Skip to content

Caravan — available extensions#

Every row pins a specific Zenodo version record, never the moving concept DOI, so a request is reproducible. Sizes and checksums were read from the Zenodo REST API on 2026-08-01; re-check with earthlens datasets refresh caravan.

All seven extensions are CC-BY-4.0.

dataset= Catchments Period Record Archive Size Range-readable
grdc 5,356 1950–2023 15349031 (v0.6) GRDC_Caravan_extension_csv.zip 8.84 GB ✅
germany 1,887 1951–2020 14755229 (v1.1.1) caravan_de.zip 6.09 GB ✅
spain 269 1990–2020 15040948 (v1.1.0) CAMELS-ES_v110.zip 0.40 GB ✅
czechia 249 1950–2025 17769325 Caravan-Extension-CZ.zip 0.91 GB ✅
denmark 308 1981–2020 15200118 Caravan_extension_DK.zip 0.52 GB ✅
israel 95 1950–2024 15181680 Caravan_extension_Israel_Ver4.zip 0.29 GB ✅
base 16,299 1950–2023 15530022 (v1.6) Caravan-csv.tar.gz 28.95 GB ❌ opt-in
base (version="1.2") 6,830 1981–2020 7944025 Caravan.zip 12.52 GB ✅

Every count here is measured, base included

base is the expensive one: a .tar.gz cannot be indexed without downloading it, so its 16,299 was established by pulling the full 28.95 GB archive and indexing it (2026-08-02) — hysets 12,162, camelsbr 870, lamah 859, camels 671, camelsgb 671, camelsaus 561, camelscl 505. The Zenodo changelog implies ≈16,300 (6,830 at v0.4, +9,130 at v1.4, +CAMELS-AUS v2 at v1.6); the measurement lands one below that.

base source datasets#

base is one archive bundling seven source datasets. They are not separately downloadable and are not valid dataset= values — request base and select their catchments by id or bbox.

Source directory Dataset Catchments
camels CAMELS-US 482
camelsaus CAMELS-AUS v2 561
camelsbr CAMELS-BR 376
camelscl CAMELS-CL 314
camelsgb CAMELS-GB 408
hysets HYSETS 4,621
lamah LamaH-CE 479

(Counts measured from the v1.2 index — the newest base release that can be read without downloading it.)

Two extensions are published under their source dataset's name

czechia and spain are titled CAMELS-CZ and CAMELS-ES — the word "Caravan" never appears in either title, only inside the record. A title-based search misses them entirely, which is why earthlens datasets refresh caravan searches record content. czechia also carries the most recent series of any extension, running to mid-2025.

Known but not wrapped#

Every Caravan-related record the project announces or that a content search surfaces has been checked. These are the ones deliberately not wrapped, each with the reason:

Record Why not
Caravan MultiMet (14196771 / 14196772) zarr cubes with lead-time bands on a UTC-0 base — a gridded shape, not a per-catchment table
Caravan-Qual (lite) (21337532) CC-BY-4.0, but per-parameter water-quality tables (wqms-csv/pH.csv, …) plus zarr — verified to expose no timeseries/ members
MSWEP4Caravan (19199860) Caravan-shaped, but CC-BY-NC-4.0 — non-commercial sources stay out of the default path
Caravan extension Iceland (lamahice) published on HydroShare, not Zenodo, and CC-BY-NC-4.0
Caravan-AUS-VIC (18736844) deleted by its author on 2026-03-01; the record returns HTTP 410
GAGES II announced in the project's thread #24 but never published with a DOI
CAMELS-CH (15025258) native CAMELS layout, not Caravan's — verified by opening it

earthlens datasets refresh caravan re-checks this list on every run, so a record that later gains a Caravan archive, or a new extension entirely, shows up as discovered rather than staying invisible.

Archive layout#

Each archive holds the same inner structure, but the directory it sits under differs per record and is absent in two of them — which is why member paths are resolved from the archive's own index rather than a template.

[<root>/]timeseries/csv/<source>/<gauge_id>.csv
[<root>/]timeseries/netcdf/<source>/<gauge_id>.nc
[<root>/]attributes/<source>/attributes_other_<source>.csv       # gauge lat/lon, country, area, name
[<root>/]attributes/<source>/attributes_caravan_<source>.csv     # climate indices
[<root>/]attributes/<source>/attributes_hydroatlas_<source>.csv
[<root>/]shapefiles/<source>/<source>_basin_shapes.shp           # + .shx .dbf .prj .cpg
[<root>/]licenses/<source>/license_<source>.md
Extension Root directory
grdc GRDC_Caravan_extension_csv/ (and ..._nc/ for the netCDF archive)
israel Caravan_extension_Israel_Ver4/
czechia Caravan-Extension-CZ/
spain v110/ — a bare version directory
base (1.2) Caravan/
denmark, germany (none — members start at the archive root)

Six shapes across seven extensions, no two derivable from the extension name. czechia also uses a singular license/ directory where every other archive ships licenses/. This is why member paths are resolved from the archive's own index rather than composed from a template.

For a member-by-member account of what those files actually contain — the 41 timeseries columns and their units, the four attribute tables, the NetCDF variant's variables and global attributes — see Inside a GRDC-Caravan archive.