How to get the index corresponding to the maximum for a resampled array?

[AndrewILWilliams]

Hi there,

I have a DataArray of precipitation with time, lat, lon coordinates, and I would like to find the time index which corresponds to the maximum precipitation value in a given year.

>>> print(da_hist)

<xarray.DataArray 'pr' (time: 56978, lat: 64, lon: 128)>
dask.array<open_dataset-aac2e5d3ea1a310f230a62870ced1941pr, shape=(56978, 64, 128), dtype=float32, chunksize=(56978, 64, 128), chunktype=numpy.ndarray>
Coordinates:
  * time     (time) datetime64[ns] 1850-01-01T12:00:00 ... 2005-12-31T12:00:00
  * lat      (lat) float64 -87.86 -85.1 -82.31 -79.53 ... 79.53 82.31 85.1 87.86
  * lon      (lon) float64 0.0 2.812 5.625 8.438 ... 348.8 351.6 354.4 357.2
Attributes:
    standard_name:     precipitation_flux
    long_name:         Precipitation
    comment:           at surface; includes both liquid and solid phases from...
    units:             kg m-2 s-1
    original_name:     PRCP
    original_units:    kg/m**2/s
    history:           2011-10-04T11:32:54Z altered by CMOR: Converted units ...
    cell_methods:      time: mean
    cell_measures:     area: areacella
    associated_files:  baseURL: http://cmip-pcmdi.llnl.gov/CMIP5/dataLocation...

I thought that perhaps I could use da_hist.resample(time='1Y').argmax('time'), but it seems that argmax hasn't been implemented for resampled objects yet?

---------------------------------------------------------------------------
AttributeError                            Traceback (most recent call last)
<ipython-input-71-6c5d305923f9> in <module>
----> 1 da_hist.resample(time='1Y').argmax('time').compute()

AttributeError: 'DataArrayResample' object has no attribute 'argmax'

Does anyone know a good way around this? Thanks!

[AndrewILWilliams]

Follow up, I'd also like to be able to do this for different month groupings as well. i.e., what is the time-index corresponding to the maximum precipitation value within May-September (MJJAS) in each year.

I think I should be able to do this with some clever masking, but I'd still require to be able to use resample().argmax()...

mjjas_mask = (da_hist.time.dt.month >= 5) & (da_hist.time.dt.month <= 9)
da_hist.where(mjjas_mask).resample(time='1Y').argmax('time').compute()

[dcherian]

Yes, this is a regression. We'll have to add arg* to the GroupBy class i guess. #4476

[AndrewILWilliams]

Cool! I might have time to take a look later this month depending on how the PhD is going..

[AndrewILWilliams]

Ok, I'm confused... it seems like you USED to be able to do this in v0.15.1, but you can't now?

[keewis]

yes, we did a refactor of argmin / argmax, but the implementations for GroupBy / Resample objects (probably also Coarsen) are still missing. PRs welcome.

[DamienIrving]

Is there a work around to achieve the old xarray.core.resample.DataArrayResample.argmax functionality in the meantime? Perhaps passing numpy.argmax to xarray.core.resample.DatasetResample.reduce would work?