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
1136315478,I_kwDOAMm_X85DusxW,6272,ds.to_netcdf() changes values of variable,2042458,closed,0,,,12,2022-02-13T19:15:20Z,2024-03-25T14:56:51Z,2024-03-25T14:56:50Z,NONE,,,,"### What happened?

I am very puzzled by an odd behavior of `ds.to_netcdf()` that modifies the value of a variable by offseting partly the vlue by 44500. 

I have list of file containing meteorological variables organized in three dimensions `'time', 'longitude', latitude'`. The files are loaded using `xr.open_mfdataset('*.nc')`. No problem here. The dataset is loaded in chuncks using Dask.  

```
In [11]: ds
Out[11]: 
<xarray.Dataset>
Dimensions:    (longitude: 10, latitude: 7, level: 14, time: 35760)
Coordinates:
  * longitude  (longitude) float32 6.5 6.75 7.0 7.25 7.5 7.75 8.0 8.25 8.5 8.75
  * latitude   (latitude) float32 61.25 61.0 60.75 60.5 60.25 60.0 59.75
  * level      (level) int32 600 650 700 750 775 800 ... 900 925 950 975 1000
  * time       (time) datetime64[ns] 1978-10-01 ... 2021-08-31T23:00:00
Data variables:
    z          (time, level, latitude, longitude) float32 dask.array<chunksize=(744, 14, 7, 10), meta=np.ndarray>
    t          (time, level, latitude, longitude) float32 dask.array<chunksize=(744, 14, 7, 10), meta=np.ndarray>
    u          (time, level, latitude, longitude) float32 dask.array<chunksize=(744, 14, 7, 10), meta=np.ndarray>
    v          (time, level, latitude, longitude) float32 dask.array<chunksize=(744, 14, 7, 10), meta=np.ndarray>
    r          (time, level, latitude, longitude) float32 dask.array<chunksize=(744, 14, 7, 10), meta=np.ndarray>
    q          (time, level, latitude, longitude) float32 dask.array<chunksize=(744, 14, 7, 10), meta=np.ndarray>
Attributes:
    Conventions:  CF-1.6
    history:      2021-12-08 19:22:33 GMT by grib_to_netcdf-2.23.0: /opt/ecmw...
```

Now, I would like to save a subset of this dataset to a netcdf file as follow: `ds.isel(latitude[1,2,3], longitude=[3,4,5]).to_netcdf('sub.nc')`. So far nothing particular.

The value `z` is a `float32` which varies from 2000 to -2000 along the time dimension. After being saved in the subsample, `z` is still a `float32` but the values that are less than -1000 are being offset by 44500. However, if I do `(ds.z.isel(latitude[1,2,3], longitude=[3,4,5])*1).to_netcdf('sub.nc')`,  instead, then all values in the subsampled netcdf are fine.  I am very puzzled by this behavior. Could this be an odd behavior of dask chuncks and `to_netcdf()`?

### What did you expect to happen?

I expected no modification of the data after saving to netcdf, no matter what.

### Minimal Complete Verifiable Example

I will share files upon request. 

### Relevant log output

_No response_

### Anything else we need to know?

_No response_

### Environment


INSTALLED VERSIONS
------------------
commit: None
python: 3.9.7 (default, Sep 16 2021, 13:09:58) 
[GCC 7.5.0]
python-bits: 64
OS: Linux
OS-release: 5.13.0-28-generic
machine: x86_64
processor: x86_64
byteorder: little
LC_ALL: None
LANG: en_US.UTF-8
LOCALE: ('en_US', 'UTF-8')
libhdf5: 1.10.6
libnetcdf: 4.8.1

xarray: 0.20.1
pandas: 1.3.5
numpy: 1.20.3
scipy: 1.7.3
netCDF4: 1.5.7
pydap: None
h5netcdf: 0.11.0
h5py: 3.6.0
Nio: None
zarr: None
cftime: 1.5.1.1
nc_time_axis: None
PseudoNetCDF: None
rasterio: 1.2.8
cfgrib: None
iris: None
bottleneck: 1.3.2
dask: 2021.10.0
distributed: 2021.10.0
matplotlib: 3.5.0
cartopy: None
seaborn: 0.11.2
numbagg: None
fsspec: 2022.01.0
cupy: None
pint: None
sparse: None
setuptools: 58.0.4
pip: 21.2.4
conda: None
pytest: None
IPython: 7.31.1
sphinx: None
","{""url"": ""https://api.github.com/repos/pydata/xarray/issues/6272/reactions"", ""total_count"": 0, ""+1"": 0, ""-1"": 0, ""laugh"": 0, ""hooray"": 0, ""confused"": 0, ""heart"": 0, ""rocket"": 0, ""eyes"": 0}",,completed,13221727,issue