Use the Archive without aquascope: R, QGIS, DuckDB, Julia¶

Everything the harvest publishes is plain open files on a public Hugging Face dataset that serves HTTP range requests, so any tool that reads Parquet, GeoJSON, FlatGeobuf or PMTiles can use it in place. No account, no key. Base URL:

https://huggingface.co/datasets/Rekin226/aquascope-gauges/resolve/main/

file	what	tools
`stations.parquet`	the station catalog, GeoParquet 1.0 (WKB point geometry)	pandas, DuckDB, R arrow, GeoPandas, QGIS, GDAL
`stations.geojson`	the same as GeoJSON	anything
`obs/<variable>/<source>/<station_id>.csv.gz`	one station's daily record (`date,value`, SI)	anything
`obs/<variable>/<source>.parquet`	a whole source and variable (`station_id, date, value`)	pandas, DuckDB, R arrow
`basins/lev12.fgb`, `basins/lev12_topology.parquet`, `basins/lev12_attributes.parquet`, `basins/station_catchments.parquet`	BasinATLAS sub-basins (CC BY 4.0), routing, attributes, station catchments	GDAL, GeoPandas, DuckDB, QGIS
`basins/lev12.pmtiles`, `basins/lev06.pmtiles`	sub-basin outlines as vector tiles	MapLibre, QGIS 3.32+

R¶

library(arrow)
base <- "https://huggingface.co/datasets/Rekin226/aquascope-gauges/resolve/main/"
stations <- read_parquet(paste0(base, "stations.parquet"))          # 45k rows, WKB geometry column
flow <- read_parquet(paste0(base, "obs/discharge/uk_ea.parquet"))   # station_id, date, value (m3/s)
thames <- subset(stations, grepl("thames", name, ignore.case = TRUE) & source == "uk_ea")
head(merge(flow, thames[, c("station_id", "name")], by = "station_id"))

or with DuckDB from R (library(duckdb)), the same SQL as below. sf::st_read() opens stations.parquet and basins/lev12.fgb as spatial layers through GDAL: st_read("/vsicurl/https://.../stations.parquet").

DuckDB (CLI, Python, R, anywhere)¶

INSTALL httpfs; LOAD httpfs;
SET VARIABLE base = 'https://huggingface.co/datasets/Rekin226/aquascope-gauges/resolve/main/';
-- stations per source
SELECT source, count(*) FROM read_parquet(getvariable('base') || 'stations.parquet') GROUP BY 1 ORDER BY 2 DESC;
-- a whole source's daily discharge joined to names
SELECT s.name, o.date, o.value
FROM read_parquet(getvariable('base') || 'obs/discharge/hubeau_hydrometrie.parquet') o
JOIN read_parquet(getvariable('base') || 'stations.parquet') s USING (station_id)
WHERE s.name ILIKE '%seine%' AND o.date >= DATE '2020-01-01';
-- every gauge with its catchment's mean elevation and rainfall (BasinATLAS)
SELECT source, station_id, elevation_m, precipitation_mm_yr, up_area
FROM read_parquet(getvariable('base') || 'basins/station_catchments.parquet') LIMIT 10;

Older DuckDB versions: write the URL out instead of getvariable.

QGIS¶

Stations: Layer → Add Layer → Add Vector Layer → Protocol HTTP(S), URI https://huggingface.co/datasets/Rekin226/aquascope-gauges/resolve/main/stations.parquet (QGIS 3.28+ with GDAL's Parquet driver; or use stations.geojson). Or drag integrations/qgis/aquascope_gauges.qlr into the map: the catalog styled by source, plus the sub-basin outlines.
Sub-basins: Layer → Add Layer → Add Vector Tile Layer → New generic connection, URL https://huggingface.co/datasets/Rekin226/aquascope-gauges/resolve/main/basins/lev12.pmtiles (QGIS 3.32+ reads PMTiles over HTTP), or the FlatGeobuf as a vector layer through /vsicurl/…/basins/lev12.fgb (spatially indexed, so panning only fetches what is on screen).
A station's record: the agency link is in the url attribute; the daily series is …/obs/<variable>/<source>/<station_id>.csv.gz (Add Delimited Text Layer accepts an HTTP URL). The Explorer at https://rekin226-aquascope-explorer.static.hf.space/#s=/ computes flood frequency and trend for it in the browser.

Julia¶

using Parquet2, DataFrames, HTTP
base = "https://huggingface.co/datasets/Rekin226/aquascope-gauges/resolve/main/"
stations = DataFrame(Parquet2.Dataset(HTTP.get(base * "stations.parquet").body))

Terms¶

The catalog is factual metadata with a link back to each agency; observations are mirrored only for sources whose licence allows it (the license and redistributable columns say which); BasinATLAS is CC BY 4.0 (Linke et al. 2019). The dataset card lists the attribution for each source.