ECMWF ensemble inconsistent dataset order

[williamhobbs]

For some ECMWF ensemble fxx values, the first dataset is all members and the second is the mean, but for other fxx values, the order is swapped. Is this expected?

Started with https://herbie.readthedocs.io/en/latest/gallery/ecmwf_models/ecmwf.html#Ensemble-Forecast-Products.

Here's an example:

init_time = '2024-04-10 00:00'
fxx=15
ds = Herbie(date=init_time, model='ifs',product='enfo',fxx=fxx).xarray(":100[uv]:")

ds[0] returns all 50 members:

and ds[1] returns (presumably?) the mean:

But, if I change fxx, I sometimes get the opposite:

fxx=18
ds = Herbie(date=init_time, model='ifs',product='enfo',fxx=fxx).xarray(":100[uv]:")

[williamhobbs]

As a workaround to get just the dataset with all the members, I check to see which one has number in the dimensions:

if 'number' in list(ds[1].dims):
    ds = ds[1] 
else:
    ds = ds[0]

I'm sure there's a better way, but I think this works. And it could be extended a bit to re-order the datasets into a consistent order.