How to change the order in dimensions xarray.Dataset?

Question

I am creating a xarray dataset as below:

import numpy as np
import xarray as xr

x_example = np.random.rand(1488,)
y_example = np.random.rand(1331,)
time_example = np.random.rand(120,)
rainfall_example = np.random.rand(120, 1331, 1488)

rainfall_dataset = xr.Dataset(
    data_vars=dict(
        rainfall_depth=(['time', 'y', 'x'], rainfall_example),
    ),
    coords=dict(
        time=(['time'], time_example),
        x=(['x'], x_example),
        y=(['y'], y_example)
    )
)

The results are like this

And the dimensions when I run rainfall_example.dims are like this Frozen({'time': 120, 'y': 1331, 'x': 1488}) (this can also be seen in the above results). I know the xarray.Dataset.dims cannot be modified according to here

My question is: How can we change the order of those dimensions into the dimensions like this Frozen({'time': 120, 'x': 1488, 'y': 1331}) without changing anything else (everything will be the same only the order in dimensions is changed)?

Selfsustaining

related: #69745606 – Cheep

Answer 1

You can reorder your coordinates and variables by selecting them both in order using a list:

In [3]: rainfall_dataset[["time", "y", "x", "rainfall_depth"]]
Out[3]:
<xarray.Dataset>
Dimensions:         (time: 120, y: 1331, x: 1488)
Coordinates:
  * time            (time) float64 0.2848 0.7556 0.9501 ... 0.694 0.734 0.198
  * y               (y) float64 0.1941 0.1132 0.2504 ... 0.1501 0.5085 0.006135
  * x               (x) float64 0.2776 0.4504 0.1886 ... 0.4071 0.3327 0.5555
Data variables:
    rainfall_depth  (time, y, x) float64 ...

Cheep

Thanks @Michael, this way will change the shape of my rainfall_depth variable (from (time, y, x) to (time, x, y)), so it does not work for me unfortunately. As I mentioned, I wish to keep everything the same, only the order of in dimensions is changed. – Selfsustaining

So sorry - I misread your question. Does simply calling transpose work? – Cheep

Or perhaps transpose, followed by re-ordering the coordinates back to the original using this list method I describe? – Cheep

Do you mean this rainfall_dataset.transpose()? If it is, it did not work either. It still changes the shape of my rainfall_depth variable and the order of dimensions are still the same. – Selfsustaining

Sorry - I guess I don’t understand your question. Can you describe exactly what you want as a result? – Cheep

It actually works when I tried as you suggested re-ordering the coordinates rainfall_dataset.transpose("time", "y", "x")[["time", "x", "y", "rainfall_depth"]]. And this is exactly what I am looking for. Thank you so much @Michael. If it's possible, you can change the answer and then I will upvote for it. – Selfsustaining

Got it. Then you don’t actually need the transpose. Updated my answer. – Cheep

For some reason this seems to drop all the data variables out of my dataset... – Shelbashelbi

Hmm. It shouldn’t! Feel free to ask another question and post your code and results if you don’t figure it out – Cheep

the reorder list works for the "upper level". But how do i make the rainfall_depth DataArray follow the new order? ie rainfall_depth (time, x, y)? – Melioration