@shoyer thanks for the tips, I think that getting the lons and lats from dask is probably the most elegant method.

After trying various things I am still struggling with dask, and in particular on how to apply elemwise to functions like np.meshgrid (I get shape broadcasting errors). To get me started, I'd be grateful for an example on how to use dask to replace the code snippet below:

```python import xarray as xr import numpy as np

ds = xr.DataArray(np.zeros((2, 3)), coords={'x': np.arange(3), 'y': np.arange(2)}, dims=['y', 'x']).to_dataset(name='data')

non-dask version

lon, lat = np.meshgrid(ds.x, ds.y) ds['lon'] = (('y', 'x'), lon) ds['lat'] = (('y', 'x'), lat) ds.set_coords(['lon', 'lat'], inplace=True) print(ds) ```

Thanks a lot!