Indicator register

Stated by a source and linked to it field by field; never estimated. See the data dictionary.

Stated by a source and linked to it field by field; never estimated. See the data dictionary.

Stated by a source and linked to it field by field; never estimated. See the data dictionary.

Stated by a source and linked to it field by field; never estimated. See the data dictionary.

Computed by the atlas from the assembled record (geometry, location trust, country, identity, address, provenance, capacity); see the data dictionary.

Point-in-polygon join of the facility coordinates to the Aqueduct 4.0 'baseline_annual' layer (HydroBASINS level-6 sub-basins intersected with GADM level-1 units). Points falling just outside every polygon (coast) take the nearest polygon within 5 km (EPSG:8857). Aqueduct -9999 ('No Data') is stored as null. Scores are 0–5 as published; values describe the whole sub-basin, not the facility's own water use.

Point-in-polygon join of the facility coordinates to the Aqueduct 4.0 'baseline_annual' layer (HydroBASINS level-6 sub-basins intersected with GADM level-1 units). Points falling just outside every polygon (coast) take the nearest polygon within 5 km (EPSG:8857). Aqueduct -9999 ('No Data') is stored as null. Scores are 0–5 as published; values describe the whole sub-basin, not the facility's own water use.

Point-in-polygon join of the facility coordinates to the Aqueduct 4.0 'baseline_annual' layer (HydroBASINS level-6 sub-basins intersected with GADM level-1 units). Points falling just outside every polygon (coast) take the nearest polygon within 5 km (EPSG:8857). Aqueduct -9999 ('No Data') is stored as null. Scores are 0–5 as published; values describe the whole sub-basin, not the facility's own water use.

Point-in-polygon join of the facility coordinates to the Aqueduct 4.0 'baseline_annual' layer (HydroBASINS level-6 sub-basins intersected with GADM level-1 units). Points falling just outside every polygon (coast) take the nearest polygon within 5 km (EPSG:8857). Aqueduct -9999 ('No Data') is stored as null. Scores are 0–5 as published; values describe the whole sub-basin, not the facility's own water use.

Point-in-polygon join of the facility coordinates to the Aqueduct 4.0 'baseline_annual' layer (HydroBASINS level-6 sub-basins intersected with GADM level-1 units). Points falling just outside every polygon (coast) take the nearest polygon within 5 km (EPSG:8857). Aqueduct -9999 ('No Data') is stored as null. Scores are 0–5 as published; values describe the whole sub-basin, not the facility's own water use. Future layer joined on pfaf_id.

Daily T2M_MAX / T2M_MIN (2 m air temperature) from POWER. Values are for the MERRA-2 grid cell (0.5° lat x 0.625° lon, ~50 x 60 km) containing the facility, queried at the cell centre: a regional climate, not the micro-climate of the site; biased for sites in steep terrain or on coasts. Annual values are averages over 2015–2024 (days with -999 fill values are excluded and the average rescaled to 365.25 days; cells with <90% valid days are null).

Hourly dry-bulb temperature is NOT observed: it is reconstructed from daily Tmax/Tmin with a standard sinusoidal diurnal profile (minimum at 05:00, maximum at 15:00 local solar time; half-cosine rise 05→15 h, half-cosine fall 15→05 h towards the next day's minimum). This ignores humidity, cloud and wind effects on the diurnal cycle and the time zone offset of POWER's LST days; expect errors of a few hundred hours/year, largest in climates whose threshold lies near the daily mean. Values are for the MERRA-2 grid cell (0.5° lat x 0.625° lon, ~50 x 60 km) containing the facility, queried at the cell centre: a regional climate, not the micro-climate of the site; biased for sites in steep terrain or on coasts. Annual values are averages over 2015–2024 (days with -999 fill values are excluded and the average rescaled to 365.25 days; cells with <90% valid days are null).

Hourly dry-bulb temperature is NOT observed: it is reconstructed from daily Tmax/Tmin with a standard sinusoidal diurnal profile (minimum at 05:00, maximum at 15:00 local solar time; half-cosine rise 05→15 h, half-cosine fall 15→05 h towards the next day's minimum). This ignores humidity, cloud and wind effects on the diurnal cycle and the time zone offset of POWER's LST days; expect errors of a few hundred hours/year, largest in climates whose threshold lies near the daily mean. Values are for the MERRA-2 grid cell (0.5° lat x 0.625° lon, ~50 x 60 km) containing the facility, queried at the cell centre: a regional climate, not the micro-climate of the site; biased for sites in steep terrain or on coasts. Annual values are averages over 2015–2024 (days with -999 fill values are excluded and the average rescaled to 365.25 days; cells with <90% valid days are null).

