# Resources: sources, dates and reuse

## Minerals

Source: [USGS Mineral Resources Data System](https://mrdata.usgs.gov/mrds/), public-domain US government catalogue. Archive: `https://mrdata.usgs.gov/mrds/mrds-csv.zip`, publisher last-modified 23 August 2022; metadata edition 20160315. Systematic updates ceased in 2011. Retrieved 10 November 2026.

The archive contains 304,632 records. This derived map selects 184,406 located reports in 22 commodity groups and excludes 2,283 selected processing-plant records. Coordinates, source status, quality code, production-size code and years retain their historical meanings. These are reports of occurrences, prospects and mines; neither current operating status nor reserve estimates. Quality codes are not positional uncertainty in metres. Coverage is particularly uneven outside the United States.

The global overview groups unique reports in 144 longitude by 72 sine-latitude cells of equal spherical area. Colour identifies the dominant selected commodity and brightness shows record density. A multi-commodity record counts once in total density and can count in several commodity totals. Heat does not estimate ore concentration, economic value, reserves or unexplored resources. Regional records and selected profiles load separately.

## Oil and gas

Source: [Global Energy Monitor, Global Oil and Gas Extraction Tracker](https://globalenergymonitor.org/projects/global-oil-gas-extraction-tracker/), March 2026 map release, [CC BY 4.0](https://globalenergymonitor.org/creative-commons-license/). Public map data: `https://publicgemdata.nyc3.cdn.digitaloceanspaces.com/interim_maps/goget_map_2026-03.geojson`. Retrieved 10 November 2026.

The source contains 7,673 records: 6,035 exact and 1,021 approximate field/asset locations are plotted; 617 country-only positions are excluded from the map. Centroids are not well locations, underground deposit boundaries or live extraction. The source's inclusion thresholds and historical inclusions limit coverage. Oil production remains million barrels/year and gas production million cubic metres/year, each with its own source year. Missing values remain missing. No reserves or oil/gas fuel type are inferred from production or names. Joint-country source labels are preserved in details.

## Forests, surface water and farmland

Source: [ESA WorldCover 2021 v200](https://esa-worldcover.org/en/data-access), [CC BY 4.0](https://creativecommons.org/licenses/by/4.0/). © ESA WorldCover project 2021 / Contains modified Copernicus Sentinel data (2021) processed by ESA WorldCover consortium.

Citation: Zanaga, D. et al. (2022), *ESA WorldCover 10 m 2021 v200*, [doi:10.5281/zenodo.7254221](https://doi.org/10.5281/zenodo.7254221).

This is a dated land-cover classification, not a live resource inventory. Forests maps tree cover (10) and mangrove (95), including plantations. Farmland maps cropland (40), excluding much pasture. Water maps permanent surface-water bodies (80), not groundwater, intermittent water, potable supply or water rights. Published global overall accuracy is 76.7%; it is not an individual-cell confidence score.

All 2,651 publisher tiles contributed to the 0.25° nearest-sample global categorical overview. This display sampling is about 28 km at the equator; it does not display every native 10 m pixel. Settled regional views request categorical COG overviews, with a bounded 384-pixel patch and up to four tiles. Point inspection reads a single native cell from the original grid. Unavailable point reads remain unavailable, without substituting the coarse overview. Display colours are never decoded as measurements. 2020 and 2021 algorithms differ, so these layers do not claim change detection.

The image fallback uses the official 2021 Sentinel-2 annual true-colour composite WMS. It is labelled **Area view · 2021**, centred on the selected coordinates, and is not an identity-verified photograph of a facility, mine or underground deposit. Its date is independent of map imagery, traffic clocks and statistical report years.

## Reproducibility and software

Source hashes and derived file hashes are recorded in `Release_Documentation/global-statistics-atlas/resources-source-manifest.json`. Rebuild with `Website_Editor/tools/refresh_atlas_resources.py` and the recorded source files and tile samples. The decoder is the unmodified npm GeoTIFF.js 3.0.5 browser bundle; MIT and dependency notices accompany it. Remote range requests omit credentials, reject full-file responses and retain strict byte/time/cache limits. Large regional records and profiles are cached on use, not eagerly installed with the site.


## Annual land cover, 2017–2025

Source: Impact Observatory, Microsoft and Esri, Sentinel-2 10 m Land Use / Land Cover v003. [Official map programme](https://www.impactobservatory.com/maps-for-good/), [Esri publisher item](https://www.arcgis.com/home/item.html?id=cfcb7609de5f478eb7666240902d4d3d), [fixed public image service](https://ic.imagery1.arcgis.com/arcgis/rest/services/Sentinel2_10m_LandCover/ImageServer). Source maps: CC BY 4.0. The public service has its own published Esri service terms. Retrieved 10 November 2026; the publisher item explicitly describes annual 2017–2025 maps.

The nine bundled 1440 × 720 global grids are nearest-neighbour categorical samples at 0.25°, not native 10 m global images. BIP exports request one U8 raw band, no rendering function, nearest interpolation and the explicit numeric Year filter. Pixel validity is decoded from the appended MSB-first bit mask. We do not derive classifications from rendered RGB colours. Regional exports use the same year, square geographic extents and a 384 × 384 budget. Point inspection requests all nine annual raw provider samples at nominal 10 m; it does not claim the original source cell footprint. Invalid, duplicate, displaced, cloud and missing samples are not imputed. Year changes cancel earlier reads.

Trees include plantations; crops exclude much pasture. Annual water is not groundwater, drinking-water availability or all temporary floods. Flooded vegetation can include rice and wetlands. Rangeland is not a farm-boundary inventory. An annual class change is not by itself a measured forest-loss event. 2017 uses fewer source observations. Classification rules and coverage differ from the explicit ESA WorldCover 2021 reference. The imagery/photo clock remains separate.

## Recent iron ore assets

[GEM Global Iron Ore Mines Tracker](https://globalenergymonitor.org/projects/global-iron-ore-mines-tracker/), September 2026 V1, CC BY 4.0. Official public map: `https://publicgemdata.nyc3.cdn.digitaloceanspaces.com/Current_maps/giomt/giomt_map_2026-09.geojson`. Retrieved 10 November 2026. The output contains 980 located assets in 77 countries / areas: 744 source-exact and 236 approximate reference locations. Older retired assets are not complete. Exact means the publisher's visual reference location, not a surveyed deposit boundary.

Original capacity is thousand tonnes of ore/year; reserves and resources are thousand tonnes of ore. Display converts each separately to million tonnes/year or million tonnes. Capacity is a design rate, not annual production. Resources include reserves and must not be added to them. Some complex totals are equally allocated between pits by the publisher. Individual quantity observation years are absent from this map and remain unreported. These assets retain GEM IDs independently from legacy MRDS, so counts must not be added as unique deposits.

Reproduction: `Website_Editor/tools/refresh_atlas_resource_updates.py` with reviewed captured source files. File URLs, byte counts, SHA-256 hashes, class counts and source receipts are in `resources-data/updates.json` and `Release_Documentation/global-statistics-atlas/resource-updates-source-manifest.json`.
