沿新维度组合 xarray 数据集变量而不扩展每个变量

Question

I have an xarray with dimensions lat and lon. It includes 17 variables, each one corresponding with a different time step. I would like to expand the dimensions to include time and reduce the number of variables to only one, with 17 time steps. I've tried implementing this with the following code, but I end up with n variables, with each variable including all the time steps and lat, long dimensions.

The original dataset, ds:

all_dates = ['2019-03-29','2019-05-10','2019-06-21','2019-07-19','2019-08-30','2019-10-11','2019-11-22','2020-01-03','2020-02-14','2020-03-27','2020-05-08','2020-06-19','2020-07-17','2020-08-28','2020-10-09','2021-10-08','2021-11-19']

date_list = pd.to_datetime(all_dates)
ds2 = ds.expand_dims(time=date_list)

The new dataset includes an array of 17 timesteps for each data variable, instead of one data variable with 17 time steps:

Is there a way to add a time dimension, and reduce the dataset's n variables to include a single variable corresponding with the n time steps in the time dimension?

Answer 1

You're looking for xr.Dataset.to_array . expand_dims does exactly what it's doing in your post - it copies each array in your data so that they each include the full time dimnension. But you're essentially looking to concatenate all arrays into a single one along the new dimension.

# concatenate the arrays, assign the name "all_images", return to a Dataset,
# and assign the time dimension
ds.to_array("time", name="all_images").to_dataset().assign_coords(
    time=date_list
)

You could also do this with xr.concat by pulling out each array and concatenating them. Using concat may be a more careful method to ensure the correct ordering since you're relabeling your arrays.

# concatenate each image with an explicit order while assigning
# them the coordinate from date_list
xr.concat(
    [ds[f"image{i}" for i in range(1, len(date_list) + 1)],
    dim=pd.Index(date_list, name="time"),
).to_dataset(name="all_images")