Dataset · version 5 · 2000–2026

Monthly Precipitation of the Atacama

310 rain gauges with data, from 11 institutions in Chile, Peru, Bolivia and Argentina, from the Altiplano to the Pacific coast. The measured series and the series homogenized with climatol, with everything needed to repeat the procedure.

Tranque Sloman coast

Visviri Altiplano

The same 60 months (2000–2004) at both, one mark per month: rain, measured zero. Visviri records rain in 45 of them and reaches 171.5 mm; Tranque Sloman, in none. Those zeros are not missing data: they are the measurement, and they make up 65 % of this database.

How to cite

Monthly Precipitation of the Atacama (PPM_DB3), version 5 [dataset]. 2026. https://atacama.gfuentes.cl

BibTeX RIS

Provisional citation: the dataset does not have a DOI yet. State the version, because each one changes the homogenized values.

310

stations with data

320

measured months
2000–2026

49,732

measured values

65 %

measured at 0.0 mm

Data access

The CSV files are encoded in UTF-8 with BOM, so Excel opens them without breaking accented characters, and all of them use the same codigo_estacion, which joins with the catalog. The SQLite database contains everything else: the provenance of each coordinate, quality marks, cleaning actions and the parameters of each climatol run.

File	Content	Size
`ppm_observado.csv`	Measured precipitation, not infilled or corrected. 49,732 rows.	CSV · 924 KB	Download
`ppm_homogeneizado.csv`	Series homogenized with climatol, with a flag per cell (0 observed, 1 infilled, 2 corrected). 78,000 rows, up to 2025.	CSV · 1.6 MB	Download
`estaciones.csv`	Catalog: current coordinates, institution, country, altitude and rainfall regime. 317 rows.	CSV · 21 KB	Download
`ppm_atacama.sqlite`	The complete database: the series, their provenance, quality marks and the parameters of each climatol run.	SQLite · 23.8 MB	Download
`ppm_atacama.zip`	All of the above, plus a CSV copy of each table and the README.	ZIP · 5.9 MB	Download
`esquema.sql`	Annotated table definitions: why each one is the way it is.	SQL · 14 KB	Download
`parametros_v5.json`	climatol parameters declared to replicate the homogenization.	JSON · 15 KB	Download

Variables

Column	Unit	Description
`codigo_estacion`	—	Identifier from the source network. Joins with `estaciones.csv`, which catalogs 317, including 7 without a series (one excluded on purpose).
`mes`	AAAA-MM	Month of the record; sorts and compares correctly as text.
`ppm_mm`	mm	Accumulated precipitation for the month.
`bandera`	—	Homogenized file only: 0 observed, 1 infilled, 2 corrected.

Before using it

A zero is a data value. 65 % of the measured cells are exactly 0.0 mm: 32,207 of 49,732. A month measured as zero has its row with ppm_mm equal to 0; a month without measurement has none. Treating zero as missing destroys precisely what characterizes this desert.

Measurement and infilling are not the same. The homogenized series is complete because climatol estimates what is missing, and 39.0 % of its cells are infilled. For trends or comparisons between stations, filter bandera = 0.

Flags and quality marks

Each homogenized value carries a flag, and suspicions about the measurement are stored as separate marks: a mark does not delete the value, it points it out. They are in the observacion and marca_calidad tables of the database.

Code	Where	Meaning	Count
`bandera = 0`	homogenized	Observed: the measured value, unchanged.	47,316 cells
`bandera = 1`	homogenized	Infilled: month without measurement, estimated by climatol from neighboring stations of the same regime.	30,459 cells
`bandera = 2`	homogenized	Corrected: measured value adjusted for a detected break.	225 cells
`revisar`	marca_calidad	Warning on the whole series: series shorter than 5 years, altitude not published by the source, no data in the period, or complete years at 0.0 mm where the annual median is tens of mm (possible non-reporting). The v_medicion_limpia view excludes these series.	114 marks · 114 stations
`implausible`	marca_calidad	Month more than 6 times higher than its neighbors within 60 km and 800 m of elevation difference. The value is kept; the v_medicion_limpia view excludes it.	42 marks · 18 stations
`cambio_variable`	marca_calidad	The whole series records precipitation that is not comparable with that of its neighbors (thousands of mm per year next to stations with tens). It is kept, flagged, and left out of the series.	1 mark · 1 station

Methods and documentation

Before and after climatol

Twelve stations month by month, the map of the four regimes and what the homogenization changed: 4 breaks detected in the whole database.

See the figure

How it was made

From the station networks to this page, step by step, with the checks and decisions behind the database.

See the process

How it is stored

The database tables, their columns and relationships: where each value comes from and which version produced it.

See the data model

Query it without downloading

The same tables are in Cloudflare D1 behind a public API, for when you only need a subset: one station, one regime, a range of months. It returns JSON, or CSV with &formato=csv, and supports CORS.

Path	Returns
`/api/resumen`	Rows, stations and range of each product.
`/api/estaciones`	Catalog; filters `regimen`, `institucion`, `pais`, `bbox`, `q`.
`/api/estacion/{codigo}`	Record: coverage, quality marks and what each source says about where it is.
`/api/serie`	Monthly series; `estacion`, `producto`, `desde`, `hasta`, `bandera`, `formato`.
`/api/corridas`	The exact parameters of each climatol run.

curl "https://atacama.gfuentes.cl/api/serie?estacion=3450004&desde=2017-05&hasta=2017-08&formato=csv"
codigo_estacion,mes,ppm_mm,bandera
3450004,2017-05,0,0
3450004,2017-06,316.6,0        # the June 2017 flash flood, intact
3450004,2017-08,9.6,0          # July is missing: it was not measured

Available products: observado, homogeneizado, chirps and era5. There is a limit of 50 stations per request; for the full dataset use the files above, which do not consume quota.

Provenance and license

The series come from the DGA and DMC of Chile, the SENAMHI of Peru and of Bolivia, CEAZA, INIA, GHCN-NOAA and private operators; the fuente table of the database holds the URL and access date of each one. The homogenization uses climatol 4.5-0 (Guijarro).

License and redistribution: pending. DGA, DMC, CEAZA and GHCN publish openly, but the SENAMHI Bolivia stations do not come from an open portal and the DMC stations were obtained with a personal access key. Until this is confirmed with each institution, this release is provisional and should not be cited as a definitive source.

Version history

Version	Date	What changed	Status
v5	2026-10-01	Series with fewer than 48 measured months are no longer infilled; unreliable series inherit the regime of their reliable neighbors; annual totals require all twelve months measured; two-stage detection threshold (inht), taken from the exploratory analysis.	current
v4	2026-08-25	Homogenization by four rainfall regimes. Superseded: it assigned the regime of short series with missing months counted as 0 mm.	superseded