This adds a short docstring example to xr.open_mfdataset to showcase how to pass arguments into preprocess for xr.open_mfdataset.

It wasn't very clear before how to do this, and this hopefully will show an easy and minimal pathway to doing so.

• [X] Closes https://github.com/pydata/xarray/discussions/6820

What happened:

When bottleneck is installed in an environment, it seems to ignore the min_periods kwarg on ds.rolling(...).

What you expected to happen:

When using ds.rolling(..., min_periods=1), it should be able to handle an array of length 1. Without bottleneck installed, it returns the original value of a length 1 array. With bottleneck installed, the error is:

python-traceback ValueError: Moving window (=2) must between 1 and 1, inclusive

Minimal Complete Verifiable Example:

With bottleneck installed to environment:

python import xarray as xr ds = xr.DataArray([1], dims='time') ds.rolling(time=2, center=True, min_periods=1).mean()

python-traceback ValueError: Moving window (=2) must between 1 and 1, inclusive

Without bottleneck installed to environment:

```python import xarray as xr ds = xr.DataArray([1], dims='time') ds.rolling(time=2, center=True, min_periods=1).mean()

<xarray.DataArray (time: 1)> array([1.]) Dimensions without coordinates: time ```

Anything else we need to know?:

In an applied case, this came up while working on .groupby('time.dayofyear').map(_rolling), where we map a rolling mean function over a defined N days with min_periods=1. Some climatological days (like leap years) will not have the N day requirement, so the min_period catch handles that, but with bottleneck installed it breaks due to the above issue.

Environment:

This PR adds an example to the xr.map_blocks() docs.

It is modeled off of @rabernat's example here: https://nbviewer.jupyter.org/gist/rabernat/30e7b747f0e3583b5b776e4093266114 and was encouraged to be opened via the gitter.

It might be slightly large for an example, but I think it shows an applied case to use this in and clears up some common pitfalls (e.g. how to pass args/kwargs).

This simply adds the map_blocks example from https://github.com/pydata/xarray/pull/3667 to the changelog.

This adds climpred to the xarray related projects page.

Code Sample, a copy-pastable example if possible

```python import numpy as np

generate data

dateVar = np.arange('2005-02', '2005-06', dtype='datetime64[D]') t = len(dateVar) floatVar = np.random.rand(t, 100) indexVar = np.arange(100) intVar = np.random.randint(1, high=10, size=(t, 100))

create dataset

A = xr.DataArray(floatVar, dims=['time', 'N']) A.name = 'floatVar' B = xr.DataArray(indexVar, dims=['N']) B.name = 'indexVar' C = xr.DataArray(intVar, dims=['time', 'N']) C.name = 'intVar' D = xr.DataArray(dateVar, dims=['time']) D.name = 'dateVar' ds = xr.merge([A,B,C,D]) print(ds)

<xarray.Dataset> Dimensions: (N: 100, time: 120) Dimensions without coordinates: N, time Data variables: floatVar (time, N) float64 0.4223 0.5019 0.8522 ... 0.9338 0.5833 0.09859 indexVar (N) int64 0 1 2 3 4 5 6 7 8 9 10 ... 90 91 92 93 94 95 96 97 98 99 intVar (time, N) int64 9 2 3 6 8 4 8 7 6 4 2 6 ... 3 1 8 3 8 3 5 3 1 6 7 dateVar (time) datetime64[ns] 2005-02-01 2005-02-02 ... 2005-05-31

apply where function

ds.where(ds.indexVar > 50, drop=True)

<xarray.Dataset> Dimensions: (N: 49, time: 120) Dimensions without coordinates: N, time Data variables: floatVar (time, N) float64 0.3381 0.04735 0.464 ... 0.5571 0.5297 0.8106 indexVar (N) float64 51.0 52.0 53.0 54.0 55.0 ... 95.0 96.0 97.0 98.0 99.0 intVar (time, N) float64 5.0 2.0 9.0 5.0 5.0 1.0 ... 1.0 6.0 5.0 4.0 3.0 dateVar (time, N) datetime64[ns] 2005-02-01 2005-02-01 ... 2005-05-31 ```

Problem description

This is motivated by a use-case of dimensions (Time, nParticle) for a Lagrangian particle simulation. In the above code snippet, I filter by some condition on indexVar (e.g., some type of particle).

For variables that contain the same dimension as the one in indexVar (N), it broadcasts fine and there's no dimension changes. In the case of dateVar, which only has dimension time, there is a dim expansion to add N.

Further, data-types are changed (indexVar and intVar from int64 to float64). In my use-case, a variable of type S64 was converted to O.

So the two major issues here: 1. where mis-broadcasts certain variables by adding the filtered dimension (in this case N to dateVar). It should instead ignore variables that don't contain the variable being filtered. 2. where changes data-types of variables after the filter is applied.

Expected Output

python <xarray.Dataset> Dimensions: (N: 49, time: 120) Dimensions without coordinates: N, time Data variables: floatVar (time, N) float64 0.3381 0.04735 0.464 ... 0.5571 0.5297 0.8106 indexVar (N) int64 51 52 53 54 55 ... 95 96 97 98 99 intVar (time, N) int64 5 2 9 5 5 1 ... 1 6 5 4 3 dateVar (time) datetime64[ns] 2005-02-01 2005-02-02 ... 2005-05-31

Output of xr.show_versions()