Incorrect 'cumsum' results when chunking before loading.

[muttener]

What happened?

I wanted to calculate a cumsum on a dataset using dask arrays. The results differ from a cumsum on the same data opened as a pure numpy array.

An example plot of the difference between the two approaches (cumsum over time):

There are 3 chunks in time. The results start to diverge at the start of the second chunk.

What did you expect to happen?

Using pure numpy arrays or dask arrays should yield results that are numerically close(r).

Minimal Complete Verifiable Example

import xarray as xr
xr.show_versions()

from pathlib import Path

import matplotlib.pyplot as plt
import numpy as np

# Save a non-trivial dataset to disk.
# Define explicit chunksize to prevent a UserWarning coming from xarray.namedarray.utils
# (""The specified chunks separate the stored chunks along ..")
path = Path("~/tmp/random.h5")
shape = (200, 1500)
data = 1000 * np.random.rand(*shape).astype("float32")
ds = xr.DataArray(data, dims=("position", "time"), name="data").to_dataset()
encoding = {
    "data": {
        "chunksizes": (200, 500),
    }
}
ds.to_netcdf(path, engine="h5netcdf", encoding=encoding)


# Open the dataset using numpy and dask arrays.
ds = xr.open_dataset(
    path,
    engine="h5netcdf",
    chunks=None,
)
chunk_size = 500
ds_chunked = xr.open_dataset(
    path,
    engine="h5netcdf",
    chunks={"time": chunk_size},
)

# Show the difference.
(ds.cumsum(dim="time").data - ds_chunked.cumsum(dim="time").data).plot()
plt.axvline(x=chunk_size, color="red", linestyle="--", label="start of second chunk")
plt.title("Difference of `cumsum` on a numpy and a dask array.")
plt.legend()
plt.show()

Steps to reproduce

No response

MVCE confirmation

• Minimal example — the example is as focused as reasonably possible to demonstrate the underlying issue in xarray.
• Complete example — the example is self-contained, including all data and the text of any traceback.
• Verifiable example — the example copy & pastes into an IPython prompt or Binder notebook, returning the result.
• New issue — a search of GitHub Issues suggests this is not a duplicate.
• Recent environment — the issue occurs with the latest version of xarray and its dependencies.

Relevant log output

Anything else we need to know?

Changing xr.open_dataset to xr.load_dataset for the chunked dataset does yield the expected result.

Environment

Details

INSTALLED VERSIONS

commit: None
python: 3.11.14 (main, Dec 17 2025, 21:07:37) [Clang 21.1.4 ]
python-bits: 64
OS: Linux
OS-release: 6.12.57+deb13-amd64
machine: x86_64
processor:
byteorder: little
LC_ALL: None
LANG: en_US.UTF-8
LOCALE: ('en_US', 'UTF-8')
libhdf5: 2.0.0
libnetcdf: None

xarray: 2026.2.0
pandas: 3.0.1
numpy: 2.4.3
scipy: 1.17.1
netCDF4: None
pydap: None
h5netcdf: 1.8.1
h5py: 3.16.0
zarr: None
cftime: 1.6.5
nc_time_axis: None
iris: None
bottleneck: 1.6.0
dask: 2026.3.0
distributed: 2026.3.0
matplotlib: 3.10.8
cartopy: None
seaborn: None
numbagg: None
fsspec: 2026.2.0
cupy: None
pint: None
sparse: None
flox: None
numpy_groupies: None
setuptools: 80.9.0
pip: 25.3
conda: None
pytest: 9.0.2
mypy: 1.19.1
IPython: 9.10.0
sphinx: 9.0.4

[welcome]

Thanks for opening your first issue here at xarray! Be sure to follow the issue template!
If you have an idea for a solution, we would really welcome a Pull Request with proposed changes.
See the Contributing Guide for more.
It may take us a while to respond here, but we really value your contribution. Contributors like you help make xarray better.
Thank you!