Reconcile coordinate systems before extracting, and say what got nothing

[chross22]

Diagnosing "areas with observations but no depth".

The bug

attach_bathymetry() passed st_coordinates(dat) straight to terra::extract(). Those coordinates are in whatever system the object carries, and extract() reads them as the raster's own. Observations in a projected CRS were read as though their metres were degrees — every point landed outside a lon/lat grid and came back NA.

Three points, identical but for their CRS:

EPSG:4326     DEPTH: 100  150  200
EPSG:32619    DEPTH:  NA   NA   NA      coords sent to extract: 500000, 4649776

That is exactly "observations with no depth in an area that plainly has depth", with nothing said about it.

sf input is now reprojected to the raster's CRS first. No CRS at all warns rather than being assumed correct.

A plain data frame carries no CRS to check, so magnitude is the only available signal — an unambiguous one, since a longitude cannot exceed 180. Those are refused, with the conversion to make.

NA depths now say why

The two causes have different remedies, and neither is guessable from a column of NAs:

2 of 40 point(s) got no bathymetry.
  1 fall outside the grid, which spans -70 to -66 lon and 41 to 44 lat.
    Widen `bounding_box` in fetch_bathymetry() to cover them.
  1 fall inside the grid on cells with no depth, which ETOPO treats as land.
    Inshore stations do this: a cell is land if most of it is, so a station in
    a narrow bay can sit in one. A finer `resolution` reduces it.

That second cause may well be what you are seeing if the missing points are coastal — it is not a bug, and no CRS fix will move it. A finer resolution will.

matchData

Already reprojected correctly — verified projected and geographic observations match the same cells. But a missing CRS failed with sf's "crs not found: is it missing?", which is true and says neither which object nor what to do. It now names both sides.

Verification

• 6 new tests: reprojection gives identical depths, missing CRS warns, projected data frames are refused, both NA causes are reported distinctly, a fully-matched set stays silent, and matchData names whichever side lacks a CRS.
• R CMD check — Status: OK. Full suite green. Installed into both R 4.6.1 and 4.3.2.

🤖 Generated with Claude Code