Daily T2M_MAX from POWER. Values are for the MERRA-2 grid cell (0.5° lat x 0.625° lon, ~50 x 60 km) containing the facility, queried at the cell centre: a regional climate, not the micro-climate of the site; biased for sites in steep terrain or on coasts. Annual values are averages over 2015–2024 (days with -999 fill values are excluded and the average rescaled to 365.25 days; cells with <90% valid days are null).

Daily T2MWET from POWER; percentile over all valid days of 2015–2024. Values are for the MERRA-2 grid cell (0.5° lat x 0.625° lon, ~50 x 60 km) containing the facility, queried at the cell centre: a regional climate, not the micro-climate of the site; biased for sites in steep terrain or on coasts. Annual values are averages over 2015–2024 (days with -999 fill values are excluded and the average rescaled to 365.25 days; cells with <90% valid days are null).

Same computation as 'cdd' on future daily series. Pattern-scaled delta change, month by month and separately for Tmax and Tmin: delta(m) = median over the GCMs of WorldClim-CMIP6 2041–2060 (m) − WorldClim 1970–2000 (m), both read at the same 10′ pixel (nearest pixel valid in all rasters within 5 pixels for coastal sites). Only the part of this change still to come after 2015–2024 is applied: every 2015–2024 POWER daily value is shifted by K·delta(m), K = (GSAT 2041–2060 − GSAT 2015–2024)/(GSAT 2041–2060 − GSAT 1970–2000) = (2.0 − 1.24)/(2.0 − 0.4) = 0.475 (global surface air temperature above 1850–1900: IPCC AR6 SSP2-4.5 best estimate, Forster et al. 2025 observed 2015–2024, observed 1970–2000). The indicator is then recomputed exactly as for the present. Keeps observed day-to-day variability (no change in variance, humidity or diurnal range beyond the Tmax/Tmin deltas is modelled); assumes local warming scales with global warming (pattern scaling). Avoids the cross-dataset bias of pure climatology replacement (MERRA-2 cell vs WorldClim pixel absolute levels). Values are for the MERRA-2 grid cell (0.5° lat x 0.625° lon, ~50 x 60 km) containing the facility, queried at the cell centre: a regional climate, not the micro-climate of the site; biased for sites in steep terrain or on coasts. Annual values are averages over 2015–2024 (days with -999 fill values are excluded and the average rescaled to 365.25 days; cells with <90% valid days are null).

Same computation as 'fc18_h' on future daily series. Pattern-scaled delta change, month by month and separately for Tmax and Tmin: delta(m) = median over the GCMs of WorldClim-CMIP6 2041–2060 (m) − WorldClim 1970–2000 (m), both read at the same 10′ pixel (nearest pixel valid in all rasters within 5 pixels for coastal sites). Only the part of this change still to come after 2015–2024 is applied: every 2015–2024 POWER daily value is shifted by K·delta(m), K = (GSAT 2041–2060 − GSAT 2015–2024)/(GSAT 2041–2060 − GSAT 1970–2000) = (2.0 − 1.24)/(2.0 − 0.4) = 0.475 (global surface air temperature above 1850–1900: IPCC AR6 SSP2-4.5 best estimate, Forster et al. 2025 observed 2015–2024, observed 1970–2000). The indicator is then recomputed exactly as for the present. Keeps observed day-to-day variability (no change in variance, humidity or diurnal range beyond the Tmax/Tmin deltas is modelled); assumes local warming scales with global warming (pattern scaling). Avoids the cross-dataset bias of pure climatology replacement (MERRA-2 cell vs WorldClim pixel absolute levels). Values are for the MERRA-2 grid cell (0.5° lat x 0.625° lon, ~50 x 60 km) containing the facility, queried at the cell centre: a regional climate, not the micro-climate of the site; biased for sites in steep terrain or on coasts. Annual values are averages over 2015–2024 (days with -999 fill values are excluded and the average rescaled to 365.25 days; cells with <90% valid days are null).

Same computation as 'fc24_h' on future daily series. Pattern-scaled delta change, month by month and separately for Tmax and Tmin: delta(m) = median over the GCMs of WorldClim-CMIP6 2041–2060 (m) − WorldClim 1970–2000 (m), both read at the same 10′ pixel (nearest pixel valid in all rasters within 5 pixels for coastal sites). Only the part of this change still to come after 2015–2024 is applied: every 2015–2024 POWER daily value is shifted by K·delta(m), K = (GSAT 2041–2060 − GSAT 2015–2024)/(GSAT 2041–2060 − GSAT 1970–2000) = (2.0 − 1.24)/(2.0 − 0.4) = 0.475 (global surface air temperature above 1850–1900: IPCC AR6 SSP2-4.5 best estimate, Forster et al. 2025 observed 2015–2024, observed 1970–2000). The indicator is then recomputed exactly as for the present. Keeps observed day-to-day variability (no change in variance, humidity or diurnal range beyond the Tmax/Tmin deltas is modelled); assumes local warming scales with global warming (pattern scaling). Avoids the cross-dataset bias of pure climatology replacement (MERRA-2 cell vs WorldClim pixel absolute levels). Values are for the MERRA-2 grid cell (0.5° lat x 0.625° lon, ~50 x 60 km) containing the facility, queried at the cell centre: a regional climate, not the micro-climate of the site; biased for sites in steep terrain or on coasts. Annual values are averages over 2015–2024 (days with -999 fill values are excluded and the average rescaled to 365.25 days; cells with <90% valid days are null).

