CTD Cast Files

Apps that read this dataset from the release. The icons after a name are the app’s lenses — the spatial grain it shows the data at.

Every way to have the bytes, by source. Nothing here asks you to register first.

Tables from the release (Parquet)

The release’s own tables, as the parquet objects it is frozen from — the same bytes every app and package below reads. A table this dataset shares with others holds every dataset’s rows, so filter on dataset_key; a partition holds only this dataset’s. since is the release whose rows these are: an unchanged table keeps its object.

CF netCDF

One self-describing file, for a tool that reads netCDF.

ERDDAP (erddap.calcofi.io)

One ERDDAP dataset per grain. Subset in the browser or query it from a script; for netCDF take the CF file above.

full resolution (pre-thinning)

the unthinned series — every bin as the instrument recorded it, before the release's depth thinning

observations

one row per measurement — value, units and quality flag — joined to the sampling event it was taken on

sampling events

one row per cast, tow or transect: when, where and how it was sampled, with the effort that scales it

Legacy ids, still answering until 2026-12-04: calcofi_ctd_thin (replaced by calcofi_ctd-cast) · calcofi_ctd_measurement (replaced by calcofi_ctd-cast_full)

From the provider

The dataset as its provider publishes it, before CalCOFI ingested it.

The same release, from a script or a browser SQL shell.

Packages

The whole release, pinned to a version, with the citation one call away.

con <- calcofi4r::cc_get_db()
calcofi4r::cc_cite("calcofi_ctd-cast")

con = calcofi4py.cc_get_db()
calcofi4py.cite("calcofi_ctd-cast")

DuckDB, anywhere

No CalCOFI package needed: each table above is a plain parquet object, readable by any DuckDB (or Arrow, pandas, Spark) from its URL. The path carries a content hash — a table whose rows did not change between releases keeps the same object, so nothing unchanged is stored or downloaded twice. Swap in any table above; one shared with other datasets needs WHERE dataset_key = 'calcofi_ctd-cast'. Every object of every release is listed in db-schema.

SELECT *
FROM read_parquet('https://storage.googleapis.com/calcofi-db/ducklake/tables/obs/dataset_key=calcofi_ctd-cast/14f55db21599862db4a1823e/data_0.parquet')
LIMIT 100;

db-query, in the browser

__TBL:obs__ is db-query’s name for the pinned release’s obs object — the hashed path above, resolved for you — so the same SQL keeps working when a release changes the object.

-- calcofi_ctd-cast in the CalCOFI release v2026.09.11
SELECT *
FROM __TBL:obs__
WHERE dataset_key = 'calcofi_ctd-cast'
LIMIT 100;

Records about the data, in the standards each portal harvests.

Where this dataset is registered outside calcofi.io: each portal’s role, what it is for, the dataset’s status there and the identifier it is known by. The policy — which portal is the archive of record and why — is in Portals.

Policy Archive of record: none yet — a CalCOFI program dataset with no DOI-bearing deposit. EDI is the planned archive (package built and staged each release); NCEI planned through the SWFSC data manager. OBIS does not apply (no occurrences).