# data-geo > Open geospatial datasets whose publishers serve them only as whole-file downloads, re-published cloud-native on Cloudflare R2 with a small API. Datasets: WorldPop's gridded population (release R2025A, 100 m, every country, 2015–2030), LandScan Mosaic Time Series (Oak Ridge National Laboratory, 3 arc-second ambient population, 1975–2024; point, polygon and catchment, section below), and EOG's VIIRS annual night lights (VNL V2.1 and V2.2, 15 arc-seconds, 2012–2025; polygon, section below). Files are read directly with HTTP range requests (GDAL, rasterio, xarray, DuckDB, QGIS); the API answers totals for administrative units, the value under a point, the population inside any polygon and within distances of a point, for one year or several with the change between them, each optionally by sex and age group with demographic indicators. Data © WorldPop, R2025A v1, CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/): cite WorldPop for every figure (population doi:10.5258/SOTON/WP00839, age and sex WP00841, under 18 WP00847). Full citations, the changes data-geo made and WorldPop's notices: [licence and citation](https://data-geo.baobabtech.app/worldpop/how-it-works.md#licence-and-citation). Base URL: https://data-geo.baobabtech.app How it works, in full: [/worldpop/how-it-works.md](https://data-geo.baobabtech.app/worldpop/how-it-works.md). The architecture, how each figure is computed and checked, measured speeds, and what we learned building it. Read it before reasoning about precision, speed or limits. ## One dataset per answer Each endpoint answers from its own dataset only. Rainfall answers carry no population and population answers carry no rainfall: to combine them (people in a district with low rainfall, rainfall weighted by where people live), call each endpoint for the same shape and period and join the answers yourself. data-geo does not mix datasets on the server. The one exception is the night-lights `population` block, built before this rule. ## Pick the cheapest route - Population of a named administrative unit (country, region, district): `GET /v1/worldpop/totals` or `/total`. A lookup, no raster read, milliseconds. - Population inside any shape you have as GeoJSON: `POST /v1/worldpop/polygon`. Exact; about a second for a city, about 5–8 seconds for a whole country. With every age and sex group, a city takes 1–3 s and a country's whole border much longer (Ethiopia 13–16 s): for an administrative unit, use `/totals` with `groups`. - People within distances of a point (a clinic, a school, a city centre): `GET /v1/worldpop/catchment?lon=&lat=&radii=1,5,10`. Each radius is a geodesic circle; about a second. - The value of one 100 m cell: `GET /v1/worldpop/point`. - Change over time: `years=2015-2030` (or `years=2015,2020,2025`) in place of `year` on any data endpoint gives every year, the change from first to last and the peak in one request. For an administrative unit this is still a lookup. A polygon reads its edge once per year, so a large one over many years is refused (413): use `/totals`, or read the Zarr cube, which holds all 16 years in each chunk. - Median age, sex ratio, dependency ratios and age shares: `indicators=true` on any data endpoint (years with age and sex data: 2015, 2020, 2025, 2030). - Rainfall (CHIRPS, monthly from 1981) over a shape or at a point, against the 1991–2020 normal: `POST /v1/chirps/polygon`, `GET /v1/chirps/point` (section below). - Anything the API does not compute: read the files (below), or WorldPop's own API (below) for what this mirror does not serve. ## API All responses are JSON. Errors carry `error`, and a 404 means the thing asked for is not published: report it as missing, do not substitute another figure. - `GET /v1` — this index as JSON. - `GET /v1/catalog` — datasets, releases, years and files. - `GET /v1/worldpop/totals?iso3=KEN&level=ADM1&year=2026` — every