the TRI's medium-probability aléa. Zone 2 is the low-probability one less zone 3; zone 1 is a populated cell with at least half its area inside the TRI perimeters. Any other French cell no band reaches has no block.
• Netherlands. The Netherlands' zone 3 runs to 1 in 300 a year and zone 2 to 1 in 3,000, not the EA's 1 in 100 and 1 in 1,000. The Netherlands' zone 1 is a populated cell the LIWO raster reaches with no pixel in classes 3-5 — assessed and below 1/3,000 a year.
• United States. Zone 3 is FEMA's Special Flood Hazard Area. Zone 2 is the 0.2% annual-chance band, about 1 in 500, also narrower than the EA's 1 in 1,000. The US flood.source is always fema_nfhl_zones_only: the NFHL Availability extent that would place zone 1 is not loaded, so no US cell carries zone 1, and a US cell outside both bands has no block at all.
• Australia. flood.source is always vicmap_planning_overlays_zones_only, act_flood_extent_zones_only or brisbane_flood_awareness_zones_only: Victoria's flood planning overlays (LSIO and FO) and the ACT's flood extent are zone 3, each 1 in 100 a year; neither maps a wider band. Brisbane's Flood Awareness map is zone 3 for its High and Medium bands (1 in 20 and 1 in 100 a year) and zone 2 for its Low band (1 in 500 a year, stated on Council's River, Creek and Storm Tide maps rather than Overall, and narrower than the EA's 1 in 1,000); its Very Low band maps to no zone. No Australian cell carries zone 1; a cell outside every mapped band has no block. A cell straddling Victoria's border carries Victoria's share of the cell only.

Deprivation

Deprivation covers Great Britain, Spain, metropolitan France, the Netherlands, the US and Australia. Each nation ranks only its own small areas, so decile 3 in one is not decile 3 in another. Read source before comparing cells across a border. The area summary refuses to average across nations. See Total an area.

In each country, every deprivation index reaches a cell through the census table's small-area code — an LSOA in England and Wales, a sección in Spain, an IRIS in France, a buurt (cascading to wijk, then gemeente) in the Netherlands, a census tract in the US, an SA1 in Australia, and in Scotland a 2022 Output Area, which SIMD reaches through its 2011 Data Zone in serve.sc_geography_bridge — so a cell with no census row carries no deprivation block in any of them.

• Great Britain. SIMD publishes ranks only, so a Scottish decile is cut from the rank.
• France. kind is edi_score. A higher score is more deprived.
• Netherlands. CBS's SES-WOA 2023 score, kind ses_woa_score. It is a composite of household wealth, education and employment history by buurt. Its scale runs the opposite way to EDI's, so we rank it lowest score first. Decile 1 is still the most deprived tenth. score is the figure CBS published, with its sign unchanged. 70% of buurten are scored, covering 98.3% of households. The rest inherit their wijk's score, or their gemeente's where the wijk is unscored too.
• Spain. INE publishes no deprivation index on current sección boundaries. A Spanish cell carries ADRH mean net income per person, kind income_decile_adrh, with score in euros. We rank the figures and cut them into deciles. Decile 1 is the tenth with the lowest income, so it is an income decile and measures something different from the deprivation deciles elsewhere. A summary over Spanish cells returns that as depriv_decile_mean, labelled by the depriv_kind the API reference describes.
• Australia. ABS's SEIFA 2021 Index of Relative Socio-economic Disadvantage by SA1, kind irsd_score. score, rank and decile are the ABS's own published columns, not recomputed here — DeprivationBlock in the API reference gives the ranking direction.
• United States. CDC/ATSDR's SVI is a social vulnerability index. It ranks each census tract's overall percentile across socioeconomic, household, housing and transportation, and minority-status themes. kind is svi_percentile and score is the published RPL_THEMES, from 0 to 1. A US decile of 1 is the most vulnerable tenth of tracts, the same direction as decile everywhere else. Rank and decile are cut nationally across every published tract. They compare across states. They do not compare with another country's index. A cell whose row names a tract that published no readable RPL_THEMES has no block.

Broadband

Broadband reaches a cell through the census table, so a cell with no census row has no broadband block. The gap is per cell, so the rest of the country keeps its blocks.

• United Kingdom. Ofcom publishes broadband for the whole UK. It reaches cells by output-area code, so a cell with no census row gets no block however well Ofcom covers it.
• France. ARCEP, by INSEE commune, with source arcep_mci_<year>_t<quarter>. Fields that ARCEP's thresholds do not match are null. Read a null there as no figure. See BroadbandBlock in the API reference.
• Spain. SETELECO, source seteleco_2025_06, by INE municipio code. A municipio is far coarser than a resolution 9 cell, so thousands of Spanish cells share one set of figures. BroadbandBlock in the API reference says which fields Spain leaves null and why no_decent_pct is a proxy there.

