Measure Observed Temporal and Spatial Coverage per Dataset

Derives each dataset's real extent from the assembled core (sample + obs) instead of the coverage_temporal / coverage_spatial strings an ingest asserts in its calcofi.dataset_meta YAML.

Usage

observed_coverage(con, tables = c("sample", "obs"), digits = 1)

Arguments

con

DuckDB connection holding the assembled core.

tables

Tables to measure, in order. Each contributes whichever of datetime / latitude / longitude it actually has; a table absent from the connection is skipped rather than erroring.

digits

Decimal places for the formatted bbox label.

Value

Tibble, one row per dataset_key, sorted by key: time_min/time_max ("YYYY-MM"), lat_min/lat_max/lon_min/ lon_max (numeric), and the display labels coverage_temporal_observed / coverage_spatial_observed.

Details

Why measure rather than assert. A hand-written extent cannot help going stale — it is authored once and the data grows underneath it. Checked against release v2026.08.06, the asserted temporal string was wrong for 7 of 15 datasets: cce-lter_zoodb claimed coverage through 2021-05 when its data ends 2015-04, calcofi_phyllosoma stopped a year short of its own rows, and three datasets said "present" while in fact stalling in 2019, 2022 and 2023.

NaN is not NULL. A NaN coordinate survives IS NOT NULL, and min()/max() propagate it, so a single poisoned row would blow a dataset's whole bounding box out to NaN while every nullity check passed. The coordinate filter is isfinite(), which rejects NaN and ±Inf alike. See the same trap in append_sample(), which normalizes these at write time.

Absent beats invented. A dataset with no usable datetimes gets NA for the temporal half, not a guess — calcofi_phytoplankton is region-pooled and carries coordinates but no datetime, so it legitimately measures spatially and not temporally. Callers fall back to a declared static value there.