unit of one country at one level (`ADM0`, `ADM1`, `ADM2`), with geoBoundaries `shape_id`, `shape_name` and `population`. - `GET /v1/worldpop/total?iso3=KEN&unit={shape_id}&year=2026` — one unit, with its boundary source, year and licence. - `GET /v1/worldpop/point?lon=38.7578&lat=9.03&year=2026` — the 100 m cell under the point: people in that cell, and which country file holds it. `population: null` means the cell has no value (WorldPop places people only on built-up cells). - `POST /v1/worldpop/polygon?year=2026` with a GeoJSON Polygon, MultiPolygon, Feature or FeatureCollection in lon/lat as the body — `population`, `cells` (100 m cells with a value inside), and `method` (how many 10 km, 1 km and 100 m cells were used). Holes and multipolygons are handled. A polygon too large to answer in one request returns 413 with the reason: split it, or use the admin totals. - `GET /v1/worldpop/catchment?lon=36.8219&lat=-1.2921&radii=1,5,10&year=2025` — `rings` (everyone within each radius, cumulative) and `bands` (between one radius and the previous). Up to 10 radii, each at most 100 km. Each radius is a geodesic circle drawn as a 256-vertex polygon (99.99% of the circle's area), then counted like any polygon. - Optional on every data endpoint: `release=r2025a` (the default). - Optional on every data endpoint: `years` in place of `year`, `groups` (a breakdown by sex and age) and `indicators=true` (below). ### Several years: `years` `years=2015,2020,2025`, `years=2015-2030`, or both forms mixed; each 2015–2030, at most 16; give `year` or `years`, not both (400). With `years`, the per-year fields move into `years: [{year, population, groups?, indicators?}, …]` in ascending order; with two or more years the answer adds `change` (`from_year`, `to_year`, `absolute`, `percent`, `annual_growth_percent`, compound) and `peak` (`year`, `population`), and with `groups` the same per group under `change.groups` and `peak.groups`. With `year` the answer is unchanged. With `groups` or `indicators`, every year asked must have age and sex data, or the answer is 404 with `missing_years`, never a partial answer. A polygon or catchment over several years reads its edge once per year: past a budget of edge cells times years (all of Ethiopia: 9 years, or 4 with `groups`; about a minute, and more with groups) it answers 413 with a hint. ### Indicators: `indicators=true` Adds `indicators` and `indicator_definitions`, computed from the 42 age and sex sums of the same cells: `median_age` (linear interpolation inside the five-year group holding the half-way person, 90 and over taken as 90–100), `sex_ratio` (males per 100 females), `dependency_ratio` {`total`, `young`, `old`} (0–14, 65+ and both per 100 aged 15–64), `aging_index` (65+ per 100 aged 0–14), `child_woman_ratio` (0–4 per 100 women 15–49) and `shares` {`under_5`, `under_15`, `under_18`, `age_15_64`, `age_65_plus`, `women_15_49`} of everyone with age and sex data. Names and definitions are those of WorldPop's API v3 `POST /indicators`, so the two compare field by field; `under_18` is ours (WorldPop's own under-18 count, null where unknown). Ratios to 2 decimals, shares to 4, null where a denominator is 0. ### By sex and age: `groups` Add `groups=` to `totals`, `total`, `point` or `polygon`, and the answer gains `groups` (people per group) and `group_definitions` (what each sum adds up). - `groups=summary` — `female`, `male`, `under5`, `under18`, `women_15_49`, `age_15_64`, `age_65_plus`. - `groups=all` — the 42 stored bands: `f_00`, `f_01`, `f_05`, `f_10`, … `f_90`, the same for `m_`, then `f_under18`, `m_under18`. Age group `00` is under 1, `01` is 1–4, `05` is 5–9, then every five years; `90` is 90 and over. - Or a comma list of any bands and sums: `female`, `male`, `under1`, `under5`, `under15`, `under18`, `age_15_64`, `age_65_plus`, `women_15_49`, `t_00` … `t_90` (both