Crime

• England and Wales. A cell inherits its LSOA's count.
• Northern Ireland. PSNI's rows carry no LSOA, so a cell counts the offences mapped inside it. source is police_uk_psni_* and lsoa_code is null.
• Scotland. Police Scotland's quarterly figures by 2022 ward, source police_scotland_mmw_*. Police Scotland publishes these figures as management information, without National Statistics status. Every cell in a ward carries that ward's rate, so it cannot tell one street from the next. Wards Police Scotland still names on pre-2022 boundaries have no block. Scottish wards are reached through the census table, as broadband is. Data updated without Scotland's census, NRS 2022, has no Scottish crime.
• Spain. The Ministerio del Interior's annual balance (source interior_sec_2025), whose own title gives its extent: capitales de provincia y municipios de más de 20.000 habitantes. Its figures are provisional. The balance marks its year as pending consolidation, so a later update can move them under the same source. Every cell of a municipio carries that municipio's whole calendar-year count in total_12m, so never add it up across cells. A smaller municipio has no block rather than a zero or its province's total.
• France. SSMSI's commune base (source ssmsi_2025), counted the same way as Spain. ssmsi_2025_partial sums only the indicators SSMSI published. A commune with none published has no block.
• Netherlands. Politie/CBS table 47022NED's "Totaal misdrijven" (source politie_47022ned), a trailing 12-published-month window like the UK's rather than a calendar year. It reaches cells through the census table by buurt, and lsoa_code carries the buurt code. A buurt with a real 0 across the window reads 0. Two things leave a buurt with no block. The table is on 2025 buurt codes against the census's 2024 map, so a buurt with no row in the feed at all has none. A buurt missing, repeating or nulling a month inside the window has none either, rather than a partial sum. 170 buurten, about 0.8% of residents and 3,968 cells, have no block in the data behind this page. That is the whole gap.
• Australia. New South Wales, Victoria and South Australia are by suburb (ABS SAL 2021), lsoa_code carrying the SAL code: New South Wales is the 12 months ending with the last month of BOCSAR's cached release (source bocsar_<yyyy_mm>, that month), Victoria the Crime Statistics Agency's year ending June (source csa_ye_2026_06), South Australia SAPOL's financial year to June (source sapol_person_property_2025_26), Queensland by police division (lsoa_code the division id) over the 12 months ending with QPS's last month with data (source qps_divisions_<yyyy_mm>, that month). See CrimeBlock in the API reference.
• United States. US crime comes in two tiers: fbi_cde_agency is the FBI Crime Data Explorer count of the agencies reporting an area's latest 12 complete months (consecutive, each non-null and, when the median of those of the 24 months to the window's end with a count is 10 or more, above half of it) over that whole area — a city agency's TIGER place or county subdivision, or a county or county-wide agency's county less the areas a city agency with 12 complete months resolved to — and <city>_pd is a police feed's rows counted in each cell, which replaces the agency row unless the agency's 12 months end later — a county force's feed only outside those same areas; basis below says what each row counts. Each tier's force, window, total_12m and per_1000 read differently. See CrimeBlock in the API reference. Beyond those two tiers, state police, university, tribal and other agencies get no area of their own, so a US cell has no block unless a loaded police feed, or an agency with 12 complete months whose name matches a TIGER place or a county subdivision in a county it is listed under, or whose listed county matches a TIGER county, covers it; the agency rows are gated on serve.cell_population, so an uninhabited cell has no agency row, and a police feed reaches one only where it counted a row, with a null per_1000.

crime.basis is defined on CrimeBasis in the API reference, and is null where this build states no basis for a row's source.

What a missing value means

A missing block never means zero. What it does mean depends on the measure.

• Flood. A flood block carries zone 1 where a publisher assessed the cell and no mapped zone reaches it, but only for inhabited cells. An absent block means the cell was never assessed, or nobody lives there. Showing it as no flood risk is wrong. Check flood.source first. Where it ends _zones_only, we wrote no zone 1 for that country, so absence there means nothing.
• Noise. noise is absent on uninhabited cells, and on cells no publisher modelled. Whether a low reading means assessed and quiet depends on the country. See Noise.
• Air quality. Absent on uninhabited cells, and outside each publisher's mapping domain. See Air quality.
• Deprivation, broadband and crime. Absent on cells with no census row, and where the publisher has no figure. See each section above.
• A field inside a block. null means the publisher has no figure for that field. Each block's schema in the API reference says what its own nulls mean.

A block you filtered out with layer= is left out of the response. There is no key for it, and so no null.

Query environment data

GET /v1/h3/environment takes exactly one of these.

