I’m guessing it has to do with the ESRI / GDAL difference you mention. I just loaded the image and the Natural Earth coastline data in QGIS, which uses GDAL, and the datasets match up.

I loaded your curv-fp.tif-file over the GEBCO bathymetry in ArcGIS and find there is a (I think)
continuous offset of about 20km. Your data seems to be shifted 20km to the north. I compared with
my curv-tif-file I converted in GMT and find the same offset.
I don’t understand what is happening when you reencode the data in CDO, but maybe something during
projecting back and forth produced that offset.
Or it is an issue of the different definitions of web mercator in ESRI and GDAL ?!- see:
https://trac.osgeo.org/gdal/ticket/3962
Maybe you could check that and maybe tell me I am wrong :),

cheers,
Anna

Thanks for the tip. I’m familiar with GDAL but not GMT, so I only used GMT for what I couldn’t do with GDAL.

The image seems fine to me, although, for whatever reason, the download took a very long time to start.