Same computation as 'hot_days' on future daily series. Pattern-scaled delta change, month by month and separately for Tmax and Tmin: delta(m) = median over the GCMs of WorldClim-CMIP6 2041–2060 (m) − WorldClim 1970–2000 (m), both read at the same 10′ pixel (nearest pixel valid in all rasters within 5 pixels for coastal sites). Only the part of this change still to come after 2015–2024 is applied: every 2015–2024 POWER daily value is shifted by K·delta(m), K = (GSAT 2041–2060 − GSAT 2015–2024)/(GSAT 2041–2060 − GSAT 1970–2000) = (2.0 − 1.24)/(2.0 − 0.4) = 0.475 (global surface air temperature above 1850–1900: IPCC AR6 SSP2-4.5 best estimate, Forster et al. 2025 observed 2015–2024, observed 1970–2000). The indicator is then recomputed exactly as for the present. Keeps observed day-to-day variability (no change in variance, humidity or diurnal range beyond the Tmax/Tmin deltas is modelled); assumes local warming scales with global warming (pattern scaling). Avoids the cross-dataset bias of pure climatology replacement (MERRA-2 cell vs WorldClim pixel absolute levels). Values are for the MERRA-2 grid cell (0.5° lat x 0.625° lon, ~50 x 60 km) containing the facility, queried at the cell centre: a regional climate, not the micro-climate of the site; biased for sites in steep terrain or on coasts. Annual values are averages over 2015–2024 (days with -999 fill values are excluded and the average rescaled to 365.25 days; cells with <90% valid days are null).

Point-in-polygon join of the facility coordinates to the Aqueduct 4.0 'baseline_annual' layer (HydroBASINS level-6 sub-basins intersected with GADM level-1 units). Points falling just outside every polygon (coast) take the nearest polygon within 5 km (EPSG:8857). Aqueduct -9999 ('No Data') is stored as null. Scores are 0–5 as published; values describe the whole sub-basin, not the facility's own water use.

Point-in-polygon join of the facility coordinates to the Aqueduct 4.0 'baseline_annual' layer (HydroBASINS level-6 sub-basins intersected with GADM level-1 units). Points falling just outside every polygon (coast) take the nearest polygon within 5 km (EPSG:8857). Aqueduct -9999 ('No Data') is stored as null. Scores are 0–5 as published; values describe the whole sub-basin, not the facility's own water use.

Count of filtered detections within 10 km (cKDTree on 3-D unit vectors, chord radius) divided by 11 years. Active-fire detections (thermal anomalies at satellite overpass, ~1 km MODIS pixels, nominal position = pixel centre) are not burned area and not fire hazard: they include agricultural and prescribed burning, miss small or short fires and fires under cloud, and two daily overpasses per satellite can detect the same fire several times. Filter: type 0 (presumed vegetation fire; volcanoes, static industrial sources and offshore flares removed), confidence ≥ 30. Detections from every downloaded country file are pooled (so cross-border fires count when the neighbour country also hosts a facility). Sites in countries without a FIRMS archive (Faroe Islands, Liechtenstein, Nauru, North Macedonia, Saint-Barthélemy, Tuvalu) are null.

Distinct acquisition years among detections within 10 km. Active-fire detections (thermal anomalies at satellite overpass, ~1 km MODIS pixels, nominal position = pixel centre) are not burned area and not fire hazard: they include agricultural and prescribed burning, miss small or short fires and fires under cloud, and two daily overpasses per satellite can detect the same fire several times. Filter: type 0 (presumed vegetation fire; volcanoes, static industrial sources and offshore flares removed), confidence ≥ 30. Detections from every downloaded country file are pooled (so cross-border fires count when the neighbour country also hosts a facility). Sites in countries without a FIRMS archive (Faroe Islands, Liechtenstein, Nauru, North Macedonia, Saint-Barthélemy, Tuvalu) are null.