• bbox=south,west,north,east returns every cell in a viewport.
• cells=<csv> returns a set of resolution 9 cells, such as a reachable area. A cell with no data is left out. There is no 404.
• division=<id> returns every cell in an administrative area. The id comes from GET /v1/divisions. See Administrative areas.

Add layer= to request a subset of noise, air-quality, flood, deprivation, broadband and crime. The default is all six. We recommend you ask for the measures you need. The API reads and returns less, so one measure is cheaper than six.

curl -H "Authorization: Bearer $POI_API_KEY" \
  "$POI_BASE/v1/h3/environment?country=GB&layer=deprivation&bbox=51.50,-0.15,51.56,-0.05&limit=2000"

Use POST /v1/h3/environment when the cell list is long. A reachable area is often thousands of cells.

curl -X POST -H "Authorization: Bearer $POI_API_KEY" -H "Content-Type: application/json" \
  "$POI_BASE/v1/h3/environment?country=GB" \
  -d '{"cells": ["89194ad3493ffff"], "layer": "deprivation,crime,air-quality"}'

Each cell in the response looks like this, with the blocks you asked for.

{
  "h3": "89194ad3493ffff",
  "population": 907,
  "deprivation": {
    "source": "mhclg_iod_2025_eng",
    "nation": "ENG",
    "kind": "imd_rank",
    "area_code": "E01001839",
    "decile": 2,
    "rank": 5767,
    "score": 36.087
  }
}

GET /v1/h3/environment/cell/{h3} returns one cell with every block and its source. It returns 404 when the cell has no environment data at all.

CELL=89194ad3493ffff
curl -H "Authorization: Bearer $POI_API_KEY" \
  "$POI_BASE/v1/h3/environment/cell/$CELL?country=GB"

limit defaults to 500 and the maximum is 10,000. Full parameter and response shapes are in the API reference.

Total an area

POST /v1/h3/environment/aggregate takes a set of cells or a division_id and returns one summary.

curl -X POST -H "Authorization: Bearer $POI_API_KEY" \
  -H 'content-type: application/json' \
  "$POI_BASE/v1/h3/environment/aggregate?country=GB" \
  -d '{"cells": ["89194ad3493ffff", "89195da4a93ffff"]}'

Averages are population-weighted, with three exceptions. gigabit_pct_mean and mean_down_mbps_mean weigh each source area once rather than each person. A premises figure averaged by people would answer a different question. AreaEnvironment in the API reference says what one area weighs in each country. flood_zone3_area_pct_mean is not population-weighted at all. It is a plain mean over cells that carry a flood block, and equal-area resolution 9 cells make that area-weighted.

Exposure is reported as counts, population_in_flood_zone3 and population_over_55db, rather than as averages. Both count the modelled population. residents_in_flood_zone3 and residents_over_55db count census residents instead, beside a census residents total. They are null whenever residents is; AreaEnvironment in the API reference says when that is.

Every average ships with a cells_with_<layer> denominator. A mean over a third of an area is a different claim from a mean over all of it.

depriv_decile_mean is returned only when every contributing cell shares one deprivation source. Otherwise it is null, and depriv_sources lists the mixture.

Attribution

Contains public sector information licensed under the Open Government Licence v3.0.

French broadband: ARCEP, Ma connexion internet, licensed under the Licence Ouverte / Open Licence 2.0 (Etalab).

Cobertura de banda ancha — Secretaría de Estado de Telecomunicaciones e Infraestructuras Digitales (MTDFP), reused under the general conditions of Real Decreto 1495/2011 — the catalogue entry (publisher E05251701) names no licence of its own, confirmed 2026-09-13 via datos.gob.es/aviso-legal and digital.gob.es/nota-legal. Values are joined to cells by INE municipio code.

Air quality for France, Spain and the Netherlands contains modified European Environment Agency data — the EEA's 1 km interpolated air-quality maps for 2024, licensed under CC BY 4.0. The modification: values are point-sampled from those rasters at each cell's centroid rather than area-averaged over the cell.

Origen de los datos: Portal Estadístico de Criminalidad (Ministerio del Interior), reused under the general conditions of Real Decreto 1495/2011.

Flood risk for Spain contains data from the Ministerio para la Transición Ecológica y el Reto Demográfico (MITECO), Sistema Nacional de Cartografía de Zonas Inundables, licensed under CC BY 4.0. The modification: the published láminas are reduced to the share of each res-9 cell they cover.

Flood risk for France contains data from the Ministère de la Transition écologique, de la Biodiversité et des Négociations Internationales sur le Climat et la Nature — the Directive inondation TRI maps, rapportage 2020 (updated 2022-01-26), downloaded from Géorisques and licensed under the Licence Ouverte v2.0. The modification: the published aléa bands and TRI perimeters are reduced to the share of each res-9 cell they cover.

