Xarray engine: temporal dimensions - seasonal forecast

[1]:

import earthkit.data as ekd

ds_fl = ekd.from_source("sample", "seasonal_monthly.grib").to_fieldlist()

The input data contains seasonal monthly forecast. Because the length of a month varies, for this data the time.forecast_month key is better suited for describing the temporal structure than using the ``step`` keys.

This is how the first few GRIB messages look like:

[2]:

ds_fl[0:4].ls(extra_keys="time.forecast_month")

[2]:
parameter.variable time.valid_datetime time.base_datetime time.step vertical.level vertical.level_type ensemble.member geography.grid_type time.forecast_month
0 2t 1993-11-01 1993-10-01 31 days 0 surface 0 regular_ll 1
1 2t 1993-11-01 1993-10-01 31 days 0 surface 1 regular_ll 1
2 2t 1993-11-01 1993-10-01 31 days 0 surface 2 regular_ll 1
3 2t 1993-12-01 1993-10-01 61 days 0 surface 0 regular_ll 2

In order to use time.forecast_month instead of step we need to use the dim_roles option in to_xarray().

[3]:

ds = ds_fl.to_xarray(time_dims=["forecast_reference_time", "step"], dim_roles={"step": "time.forecast_month"})
ds

[3]:

<xarray.Dataset> Size: 395kB
Dimensions:                  (member: 3, forecast_reference_time: 4, step: 6,
                              latitude: 19, longitude: 36)
Coordinates:
  * member                   (member) <U1 12B '0' '1' '2'
  * forecast_reference_time  (forecast_reference_time) datetime64[ns] 32B 199...
  * step                     (step) int64 48B 1 2 3 4 5 6
  * latitude                 (latitude) float64 152B 90.0 80.0 ... -80.0 -90.0
  * longitude                (longitude) float64 288B 0.0 10.0 ... 340.0 350.0
Data variables:
    2t                       (member, forecast_reference_time, step, latitude, longitude) float64 394kB ...
Attributes:
    Conventions:  CF-1.8
    institution:  ECMWF

When we check the “step” dimension we can see its units are “months”.

[4]:

print(ds["step"])

<xarray.DataArray 'step' (step: 6)> Size: 48B
array([1, 2, 3, 4, 5, 6])
Coordinates:
  * step     (step) int64 48B 1 2 3 4 5 6
Attributes:
    units:    months

By default, the dimensions related to dimension roles are named after the roles. So, although the step dimension was generated from the “time.forecast_month” key the dimension name is still “step”. To override this use the dim_name_from_role_name=False option in to_xarray().

[5]:

ds = ds_fl.to_xarray(
    time_dims=["forecast_reference_time", "step"],
    dim_roles={"step": "time.forecast_month"},
    dim_name_from_role_name=False,
)
ds

[5]:

<xarray.Dataset> Size: 395kB
Dimensions:                  (member: 3, forecast_reference_time: 4,
                              forecast_month: 6, latitude: 19, longitude: 36)
Coordinates:
  * member                   (member) <U1 12B '0' '1' '2'
  * forecast_reference_time  (forecast_reference_time) datetime64[ns] 32B 199...
  * forecast_month           (forecast_month) int64 48B 1 2 3 4 5 6
  * latitude                 (latitude) float64 152B 90.0 80.0 ... -80.0 -90.0
  * longitude                (longitude) float64 288B 0.0 10.0 ... 340.0 350.0
Data variables:
    2t                       (member, forecast_reference_time, forecast_month, latitude, longitude) float64 394kB ...
Attributes:
    Conventions:  CF-1.8
    institution:  ECMWF
[ ]: