Data of int16 type from Zarr store is of float32 type when read back in

[mliukis]

I might be missing something or it's just an internal xarray implementation (to handle missing values in integer data types), and I would appreciate if somebody could, please, clarify.
I am saving int16 data to Zarr store and provide the type in encodings settings. When reading data back in from the Zarr store the datatype is of float32 type.

Is it expected?
Is original data still saved to the Zarr as int16 type?

I see expected dtype of int16 of the read back in data if I omit the missing_value setting for the encoding.

import xarray as xr
import os
import shutil

x = range(10)

ds = xr.Dataset(
    coords = {
        'x': ('x', x, {'name': 'x_coord'})
    }
)

a = xr.DataArray(
    data=list(range(10, 20)),
    coords=[x],
    dims=['x'],
    attrs={'stardard_name': 'test_data'}
)
print('Original data:')
a.dtype
a.values

ds['a'] = a

output_file = 'test_a.zarr'

if os.path.exists(output_file):
    shutil.rmtree(output_file)

encoding_vals = {'a': {'dtype': np.int16, 'chunks': 5, 'missing_value': -1}}

# Write to Zarr
backend = ds.to_zarr(output_file, mode='w', encoding=encoding_vals, consolidated=True)

# Read back in
with xr.open_zarr(output_file, consolidated=True) as ds_from_zarr:
    print('Read back from Zarr store:')
    print(f'Encoding: {ds_from_zarr.a.encoding}')
    print(f'Actual data dtype: {ds_from_zarr.a.dtype}')
    ds_from_zarr.a.values

Produces output:

Original data:
dtype('int64')
array([10, 11, 12, 13, 14, 15, 16, 17, 18, 19])
Read back from Zarr store:
Encoding: {'chunks': (5,), 'preferred_chunks': {'x': 5}, 'compressor': Blosc(cname='lz4', clevel=5, shuffle=SHUFFLE, blocksize=0), 'filters': None, 'missing_value': -1, 'dtype': dtype('int16')}
Actual data dtype: float32
array([10., 11., 12., 13., 14., 15., 16., 17., 18., 19.], dtype=float32)