sexes in one age group). Every group is counted over the same cells by the same rule as `population`, so the figures in one answer add up with each other. The 40 sex-by-age groups sum to `population` within 3.1e-5 of a person per 100 m cell (WorldPop computed each file separately); over a city the difference is under one person, over a country at most 120 people (the US, 2030). Nepal is the exception: in about 69,000 cells of its south (83.9–87.2°E, 26.6–27.8°N) WorldPop's groups add up to about a tenth of `population`, so its 40 groups sum to 554,616 people less (1.8%, 2030); the same in all four years. `under18` is WorldPop's own under-18 count, not a sum of five-year groups (15–19 straddles it). Status, 1 November 2026: total population answers on the API for all 16 years, 2015–2030, worldwide (242 countries). `GET /v1/catalog` gives each year's status, its countries and its global files. Age and sex answer worldwide for 2015, 2020, 2025 and 2030 (default release); other years answer 404 for `groups` and `indicators`. Admin totals cover every year, with `groups` for those four. geoBoundaries does not cover everyone: for 2030, 69.5 million people fall in no ADM0 unit, 47.4 million in no ADM1 unit and 55.3 million in no ADM2 unit (China 29.6 million at ADM0, Russia 20.5 million or 14% at ADM2, Argentina 4.0 million or 8.6%). They are counted in the country's file, not in any unit; `worldpop/r2025a/admin_totals.unassigned.json` lists them per country, level and year. 13 territories have no boundaries at all (among them Hong Kong, Macao, Puerto Rico and Western Sahara) and have no `totals`. Ethiopia and Kenya, every year, are a test set read with `release=r2025a/_test`. WorldPop publishes no under-18 files for Russia, Fiji and Greenland; there `f_under18` and `m_under18` hold no value, and any `under18` figure for an area that includes such a cell is null, never an undercount (GBR's come from WorldPop's zip). ### Examples Real answers on the default release, so you can check a call works (checked 1 November 2026; `years`, indicators and catchment 5 November). - Regions of Kenya: `GET /v1/worldpop/totals?iso3=KEN&level=ADM1&year=2026` → 47 counties; Nairobi `shape_id` 32016919B57908315498598, 5,762,597 people. - That one unit: `GET /v1/worldpop/total?iso3=KEN&unit=32016919B57908315498598&year=2026` → 5,762,597, with the boundary's source (RCMRD GeoPortal), year (2020) and licence. - One cell: `GET /v1/worldpop/point?lon=36.8219&lat=-1.2921&year=2026` → 69.39 people in that 100 m cell, from `worldpop/r2025a/100m/2026/ken.tif`. - Around a point: `GET /v1/worldpop/catchment?lon=36.8219&lat=-1.2921&radii=1,5,10&year=2025` → within 1 km 27,014, 5 km 935,402, 10 km 3,204,498. A 64-sided circle POSTed as a polygon gives 931,744 for 5 km: fewer sides cut more off the circle. - A Feature, a MultiPolygon, a hole: all accepted. A box over central Addis Ababa, 2025 → 1,606,479; the same box with a hole → 1,290,299; central Addis and central Nairobi as one MultiPolygon → 2,491,252. - A trend: the central Addis box with `years=2015,2020,2025,2030` → 1,266,919, 1,435,808, 1,606,479, 1,780,608; `change` +513,689 (+40.55%, 2.3% a year); peak 2030. Nairobi county, `/total?iso3=KEN&unit=32016919B57908315498598&years=2015,2030` → 4,622,138 to 6,203,820 (+34.22%, 1.98% a year). - Indicators: the central Addis box, 2025, `indicators=true` → `median_age` 25.3, `sex_ratio` 93.18, `dependency_ratio` 33.04 (young 28.54, old 4.51), `aging_index` 15.8, `child_woman_ratio` 22.62, `shares.under_5` 0.0773. - A whole country as a polygon: Ethiopia's boundary, 2025 → 133,610,866 in 3–5 s (cells of neighbouring countries' files that fall inside the boundary count too). For an administrative unit, `/totals` is faster and gives the same rule. - By sex and age: `GET /v1/worldpop/totals?iso3=KEN&level=ADM1&year=2025&groups=summary` → Nairobi 