Dataset.map_blocks(func, args=(), kwargs=None, template=None)[source]#

Apply a function to each block of this Dataset.


This method is experimental and its signature may change.

  • func (callable()) – User-provided function that accepts a Dataset as its first parameter. The function will receive a subset or ‘block’ of this Dataset (see below), corresponding to one chunk along each chunked dimension. func will be executed as func(subset_dataset, *subset_args, **kwargs).

    This function must return either a single DataArray or a single Dataset.

    This function cannot add a new chunked dimension.

  • args (sequence) – Passed to func after unpacking and subsetting any xarray objects by blocks. xarray objects in args must be aligned with obj, otherwise an error is raised.

  • kwargs (Mapping or None) – Passed verbatim to func after unpacking. xarray objects, if any, will not be subset to blocks. Passing dask collections in kwargs is not allowed.

  • template (DataArray, Dataset or None, optional) – xarray object representing the final result after compute is called. If not provided, the function will be first run on mocked-up data, that looks like this object but has sizes 0, to determine properties of the returned object such as dtype, variable names, attributes, new dimensions and new indexes (if any). template must be provided if the function changes the size of existing dimensions. When provided, attrs on variables in template are copied over to the result. Any attrs set by func will be ignored.


  • A single DataArray or Dataset with dask backend, reassembled from the outputs of the

  • function.


This function is designed for when func needs to manipulate a whole xarray object subset to each block. Each block is loaded into memory. In the more common case where func can work on numpy arrays, it is recommended to use apply_ufunc.

If none of the variables in this object is backed by dask arrays, calling this function is equivalent to calling func(obj, *args, **kwargs).


Calculate an anomaly from climatology using .groupby(). Using xr.map_blocks() allows for parallel operations with knowledge of xarray, its indices, and its methods like .groupby().

>>> def calculate_anomaly(da, groupby_type="time.month"):
...     gb = da.groupby(groupby_type)
...     clim = gb.mean(dim="time")
...     return gb - clim
>>> time = xr.cftime_range("1990-01", "1992-01", freq="M")
>>> month = xr.DataArray(time.month, coords={"time": time}, dims=["time"])
>>> np.random.seed(123)
>>> array = xr.DataArray(
...     np.random.rand(len(time)),
...     dims=["time"],
...     coords={"time": time, "month": month},
... ).chunk()
>>> ds = xr.Dataset({"a": array})
>>> ds.map_blocks(calculate_anomaly, template=ds).compute()
Dimensions:  (time: 24)
  * time     (time) object 1990-01-31 00:00:00 ... 1991-12-31 00:00:00
    month    (time) int64 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12
Data variables:
    a        (time) float64 0.1289 0.1132 -0.0856 ... 0.2287 0.1906 -0.05901

Note that one must explicitly use args=[] and kwargs={} to pass arguments to the function being applied in xr.map_blocks():

>>> ds.map_blocks(
...     calculate_anomaly,
...     kwargs={"groupby_type": "time.year"},
...     template=ds,
... )
Dimensions:  (time: 24)
  * time     (time) object 1990-01-31 00:00:00 ... 1991-12-31 00:00:00
    month    (time) int64 dask.array<chunksize=(24,), meta=np.ndarray>
Data variables:
    a        (time) float64 dask.array<chunksize=(24,), meta=np.ndarray>