MITgcm/xmitgcm

Dimension error with open_mdsdataset

Closed this issue · 5 comments

lily-dove commented

Hi all, I'm having a similar problem to issue #316. I'm trying to open ECCO-Darwin output stored on PODAAC. However, I'm getting an error regarding data dimensions.

ds = open_mdsdataset('/Users/lily/OneDrive/ECCO-Darwin', ignore_unknown_vars="True", default_dtype="float32")
ValueError: dimensions ('j_g', 'i_g') must have the same length as the number of data dimensions, ndim=3

I would appreciate any help or insight - this is my first time working with MITgcm output and I'm excited to approach some fun questions!

timothyas commented

Hi @lily-dove, I think adding the argument geometry="llc" should do the trick

timothyas commented

Also you may need to specify nx=270 as well. I can't remember if that's necessary or what resolution ECCO DARWIN uses

lily-dove commented

Hi @lily-dove, I think adding the argument geometry="llc" should do the trick

Thanks @timothyas - the error changes to read: "AssertionError: (50, 13, 270, 270) ['face', 'j_g', 'i_g']". Any thoughts? Still seems like a dimension error. I also added nx=270, but I'm pretty sure that's already defined in the grid meta files.

timothyas commented

Very strange. Could you list the files in the directory you're trying to read?

lily-dove commented

Hi all! Using MITgcm for very different questions now, but I figured out a potential cause of this and wanted to update.

Investigating led me to issue 283, and I realized I was trying to use xmitgcm on multi-threaded output. Solution was to re-run the model using singleCpuIo=.TRUE. in data. Didn't slow down my simulation, although it is quite a small domain.