5,653,421 people: female 2,815,510, male 2,837,911, under 5 662,549, under 18 1,865,130, women 15–49 1,837,249, 65 and over 58,270. - The central Addis box with `groups=summary`, 2025 → 1,606,479 people: female 831,617, male 774,863, under 5 124,189, under 18 452,596, women 15–49 549,033, 65 and over 54,430. - One cell with `groups=under5,women_15_49,age_65_plus`: `lon=36.8219&lat=-1.2921`, 2025 → 67.84 people, of whom 6.33 under 5, 21.87 women 15–49, 0.94 aged 65 and over. Status codes: 404 not published (report it as missing; with `years`, `missing_years` says which), 400 malformed (the message says what), 413 polygon too large, or too many years for its size (split it, ask for fewer years, or use `/totals`), 429 over a rate limit (wait `Retry-After` seconds: the next hour for the hourly limit, about a minute otherwise). Use `requests`, `httpx`, `fetch` or `curl`; Python's built-in `urllib` is currently refused (403) by Cloudflare's bot screening. ### How totals and polygons count A 100 m cell is counted when its centre is inside the unit or polygon. Every route uses this one rule, so the polygon endpoint given an admin unit's boundary returns that unit's total. Admin totals report the population in cells whose centre falls in no geoBoundaries unit (coastlines, disputed borders) separately rather than dropping it. The polygon endpoint is exact, not an estimate. It uses a sum pyramid: 1 km and 10 km grids where each cell holds the exact sum of the 100 m cells under it. Coarse cells wholly inside the polygon are taken as one stored sum; only the cells along the polygon's edge are read at 100 m. ## Files Range-readable from `https://data-geo.baobabtech.app/`. Paths under a release never change. - `worldpop/r2025a/100m/{year}/{iso3}.tif` — one Cloud-Optimized GeoTIFF per country and year. float32, EPSG:4326, people per cell, nodata −99999, 512 × 512 tiles, no overviews. Lossless: equal to WorldPop's GeoTIFF cell for cell. - `worldpop/r2025a/100m/{year}/global.vrt` — every country for one year as one raster. Open it with GDAL: `/vsicurl/https://data-geo.baobabtech.app/worldpop/r2025a/100m/2026/global.vrt`. - `worldpop/r2025a/100m.zarr` — Zarr v3 cube, `population(time, y, x)`, all 16 years in each chunk (16 × 512 × 512), sharded, NaN where there is no value. `xarray.open_zarr("https://data-geo.baobabtech.app/worldpop/r2025a/100m.zarr")`. - `worldpop/r2025a/pyramid/{year}/1km.tif`, `10km.tif` — the sum pyramid. Two float64 bands: sum of 100 m cells, count of valued 100 m cells. - `worldpop/r2025a/agesex/100m/{year}/{iso3}.tif` — age and sex, one COG per country and year with 43 bands: `population`, `f_00` … `f_90`, `m_00` … `m_90`, `f_under18`, `m_under18` (band descriptions name them). float32, 32 × 32 tiles, pixel-interleaved: reading one band reads them all, so read a window, not the whole country. `agesex/100m/{year}/global.vrt` mosaics a year; `agesex/pyramid/{year}/` holds its 1 km and 10 km sums (45 float64 bands: population, cells, each group, then cells with an under-18 value); `agesex/admin_totals/{year}.parquet` one row per unit, a column per band. 2015, 2020, 2025 and 2030 worldwide (published 31 October 2026); Ethiopia and Kenya every year under `worldpop/r2025a/_test/agesex/`. The pyramid's last band counts the cells that have an under-18 value. - `worldpop/r2025a/admin_totals.parquet` — one row per geoBoundaries unit (ADM0–ADM2), one column per year (`pop_2015` … `pop_2030`). One row group per country. `SELECT * FROM 'https://data-geo.baobabtech.app/worldpop/r2025a/admin_totals.parquet' WHERE iso3 = 'KEN'` in DuckDB. - `catalog.json`, `worldpop/latest.json`, `worldpop/r2025a/manifest.json` (every file with size and SHA-256), `worldpop/r2025a/stac/catalog.json` (STAC 1.0: one item per country and year, `items/{year}/{iso3}.json`, and `items/{year}/global.json` for the VRT