Write one chunk of a CF profile file

Writes profile-level variables (one value per profile, taken from each profile's first row), the rowSize ragged-array index, and the obs-level depth + measurement variables. Call once for a whole dataset, or repeatedly with advancing offsets to stream a table too large to materialize — the 216M-row obs_ctd_full is written one cruise partition at a time, holding ~15 MB rather than the whole table.

Usage

nc_profile_write(
  nc,
  vars,
  wide,
  profile_cols,
  obs_types,
  profile_id_col = "profile_id",
  start_profile = 1L,
  start_obs = 1L,
  strlen = 64L,
  obs_cols = "depth"
)

Arguments

nc

Open ncdf4 handle created from nc_profile_def()'s vars.

vars

The vars element of nc_profile_def().

wide

Wide data.frame, one row per (instance, observation), ordered by instance then depth (or time) so each instance's rows are contiguous. Must contain profile_id_col, every obs_cols entry, and the instance-level and obs-level columns.

profile_cols

Character vector of instance-level column names.

obs_types

Character vector of obs-level measurement column names.

profile_id_col

Column identifying the instance (profile or trajectory).

start_profile, start_obs

1-based write offsets into the profile and obs dimensions.

strlen

Fixed character length, matching nc_profile_def().

obs_cols

Coordinate columns on the observation dimension, matching nc_profile_def().

Value

list(n_profile, n_obs) — the counts written, for advancing the offsets on the next chunk.

Details

Non-contiguous profile rows are a hard stop, not a warning. A contiguous ragged array encodes each profile as a run of rowSize consecutive rows, so rows interleaved between profiles produce a file that reads cleanly and assigns depths to the wrong casts. Ordering is the caller's job (ORDER BY profile, depth); verifying it is this function's.