In Parcels we need to ensure that we don't make any assumptions as to what Xarray is using to store the data arrays (numpy, cupy, cubed, dask etc).
This makes us flexible enough to support any array backend that Xarray supports. We can expand our test suite to test against this using testing utils provided via the data-apis iniative, or any existing Hypothesis testing strategies provided from xarray.
Resources:
https://data-apis.org/
https://lucascolley.github.io/talks/pydata-paris-25-array-api/