Understanding how Zarr handles NaN and the fill_value parameter

[palmerjohn248]

I struggling to understand from the documentation how Zarr (v2) handles NaN values and the fill_value parameter.

• Say I create a Zarr array with a numpy ndarray which contains np.nan values, are these stored as NaN values in the Zarr file? or are they filled using the fill_value parameter?
• Or is the fill_value parameter only used when creating empty arrays of specified dimensions containing uninitialized parts of the array?

Thanks

[d-v-b]

• Or is the fill_value parameter only used when creating empty arrays of specified dimensions containing uninitialized parts of the array?

this. more generally, the fill_value tells zarr readers how to interpret a missing chunk as an in-memory array.