Population 2020 (100 m, Mollweide)

Modelled residential population per 100 m grid cell for 2020, mirrored from the JRC Global Human Settlement Layer, GHS-POP R2023A.

Reading it

Assets are Cloud Optimized GeoTIFFs, so a client can read only the window it needs. Item metadata for all 332 tiles sits in one STAC-GeoParquet index, items.parquet, rather than in per-item JSON.

Query recipes, including how to find the tile covering a place, are in AGENTS.md.

What was changed from the source

This is a derived, lossy product. The upstream rasters ship inside ZIP archives with no internal overviews, which prevents any client from reading part of a file over HTTP. Three changes were made:

• Float64 to Float32. Lossless for this data: the source values carry only single-precision information, and GDAL checksums matched exactly.
• LERC_ZSTD compression, per-pixel error bounded at 0.01 persons per cell. This is the lossy step. Across a 68-tile sample, the published population differs from the source by -0.000631% (-7,959 people). The largest absolute change to a tile total was 2,579 people. Among tiles holding over a million people, the largest relative shift was 0.0176%. Relative percentages are largest on nearly empty tiles because the quantization step is fixed; the absolute error there is a fraction of one person.
• Internal overviews and embedded statistics added. Neither exists upstream.

43 tiles that upstream publishes containing only nodata are omitted, since they hold no data and cannot carry the band statistics a renderer needs.

The grid, projection, model, inputs, authors, and licence are unchanged. See why this mirror exists for the measurements, and the authoritative original if you need the source bytes.

Citation

Schiavina M., Freire S., Carioli A., MacManus K. (2023): GHS-POP R2023A - GHS population grid multitemporal (1975-2030). European Commission, Joint Research Centre (JRC). doi:10.2905/2FF68A52-5B5B-4A22-8F40-C41DA8332CFE