and pyramid). These four are rewritten as files land; the data files never change. ### Reading tips - Every file answers `Access-Control-Allow-Origin: *` and allows `Range`, so a web page can read it directly (since 7 November 2026). - File names are lower case: `100m/2026/eth.tif`. `ETH.tif` redirects (301) to it; follow redirects. A missing file is a 404 with a small JSON body pointing to `/catalog.json`, which lists what is published. - A ranged read answers 206, cold or warm, for every file including those over 512 MB (fixed 24 October 2026; before that the first read of a large file at a data centre could answer 200 with the whole file). A 200 to a ranged read is a fault: report it. - GDAL: set `GDAL_DISABLE_READDIR_ON_OPEN=EMPTY_DIR`. Do not set `GDAL_HTTP_MULTIPLEX` or `GDAL_HTTP_MERGE_CONSECUTIVE_RANGES`: with them, reads were ten times slower here. - The grid: origin (−180, 90), cell 0.00083333333° as stored (not exactly 1/1200). Every country file sits on whole cells of that grid; nothing is resampled. - Limits: 2,000 API calls (`/v1`) an hour per IP, no key, no daily quota; read `X-RateLimit-Remaining` and `X-RateLimit-Reset` on every answer, and on a 429 wait `Retry-After` seconds. Requests are also limited per IP over short windows, and `POST /v1/worldpop/polygon` far more tightly than file reads. Send polygons in small, paced batches, never a tight loop; for many administrative units use `/totals`. A COG read is 4–5 requests; a country-wide Zarr read is hundreds, so read large areas with fewer, larger requests or use the API. ## LandScan Mosaic Time Series ORNL's LandScan Mosaic Time Series v1.0: one global grid per year, 1975–2024, exactly 1/1200 degree (3 arc-seconds, about 90 m at the equator), float32, people per cell. It is **ambient** population, the people in a cell over an average 24 hours, not where they sleep: a business district holds more people than in WorldPop, a suburb fewer. National totals follow the US Census Bureau's International Database, so they differ from WorldPop's (Ethiopia 2024: 118.5 million here, 130.0 million in WorldPop R2025A). 2024 is ORNL's benchmark surface; 1975–2023 are backcast from it with built-up change and population totals through time, made to be compared across years. CC BY 4.0: cite Zimmer, A., Adams, D. S., Moehl, J., Buck, W., Martin, A., & Urban, M. (2026). LandScan Mosaic Time Series, Version 1.0: 1975-2024 [Dataset]. Oak Ridge National Laboratory. https://doi.org/10.48690/lsm/zew2-1n91 — for every figure. Not ORNL's service and not endorsed by ORNL. Published years: all of 1975–2024 under `v1` since 6 November 2026; `release=v1/_test` holds 1975 and 2024 from the test. `GET /v1/catalog` lists them (dataset `landscan-mosaic`: each year's files and global total); the release's manifest (`landscan/mosaic-timeseries/v1/manifest.json`, every file with its size and SHA-256) and STAC (`landscan/mosaic-timeseries/v1/stac/catalog.json`, one item per year) sit beside the files. - `GET /v1/landscan-mosaic/point?lon=38.7578&lat=9.03&year=2024` — the cell under the point; `population: null` where it has no value. The grid covers 83.67° N to 56.54° S. - `POST /v1/landscan-mosaic/polygon?year=2024` with GeoJSON as for WorldPop — `population`, `cells`, `method`. Same rule (a cell counts when its centre is inside), same sum pyramid method, exact. - `GET /v1/landscan-mosaic/catchment?lon=36.8219&lat=-1.2921&radii=1,5,10&year=2024` — as for WorldPop. - `years` works as for WorldPop (1975–2024). No `groups` or `indicators` (400): LandScan Mosaic has no age and sex. - Files: `landscan/mosaic-timeseries/v1/90m/landscan_mosaic_timeseries_{year}.tif` — ORNL's Cloud-Optimized GeoTIFF as ORNL ships it (byte for byte; SHA-256 checked against the file in ORNL's ZIP): float32, nodata −9999, 512 × 512 tiles, DEFLATE, overviews (ORNL's, averaged, not sums). 