Flood risk for the Netherlands contains data from LIWO, Watermanagementcentrum Nederland — the plaatsgebonden overstromingskans map, published under the terms at basisinformatie-overstromingen.nl, which permit reuse with attribution — also published by Klimaateffectatlas, 2026, under CC BY 4.0. The modification: the published class raster is reduced to the share of each res-9 cell in each class.

Noise for Spain contains data from the Ministerio para la Transición Ecológica y el Reto Demográfico (MITECO), Mapas Estratégicos de Ruido round 3 (2017 data), licensed under CC BY 4.0. The modification: the published contour lines are merged and closed into rings, then rasterised onto a 50 m grid — a cell with no closed ring around it carries no reading.

Noise for France contains data from the Centre d'études et d'expertise sur les risques, l'environnement, la mobilité et l'aménagement (Cerema), strategic noise maps round 4 (2022), downloaded from data.gouv.fr and licensed under the Licence Ouverte v2.0. The modification: the published 5 dB bands are rasterised onto a 50 m grid.

Noise for the Netherlands contains data from the Rijksinstituut voor Volksgezondheid en Milieu (RIVM), Geluid in Nederland — road (2022) and rail (2023) strategic noise grids and an aviation Lden coverage (2022) — released under the Public Domain Mark 1.0. The modification: the native 10 m road and rail grids are sampled at 50 m, and the aviation coverage is fetched pre-sampled at 50 m from RIVM's own WCS.

French crime: SSMSI, base statistique communale de la délinquance enregistrée 2025, published on data.gouv.fr under the Licence Ouverte / Open Licence 2.0.

NSW crime: NSW Bureau of Crime Statistics and Research; Victorian crime: Crime Statistics Agency; South Australian crime: South Australia Police; all CC BY 4.0. The modification: offences are aggregated per suburb. Suburb assignment: Australian Bureau of Statistics, ASGS Edition 3 allocation files, CC BY 4.0.

Queensland crime and police division boundaries: Queensland Police Service, CC BY 3.0. The modification: offences are aggregated per police division.

Australian deprivation: Australian Bureau of Statistics, Socio-Economic Indexes for Areas (SEIFA) 2021, CC BY 4.0; the modification: each SA1's published score, rank and decile are joined onto the cells it covers.

NL crime contains data from Statistics Netherlands (CBS) and Politie, Geregistreerde misdrijven per wijk en buurt, table 47022NED, licensed under CC BY 4.0. The modification: the published monthly totals are summed per buurt over a trailing 12-month window and divided by the buurt's WorldPop residents for a rate; a buurt short of all 12 months is dropped rather than partially summed.

NL deprivation contains data from Statistics Netherlands (CBS), SES-WOA sociaaleconomische status per wijk en buurt 2023, licensed under CC BY 4.0. The modification: the published score is ranked and cut into deciles here, lowest first; a buurt CBS left unscored inherits its wijk's, then its gemeente's, score.

US deprivation contains data from the CDC/ATSDR Social Vulnerability Index, 2022 database, US Government Work — released into the public domain.

US air quality contains data from the Atmospheric Composition Analysis Group (ACAG), Washington University in St. Louis — the V6.GL03 PM2.5 surface for 2024, North America cut — and from Anenberg and Mohegh's global NO2 surface for 2020. The NO2 surface is licensed under CC BY 4.0; the V6GL03 PM2.5 surface's licence is unconfirmed — only an earlier ACAG release (V6.GL.02.04) states CC BY 4.0. The modification: values are point-sampled from those rasters at each cell's centroid, and NO2 is converted here from its published ppb to µg/m³ at the US EPA's ×1.88 factor (25 °C, 1 atm).

AU air quality contains the same ACAG PM2.5 release's global (GL) cut and the same Anenberg/Mohegh NO2 surface, sampled and converted the same way; the same licence position applies.

AU flood risk contains data from the Victorian Department of Transport and Planning, Vicmap Planning, licensed under CC BY 4.0, and from the ACT Government, ACTmapi 1% AEP flood extent model: CC BY 4.0 per the service's licence field; terms unverified. And from Brisbane City Council, Flood Awareness — Flood Risk Overall, licensed under CC BY 4.0. The modification: the published overlays are reduced to the share of each res-9 cell they cover.

US flood risk contains data from the Federal Emergency Management Agency (FEMA), National Flood Hazard Layer, US Government Work — released into the public domain.

US noise contains data from the Bureau of Transportation Statistics (BTS), National Transportation Noise Map 2022, US Government Work — released into the public domain.

US crime contains data from the FBI's Crime Data Explorer (Uniform Crime Reporting Program), US Government Work — released into the public domain. The city police feeds are each city's own open data, reused under that city portal's terms.