Reindexing from 2D index

[Felipe Whitaker]

Greetings,

I started using xarray recently to work with weather data. 

Today I downloaded precipitation data from two different sources: ECMWF's SEAS5 seasonal forecast, and observation data from a local weather institute (CPTEC/INPE).

Both are in different spatial resolutions, but this was easily solved by `xr.Dataset.interp` method.

The issue I am facing is that the seasonal forecast data (description below) is indexed by (number, time, step, latitude and longitude), with a `valid_time` 2D index, and I would like to just `valid_time` as an index instead of `time` and `step`, which will enable dropping duplicates (to make less requests, I queried 31 steps forward for every month, so there is some overlap) and facilitate comparison between both datasets (observation data also has a `valid_time` non primary index).

Is there anyway I can transform `valid_time` into an index for both datasets? I have tried using `reindex`, or directly coding the correct selection of days in month to merge them instead of dropping (same effect), but I had no success.

```python

xarray.Dataset

• Dimensions:
  • number: 51 
  • time: 12 
  • step: 31 
  • latitude: 8 
  • longitude: 9
• Coordinates:
  • number
    (number)
    int32
    0 1 2 3 4 5 6 ... 45 46 47 48 49 50
  • time
    (time)
    datetime64[ns]
    2021-01-01 ... 2021-12-01
  • step
    (step)
    timedelta64[ns]
    1 days 2 days ... 30 days 31 days
  • surface
    ()
    float64
    0.0
  • latitude
    (latitude)
    float64
    -16.0 -17.0 -18.0 ... -22.0 -23.0
  • longitude
    (longitude)
    float64
    -54.0 -53.0 -52.0 ... -47.0 -46.0
  • valid_time
    (time, step)
    datetime64[ns]
    2021-01-02 ... 2022-01-01
• Data variables:
  • tp
    (number, time, step, latitude, longitude)
    float32
    0.0001335 0.0002556 ... 0.3486
• Indexes: (5)
• Attributes: (7)```
  

[Felipe Whitaker]

(just noticed that the previous e-mail got badly formatted, sorry)

By chance I stumbled on a past discussing (https://github.com/ecmwf/cfgrib/issues/97#issuecomment-557190695) and noticed the `backend_kwargs`. Using it to indicate that `time` dimension could be set to `valid_time` (below) seems to have done what I wanted. Thanks!

seasonal_backend = xr.open_dataset(

    seasonal_path, engine="cfgrib", backend_kwargs=dict(time_dims=("valid_time",))

)

What isn't clear to me is whether it dropped the first or the last duplicate. From what I tested (below), it seems to have dropped the last, which is exactly what I wanted. 

seasonal_no_backend = xr.open_dataset(seasonal_path, engine="cfgrib")

dup_fev = (

    seasonal_no_backend.sel(time="2021-02-01")

    .isel(number=0, latitude=0, longitude=0, step=29)

    .tp.item()

)

dup_mar = (

    seasonal_no_backend.sel(time="2021-03-01")

    .isel(number=0, latitude=0, longitude=0, step=0)

    .tp.item()

)

seasonal_backend = xr.open_dataset(

    seasonal_path, engine="cfgrib", backend_kwargs=dict(time_dims=("valid_time",))

)

valid_mar = (

    seasonal_backend.sel(valid_time="2021-03-01T00:00:00")

    .isel(number=0, latitude=0, longitude=0)

    .tp.item()

)

dup_fev, dup_mar, valid_mar  # dup_fev == valid_mar

Thanks for the project,

Felipe