6.2–7.5 GB a year, 389 GB with the pyramids for the 50 years. Its grid starts 11 cells west of −180 and ends 9 east of 180; those 20 columns are empty. `landscan/mosaic-timeseries/v1/pyramid/{year}/1km.tif`, `10km.tif` — the sum pyramid, as WorldPop's. - Real answers (6 November 2026, `year=2024`): Ethiopia's ADM0 boundary as a polygon 118,463,104.5 people in 9,172,416 cells (about 2 s); Addis Ababa 6,374,237 (46,966 cells); `years=1975,2024` on Addis 469,508 → 6,374,237; the point 38.7578, 9.03 holds 189.7; within 1, 5 and 10 km of 36.8219, −1.2921: 151,530, 1,237,033, 3,107,233 (WorldPop R2025A 2025: 27,014, 935,402, 3,204,498: ambient population piles into the centre). Each equals a GDAL cell-by-cell count of ORNL's file. - WorldPop and LandScan Mosaic cells sit on the same lattice from (−180, 90) to within 0.002 of a cell, so the two can be compared cell for cell; but WorldPop's stored cell is 0.00083333333°, not 1/1200, and the API plans each polygon on its own dataset's cell size. ## Night lights (EOG VNL) The Earth Observation Group's annual VIIRS night lights (VNL V2, `average_masked`: EOG's background and outlier mask applied), radiance in nW/cm²/sr on a 15 arc-second grid (about 460 m at the equator), 75° N to 65° S. Every year 2012–2025, published 7 November 2026; page: https://data-geo.baobabtech.app/vnl. Each year is read from its own EOG version, V2.1 for 2012–2021 (`release` v21) and V2.2 from 2022 (v22); 2012 is April to December. CC BY 4.0. EOG's licence asks that graphs and pictures made from these data carry this notice, and large-format graphics the EOG logo: "This product was made utilizing VIIRS Nighttime Lights (VNL V2.2) data produced by the Earth Observation Group, Payne Institute for Public Policy, Colorado School of Mines." (VNL V2.1 for 2012–2021; each answer's `notice` names the right one.) Pass it on with any figure. EOG's licence document, which no longer answers at EOG, is kept at https://data-geo.baobabtech.app/vnl/eog_products_cc_license.pdf (Wayback copy of 28 September 2026). Cite Elvidge et al. (2021), Remote Sensing 13(5), 922, doi:10.3390/rs13050922. Not EOG's service and not endorsed by EOG. - `POST /v1/vnl/polygon?year=2025` (or `years=2012-2025`) with GeoJSON as for WorldPop. A VNL cell belongs to the polygon when its centre is inside. Answers: `mean_radiance` (area-weighted, nW/cm²/sr), `sum_of_lights` (radiance × cell area, nW/sr), `sum_of_values` (radiance summed over cells, as most papers' "sum of lights"), `lit_share` (share of the area EOG's mask leaves lit), `area_km2`, `cells`, `lit_cells`, and `population`: `weighted_mean_radiance` (each WorldPop R2025A 100 m cell inside takes the VNL cell under its centre, weighted by its people), `unlit_share` (people in cells the mask leaves dark: below VNL's detection, not a measure of electricity access), `people`. `population` is null for years WorldPop does not have (before 2015). `release` gives the release each year was read from (`v21`, `v22`); `notice` carries EOG's notice, naming V2.1, V2.2 or both; `method` says what was read. Exact: equal to a cell-by-cell count to 1e-13 on the test polygons (2015, 2024, 2025). - Files: `vnl/{v21|v22}/{year}/average_masked.tif` (radiance, EOG's values, every cell equal as bits), `lit_mask.tif`, `pyramid/{year}/{1km,10km}.tif` (sums), `popjoin/{year}/{1km,10km}.tif` (from 2015); listed with sizes and SHA-256 in `vnl/{release}/manifest.json`, STAC at `vnl/{release}/stac/catalog.json`, and in `/catalog.json` (dataset `vnl`). - Real answers (6 November 2026, 2024): Kenya's ADM0 boundary: `mean_radiance` 0.09995, `lit_share` 5.64%, people-weighted 5.336, 39.4% of people in unlit cells; Addis Ababa (test polygon): 20.64, people-weighted 30.85, 0.001% unlit. - No point or catchment endpoint yet; no `groups` or `indicators`. ### Monthly night lights (EOG VNL monthly), in testing EOG's monthly