Nearest-neighbour query on the pooled detections within 50 km of any site. Active-fire detections (thermal anomalies at satellite overpass, ~1 km MODIS pixels, nominal position = pixel centre) are not burned area and not fire hazard: they include agricultural and prescribed burning, miss small or short fires and fires under cloud, and two daily overpasses per satellite can detect the same fire several times. Filter: type 0 (presumed vegetation fire; volcanoes, static industrial sources and offshore flares removed), confidence ≥ 30. Detections from every downloaded country file are pooled (so cross-border fires count when the neighbour country also hosts a facility). Sites in countries without a FIRMS archive (Faroe Islands, Liechtenstein, Nauru, North Macedonia, Saint-Barthélemy, Tuvalu) are null.

Ember CO2 intensity of generation, latest complete year per country.

Ember 'Electricity demand' / Demand.

Ember Aggregate fuel / Clean (renewables + nuclear), % of generation.

Ember Aggregate fuel / Renewables, % of generation.

Distance to the nearest OSM power=line with max voltage >= 220 kV, within 25 km.

energy_gwh × country ci_g_kwh / 1000. Location-based (national average grid mix); ignores PPAs, on-site generation, sub-national grids and utilisation.

it_load_mw × pue × 8.76, only when BOTH are measured and pue is a single number in [1, 3]. Assumes 100 % IT utilisation all year → upper bound, not actual consumption. No PUE is ever assumed.

Distance from the facility point to the nearest OSM power=substation (point or area; 0 if inside) whose max voltage tag >= 110 kV, within 25 km. Substations without a voltage tag are ignored.

Sum of GHS-POP residents in a geodesic disk of radius R around the facility. Each 30" cell is split into 10x10 sub-cells; the cell's population is weighted by the fraction of sub-cell centres within R (haversine), i.e. population assumed uniform inside a cell. Negative/nodata cells count as 0. At ~1 km resolution the 1 km value is coarse (the disk covers ~3–4 cells), treat it as an order of magnitude. Residential population, not daytime/workplace population.

Sum of GHS-POP residents in a geodesic disk of radius R around the facility. Each 30" cell is split into 10x10 sub-cells; the cell's population is weighted by the fraction of sub-cell centres within R (haversine), i.e. population assumed uniform inside a cell. Negative/nodata cells count as 0. At ~1 km resolution the 1 km value is coarse (the disk covers ~3–4 cells), treat it as an order of magnitude. Residential population, not daytime/workplace population.

0 if the point lies inside an OSM landuse=residential area, else distance to the nearest one, within 3000 m; rounded to 10 m.

Distance to the nearest OSM school area (Overture land_use class 'school', i.e. amenity=school grounds); 0 if inside; within 3000 m; rounded to 10 m. Schools mapped only as a point are NOT included.

Operator QID from dataset tags (or unique exact label match on organizations); P749/P127 closure; ultimate parent via current P749, else a single current P127.

Weighted mean of 6 published pillars; validation AUC 0.745.

Weighted mean of 7 published pillars; validation AUC 0.662.

Weighted share of an operator's data centers disclosing 8 metrics in open sources.

Candidate events were identified with web searches (English and local-language press, government and court sites) across four regional sweeps (US/Canada, north-west Europe, southern/northern Europe, rest of world). An event was kept only if at least one source URL was fetched during research and its content confirmed the date, place, event type and outcome recorded; unconfirmable candidates were dropped. Summaries are written in neutral own words. Coordinates are municipality or region centroids from OpenStreetMap Nominatim unless a site is known (geo_precision says which). Links to inventory facilities (dc_uids) are made automatically for site-level events only, when operator and location match unambiguously. The list is a documented sample, not an exhaustive census.

Candidate projects were drawn from operator, government and trade-press (DCD, Reuters, CNBC, company newsrooms) announcements of data-center campuses of >=100 MW announced capacity or >=USD 1 bn announced capex. Each MW, capex, date and status value was confirmed in the text of a source fetched between 2026-09-22 and 2026-09-24; values not found in fetched text are null. MW and capex are as announced/stated by the operator or reported by the cited source, not measured; mw_announced is the largest stated build-out for the site. Capex is only filled when a USD figure is stated; other currencies and non-capex figures (economic impact, lease revenue, land purchase) are in capex_note. Status reflects the most recent dated source listed. Coordinates: Nominatim (OpenStreetMap) municipality or county centroid unless site coordinates are available from a linked atlas facility (geo_precision='site'). dc_uids link to data/details.json only when the same operator's facility with a matching name lies at the site.

Per-request figures published by AI providers, government agencies, or well-known independent analyses; each value was read in the fetched source on 2026-09-23. Figures use different system boundaries (see scope) and are not strictly comparable. Unit conversions (J to Wh, gallons to mL, annual to daily) are noted where applied.