number of dry days

[TarekAMZ]

Setup Information

• Xclim version:

Context

I'm trying to calculate the number of dry days using xclim.indicators.atmos.dry_days but I get the results as float64 (e.g., 15 days, 193 days, etc.). Is that how it supposed to be?

Steps To Reproduce

models = ['ACCESS-CM2','BCC-CSM2-MR', 'CMCC-CM2-SR5', 'CMCC-ESM2', 'CNRM-ESM2-1', 'FGOALS-g3', 'GFDL-ESM4', 'INM-CM5-0', 'KACE-1-0-G', 'MIROC6', 'MIROC-ES2L','MPI-ESM1-2-LR', 'MRI-ESM2-0']
ssps = ['historical','ssp245','ssp370','ssp585']

for model in models:
for ssp in ssps:
print(model, ssp)
#ds=xr.open_dataset('pr/daily/pr_day_'+model+''+ssp+'.nc')
ds=xr.open_dataset('pr/daily/pr_day'+model+''+ssp+'.nc')
#ds1 = units.convert_units_to(ds.pr, "mm")
ndd = xclim.indicators.atmos.dry_days(ds.pr, thresh = '0.1 mm/d', freq = 'YS')
ndd.to_netcdf('pr/daily/ndd/ndd_annual'+model+'_'+ssp+'.nc')

[aulemahal]

I'm not sure I fully understand your issue here, but I see two answers that might solve it.

Floats instead of integers

Are you surprised that ndd in your example (NOT the netcdf, but before), is in float instead of int ?

This is because xarray and xclim have special checks for missing values. If your data had missing values (for example if it stopped in november instead of december) then the incomplete periods would be flagged as invalid. Xclim would put a NaN in the output to indicate that. But NaNs are only available when your data has a float dtype. Thus, xclim (and xarray underneath) always return floats in any case where missing data is a possibility.

Timedelta instead of numbers

If instead you meant that when you open the netcdf again, you see "timedelta64[ns]" as the dtype instead of a normal number dtype (float, int), this is because xarray thinks it is smart. xr.open_dataset sees the units that xclim added to the dataset : "days" and xarray decides to interpret that litterally and returns a "timedelta" array.

To avoid that open the netCDF with xr.open_dataset('file.nc', decode_timedelta=False).