composites (v10, `avg_rade9h`, stray-light corrected where EOG has it), April 2012 to September 2026, at 15 arc-seconds. In testing since 7 November 2026 under `release=v10/_test`, months appearing as they are built; a month not built answers 404. No background mask: dark land reads near 0, often slightly negative, and fires, gas flares, aurora and boats are in. A cell with no cloud-free night that month has no value (not 0). No population join and no lit share. - `POST /v1/vnl/monthly/polygon?month=2024-06&release=v10/_test` (or `months=2024-01/2024-12,2025-06`) with GeoJSON. Answers `mean_radiance` (area-weighted over observed cells, nW/cm²/sr), `sum_of_lights` (nW/sr), `sum_of_values`, `observed_share` (share of the area with a cloud-free night that month), `observed_area_km2`, `area_km2`, `cells`, `notice` (EOG's), `method`. Exact: equal to a cell-by-cell count to 1e-13 (June 2012, six polygons). Compare months of one place; a single month's whole-country mean can be negative (noise). ## CHIRPS rainfall The Climate Hazards Center's CHIRPS v3 final monthly rainfall, 0.05° (about 5.5 km), land 60° N to 60° S, January 1981 onwards. Page: https://data-geo.baobabtech.app/chirps. CC BY 4.0 (the CHC also calls it public domain): credit the Climate Hazards Center and cite the CHIRPS3 Data Repository, https://doi.org/10.15780/G2JQ0P (2025), and Funk, C., Peterson, P., Harrison, L., et al. (2026), Sci Data 13, 718, https://doi.org/10.1038/s41597-026-07096-4. Not the CHC's service and not endorsed by it. Rainfall only: answers carry no population (see "One dataset per answer"). - `POST /v1/chirps/polygon?month=2026-09` (or `months=2026-07..2026-09`, up to 120, or a comma list) with GeoJSON as for WorldPop. Per month: `rainfall_mm` (area-weighted mean of the month's total), `normal_mm` (the 1991–2020 normal for that calendar month, same cells), `anomaly_mm`, `percent_of_normal`, `area_km2` (with a value), `area_no_value_km2` (sea, outside 60° N–60° S). Several months add `total`: sums of the monthly `rainfall_mm` and `normal_mm`, with anomaly and percent from those sums. The rule: a 100 m cell (exactly 1/1200°) counts when its centre is inside and takes the rainfall of the CHIRPS cell containing it; each CHIRPS cell is exactly 60 × 60 of them, edge on edge, so nothing is resampled; the mean is weighted by the spherical area of the counted 100 m cells. A district smaller than one 5 km cell still gets a value. Exact: equal to a GDAL cell-by-cell count to 4e-14 (Ethiopia, Kenya and four test shapes, seven months). - `GET /v1/chirps/point?lon=36.8219&lat=-1.2921&month=2026-09` (or `months=`): the CHIRPS cell containing the point, with its normal. `null` on the sea. - Admin tables, for long series or many units: `chirps/v3.0/admin/{YYYY}.parquet`, one row per geoBoundaries gbOpen unit (ADM0, ADM1, ADM2) and month: `iso3`, `level`, `shape_id`, `shape_name`, `month` (YYYY-MM), `rainfall_mm`, `normal_mm`, `anomaly_mm`, `percent_of_normal`, `area_km2`, `area_inside_km2`; one row group per country. `chirps/v3.0/admin/units.parquet`: each unit's inside area, 100 m and CHIRPS cell counts and boundary source, year and licence. The same rule as the polygon endpoint, unit by unit: each row equals what `POST /v1/chirps/polygon` answers for that boundary. DuckDB: `SELECT month, shape_name, rainfall_mm, percent_of_normal FROM 'https://data-geo.baobabtech.app/chirps/v3.0/admin/2026.parquet' WHERE iso3 = 'ETH' AND level = 'ADM2' ORDER BY month`. A new year's file is added, and the current year's rewritten, when a month is added. - The normal is **computed by data-geo**, not a CHC product: per calendar month and cell, the mean of the 30 monthly totals 1991–2020. Every month from 1981 to now is compared with it; months after 2020 are not part of it. Files: `chirps/v3.0/normals/1991-2020/chirps-v3.0.normal-1991-2020.