id,node_id,number,title,user,state,locked,assignee,milestone,comments,created_at,updated_at,closed_at,author_association,active_lock_reason,draft,pull_request,body,reactions,performed_via_github_app,state_reason,repo,type
343659822,MDU6SXNzdWUzNDM2NTk4MjI=,2304,float32 instead of float64 when decoding int16 with scale_factor netcdf var  using xarray ,971382,closed,0,,,32,2018-07-23T14:35:12Z,2024-03-15T16:31:06Z,2024-03-15T16:31:06Z,NONE,,,,"#### Code Sample :
Considering a netcdf file file with the following variable:
```
short agc_40hz(time, meas_ind) ;
        agc_40hz:_FillValue = 32767s ;
        agc_40hz:units = ""dB"" ;
        agc_40hz:scale_factor = 0.01 ;
```

Code: 

```python
from netCDF4 import Dataset
import xarray as xr

d = Dataset(""test.nc"")
a = d.variables['agc_40hz'][:].flatten()[69] ## 21.940000000000001 'numpy.float64'
x = xr.open_dataset(""test.nc"")
b = x['agc_40hz'].values.flatten()[69] ## 21.939998626708984 'numpy.float32'
abs(a - b) # 0.000001373291017
```
#### Problem description :

## Different behaviour of xarray comparing to netCDF4 Dataset
When reading the dataset with xarray we found that the decoded type was numpy.float32 instead of numpy.float64
This netcdf variable has an int16 dtype
when the variable is read with the netCDF4 library directly, it is automatically converted to numpy.float64.
in our case we loose on precision when using xarray.
We found two solutions for this:

## First solution :

This solution aims to prevent auto_maskandscale

```python
d = Dataset(""test.nc"")
a = d.variables['agc_40hz'][:].flatten()[69] ## 21.940000000000001 'numpy.float64'
x = xr.open_dataset(""test.nc"", mask_and_scale=False, decode_times=False)
b = x['agc_40hz'].values.flatten()[69] ## 21.940000000000001 'numpy.float64'
abs(a - b) # 0.000000000000000
```

Modification in xarray/backends/netCDF4_.py line 241

```python
def _disable_auto_decode_variable(var):
    """"""Disable automatic decoding on a netCDF4.Variable.

    We handle these types of decoding ourselves.
    """"""
    pass
    # var.set_auto_maskandscale(False)
    # # only added in netCDF4-python v1.2.8
    # with suppress(AttributeError):
    #     var.set_auto_chartostring(False)
```

## Second solution :

This solution uses numpy.float64 whatever integer type provided.

```python
d = Dataset(""test.nc"")
a = d.variables['agc_40hz'][:].flatten()[69] ## 21.940000000000001 'numpy.float64'
x = xr.open_dataset(""test.nc"")
b = x['agc_40hz'].values.flatten()[69] ## 21.940000000000001 'numpy.float64'
abs(a - b) # 0.000000000000000
```

Modification in xarray/core/dtypes.py line 85

```python
def maybe_promote(dtype):
    """"""Simpler equivalent of pandas.core.common._maybe_promote

    Parameters
    ----------
    dtype : np.dtype

    Returns
    -------
    dtype : Promoted dtype that can hold missing values.
    fill_value : Valid missing value for the promoted dtype.
    """"""
    # N.B. these casting rules should match pandas
    if np.issubdtype(dtype, np.floating):
        fill_value = np.nan
    elif np.issubdtype(dtype, np.integer):
        #########################
        #OLD CODE BEGIN
        #########################
        # if dtype.itemsize <= 2:
        #     dtype = np.float32
        # else:
        #     dtype = np.float64
        #########################
        #OLD CODE END
        #########################

        #########################
        #NEW CODE BEGIN
        #########################
        dtype = np.float64 # whether it's int16 or int32 we use float64
        #########################
        #NEW CODE END
        #########################
        fill_value = np.nan
    elif np.issubdtype(dtype, np.complexfloating):
        fill_value = np.nan + np.nan * 1j
    elif np.issubdtype(dtype, np.datetime64):
        fill_value = np.datetime64('NaT')
    elif np.issubdtype(dtype, np.timedelta64):
        fill_value = np.timedelta64('NaT')
    else:
        dtype = object
        fill_value = np.nan
    return np.dtype(dtype), fill_value
```

Solution number 2 would be great for us.
At this point we don't know if this modification would introduce some side effects.
Is there another way to avoid this problem ?

#### Expected Output
In our case we expect the variable to be in numpy.float64 as it is done by netCDF4.

#### Output of ``xr.show_versions()``

<details>
INSTALLED VERSIONS
------------------
commit: None
python: 3.5.2.final.0
python-bits: 64
OS: Linux
OS-release: 4.15.0-23-generic
machine: x86_64
processor: x86_64
byteorder: little
LC_ALL: None
LANG: en_US.UTF-8
LOCALE: en_US.UTF-8

xarray: 0.10.8
pandas: 0.23.3
numpy: 1.15.0rc2
scipy: 1.1.0
netCDF4: 1.4.0
h5netcdf: None
h5py: None
Nio: None
zarr: None
bottleneck: None
cyordereddict: None
dask: 0.18.1
distributed: 1.22.0
matplotlib: 2.2.2
cartopy: None
seaborn: None
setuptools: 40.0.0
pip: 10.0.1
conda: None
pytest: 3.6.3
IPython: 6.4.0
sphinx: None
</details>","{""url"": ""https://api.github.com/repos/pydata/xarray/issues/2304/reactions"", ""total_count"": 0, ""+1"": 0, ""-1"": 0, ""laugh"": 0, ""hooray"": 0, ""confused"": 0, ""heart"": 0, ""rocket"": 0, ""eyes"": 0}",,completed,13221727,issue
407746874,MDExOlB1bGxSZXF1ZXN0MjUxMTUwNTk2,2751,Ability to force float64 instead of float32 issue #2304,971382,open,0,,,16,2019-02-07T15:10:58Z,2022-06-09T14:50:17Z,,FIRST_TIMER,,0,pydata/xarray/pulls/2751,"Ability to promote to float64 instead of float32 when dealing with int variables with scale_factor.
added parameter force_promote_float64 (False by default) to open_dataset and open_dataarray that enables this behaviour when True.

 - [x] Closes #2304 
 - [x] Tests added
 - [x] Fully documented, including `whats-new.rst` for all changes and `api.rst` for new API
","{""url"": ""https://api.github.com/repos/pydata/xarray/issues/2751/reactions"", ""total_count"": 0, ""+1"": 0, ""-1"": 0, ""laugh"": 0, ""hooray"": 0, ""confused"": 0, ""heart"": 0, ""rocket"": 0, ""eyes"": 0}",,,13221727,pull