Removing nan data variable values

[gabrahamastro]

Problem

I'm trying to perform a Fourier transform of some gridded data in a dataset. However, I keep getting an error because there are some values of the data variable which are nan.

Is there a way in which I can remove all datapoints in a grid which are attributed to a nan data variable?

Meta Data of Array

Dimensions: northing: 58, easting: 71
Coordinates: easting (easting), northing (northing), upward (northing, easting)
Data Variables: magnetic_anomaly (northing, easting)

Attempts to Resolve

I have tried the method of doing ds.dropna(dim='magnetic_anomaly', how='any') but this comes up with an error, since magnetic_anomaly is not a coordinate but a variable.

I've also tried to use the where function to remove those points which have nan values in the specific data variable, but this has no effect on outputs: ds = ds.where(ds.magnetic_anomaly.notnull(), drop=True).

Libraries and Versions

I'm using Verde (v1.8.1) to generate the grid through the harmonica (v0.7.0) EquivalentSources function (though I don't think this is especially relevant to the question, since this is simply how the nan values were generated in the first place).

The xarray version is 2024.9.0.

Concluding words

I feel this may not be possible to resolve, since it's a regular grid (meaning coordinates can't simply be removed) but I'd like to ask anyway in case I can. Looking forward to any responses.

[gabrahamastro]

I believe I've determined that, since I was using an xarray.Dataset, when it was being read by numpy in a test I was running (np.isnan(grid_to_array).any()) it wasn't being read correctly, giving an error that there are nan values because of the data type.

However, when I converted it to an xarray.DataArray this problem was fixed, providing False to the test.