Deseasonalization: climatology and anomalies (s2stools.clim)

This is to compute a climatology from s2s data.

Functions

s2stools.clim.climatology(data: DataArray | Dataset, window_size: int = 15, mean_or_std: str = 'mean', ndays_clim_filter: int = 7, hide_warnings: bool = False, groupby: str = 'leadtime', dim_number_non_exist: bool = False)

Compute anomalies from the climatological mean. Deseasonalization is based on hindcasts.

Parameters:

  • data (xr.Dataset or xr.DataArray) – The raw data.

  • window_size (int) – The mean is constructed using all reftimes within this plus-minus-day-interval.

  • mean_or_std (str) – either ‘mean’ or ‘std’

  • ndays_clim_filter (int) – Apply running mean to the climatology.

  • hide_warnings (boolean) – e.g. for UKMO it can happen that one is interested in the climatology for 2020-11-01, but hindcasts are available only at 2020-11-05; by default, the climatology is computed for 2020-11-05 onwards and the missing previous 4 days are linearly extrapolated; if this happens, hide_warnings=False will warn you!

  • groupby (str) –

    must be ‘leadtime’ or ‘validtime’

    If “leadtime”, create climatology along same leadtime, e.g., different forecasts but always for leaditme=+4 days, leadtime=+5 days, …; If “validtime”, create climatology along same day-of-year, e.g., different forecasts but always for March 03, March 04, March 05. Defaults to “leadtime”.

  • dim_number_non_exist (bool) – Depracated, because it is now automatically checked whether adding dimension “number” is neceesary. If True, an ensemble member dimension “number” is added, because it is required for deseasonalization. Defaults to False.

Returns:

climatology – climatological mean or climatological std based on hindcasts

Return type:

xr.DataArray or xr.Dataset

Warning

If reftimes +- time window are not available then anomalies are still computed, but climatology is less robust.

Warning

In latest version, groupby=”validtime” is not yet implemented to work with the new xarray version. Therefore, default changed to groupby=”leadtime”.

Notes

There my be use cases where no running mean should be used, e.g., for computing anomalies of forecast variance, which grows non-linearly! Moreover, if anomalies of variance are computed, make sure that groupby is set to “leadtime”, because spread is more sensitive to the leadtime of course than to the day of year.

s2stools.clim.deseasonalize(data, window_size=15, standardize=False, ndays_clim_filter=7, hide_print=True, hide_plot=True, hide_warnings=False, return_clim_lists=False, dim_number_non_exist=False)

Compute anomalies from the climatological mean. Deseasonalization is based on hindcasts.

Parameters:

  • data (xr.Dataset or xr.DataArray) – The raw data.

  • window_size (int) – The mean is constructed using all reftimes within this plus-minus-day-interval.

  • standardize (bool) – If True, compute standardized anomalies.

  • ndays_clim_filter (int) – Apply running mean to the climatology.

  • hide_print (bool) – If False, print how many reftimes are used for each climatology. Defaults to True.

  • hide_plot (bool) – If False, plot the climatology.

  • hide_warnings (bool) –

    ???

  • return_clim_lists (bool) – If True, also return the constructed climatologies. If False, only return anomalies.

  • dim_number_non_exist (bool) – If True, an ensemble member dimension “number” is added, because it is required for deseasonalization. Defaults to False.

Returns:

Anomalies If return_clim_lists=True, return tuple anom, clim_list, climstd_list.

Return type:

xr.Dataset or xr.DataArray

Warning

Deprecated since version 0.3.0: deseasonalize will be removed in the future, it is replaced by climatology