staggered grids from xarray datasets

[ingwag]

Question

Question

Hi, I have been wondering how to correctly load staggered grids as xarray datasets directly?

My velocity data is stored in binary files on a c-grid (MITgcm indexing). Instead of making a netcdf copy, I feel like it is easier for me to read the data using some custom scripts (which I already have) and just pass on the xarray dataset (chunked with dask) directly to parcels.
However, I don't see any way to specify the indexing or anything else regarding the grid type. GridInexingType is not used anywhere in fieldset.from_xarray_dataset or Field.from_xarray. Or is this not necessary since the data dims/coords are already explicitly set in the dataset?

Thanks

[erikvansebille]

Hi @ingwag, thanks for your question. You are right that the setting of gridindexingtype is not very robust in Parcels v3. @VeckoTheGecko and I are currently working on a much more robust/user-controlable implementation in the new Parcels v4!

However, I think that even in Parcels v3, you should be able to do something like (not tested):

fieldset = FieldSet.from_xarray(...)
fieldset.U.gridindexingtype = "mitgcm"
fieldset.V.gridindexingtype = "mitgcm"

The gridindexingtype = "mitgcm" here controls the gridindexingtype used (see e.g. the code below; same for JIT-mode):

Parcels/parcels/field.py

Lines 1603 to 1612 in 1ce5908

	if self.gridindexingtype == "nemo":
	U0 = self.U.data[ti, zi, yi + 1, xi] * c4
	U1 = self.U.data[ti, zi, yi + 1, xi + 1] * c2
	V0 = self.V.data[ti, zi, yi, xi + 1] * c1
	V1 = self.V.data[ti, zi, yi + 1, xi + 1] * c3
	elif self.gridindexingtype in ["mitgcm", "croco"]:
	U0 = self.U.data[ti, zi, yi, xi] * c4
	U1 = self.U.data[ti, zi, yi, xi + 1] * c2
	V0 = self.V.data[ti, zi, yi, xi] * c1
	V1 = self.V.data[ti, zi, yi + 1, xi] * c3

Can you confirm if this works?

[ingwag]

Hi @erikvansebille,

setting it like that doesn't work.

AttributeError: property 'gridindexingtype' of 'Field' object has no setter

But passing it as a kwarg to from_xarray_dataset works.

[ingwag]

but testing with a simple simulation, one with gridindexingtype set to nemo and one to mitgcm doesn't show any difference. the output is identical.

[erikvansebille]

Hmm, that is surprising! Which version of parcels do you use? Are you running with JITParticles? Can you also test with ScipyParticles?

[ingwag]

I'm using parcels 3.1.4. And yes I have been using JITParticles. I ran a test with ScipyParticles and also had no difference. And to be precise, the distance between the simulations with mitgcm and nemo indexing is exactly 0.