{MM}.tif` (float32 mm, nodata −9999 declared, DEFLATE). - Speed (7 November 2026, live): see the page. Ten years over a country is the slowest request (each month reads one or two 512 × 512 tiles of about 850 KB); for long series over many units, read the files. - Real answers (2026-09): Kenya 27.99 mm against a normal of 24.83 (112.7%); Ethiopia 98.94 mm (90.5%); the point 36.8219, −1.2921 (Nairobi) 18.90 mm against 20.94 (90.3%), and July–September 47.03 mm against 54.46. - Files: `chirps/v3.0/monthly/chirps-v3.0.{YYYY}.{MM}.tif`, byte for byte the CHC's Cloud-Optimized GeoTIFF (only `.cog` renamed `.tif`): float32, mm in the month, EPSG:4326, 7200 × 2400, LZW, 512 × 512 tiles, overviews (the CHC's, averaged). **−9999 means no value and is not declared as nodata in these files: mask it yourself.** Every month to 2022 has 4,848,282 valued cells; from 2023 most have up to 7,714 fewer, as the CHC published them. Copied because the CHC's server sends no CORS header, so browsers cannot read it there (7 November 2026). Index: `chirps/latest.json` (newest month), `chirps/v3.0/monthly/manifest.json` (every file and normal with size and SHA-256), `chirps/v3.0/monthly/stac/catalog.json`, and `/catalog.json` (dataset `chirps`). - Daily and dekadal CHIRPS v3: WFP's COGs at https://source.coop/wfp (CORS sent), not here. ## What this is not - Not WorldPop, and not endorsed by them. A mirror of their published R2025A constrained 100 m grids, re-packaged so they can be read by range. WorldPop calls R2025A an alpha that may change, and asks that the data not be used to discriminate against, exploit, surveil or otherwise harm individuals or vulnerable populations. WorldPop's own API, on the same data: https://api.worldpop.org/v3/docs (below). - Only the constrained 100 m products (total population, and age and sex), and only the plain country files. WorldPop's both-sexes age files are not stored: they are female + male to float32 rounding, which `t_00` … `t_90` give. WorldPop's 1 km products are not mirrored. Variant folders on WorldPop's server (`*_ttp`, `*_v0`, `GBR_L2`, `NGA_L2`, `USA_States`) are not mirrored. - No reprojection, no clipped GeoTIFF downloads. ## WorldPop's own API WorldPop serves the same R2025A data through API v3 at https://api.worldpop.org/v3/ (guide https://api.worldpop.org/v3/llms.txt, Swagger /v3/docs, an MCP server at /v3/mcp). Anonymous use: 1,000 requests a day, 60 a minute, polygons up to 100,000 km²; a key, by registration and approval, raises them. Its polygon totals agree with this API's to under one person (both count a cell when its centre is inside; checked 5 November 2026). Use it for what data-geo does not serve: - age and sex in years other than 2015, 2020, 2025 and 2030; - place profiles with charts (Vega-Lite and PNG), comparisons and country maps; - `POST /exposure` (people near points, lines or polygons, weighted by distance), `POST /batch` (up to 1,000 features, ranked), catchments around many points at once; - `/search` by place name, and WorldPop's own administrative units; - clipped GeoTIFFs (`/raster`). Use data-geo for the files read by range (COGs, VRTs, the Zarr cube, sum pyramids, Parquet admin totals, STAC; on 5 November 2026 WorldPop's own files ignore ranges, and its planned AWS Open Data release was not yet listed), polygons of any size up to a whole country, geoBoundaries ADM0–ADM2 units with their source and licence, `method` in every polygon answer, and use with no key or daily quota. ## More - Pages for people: [WorldPop](https://data-geo.baobabtech.app/worldpop), [LandScan Mosaic](https://data-geo.baobabtech.app/landscan) - [How it works](https://data-geo.baobabtech.app/worldpop/how-it-works.md) (markdown; [as a web page](https://data-geo.baobabtech.app/worldpop/how-it-works)) - [OpenAPI](https://data-geo.baobabtech.app/openapi.json) - WorldPop R2025A: https://hub.worldpop.org