[Questions] Trying to understand how xclim.indicators.atmos.tn10p works

[seblehner]

Setup Information

• Xclim version: 0.56.0
• Python version: 3.12.2
• Operating System: Ubuntu 20.04.6 LTS

Context

Hello 👋
I am trying to use xclim.indicators.atmos.tn10p but I fear I am not fully understanding the logic behind how it works.

What I would like:

• provide an xr.DataArray as tasmin that contains data at daily resolution.
• provide an xr.DataArray as tasmin_per with precalculated percentiles (i.e. only dims left are (x, y) and no time dim anymore).

This throws an error because it wants to calculate the percentiles itself. However, even if the full data is provided, the following issues arose for me:

• the time dims are required to be the same, so I cannot provide a subset (e.g. a reference period) via .sel for the percentile calculation
• {tasmin_per_window} and {tasmin_per_period} are listed in the doc, but cannot be provided as kwargs, because they get passed through to .select_time which will throw an error
• lastly, I tried simply with tasmin and tasmin_per being the same data (same time dim, etc.), but the resulting output was 0 everywhere

Does anyone have any pointer as to what could be the issue, or what I am misunderstanding here?

Cheers

Steps To Reproduce

No response

[aulemahal]

Hi! I'm afraid I can't reproduce your errors. However, one potential issue I see is that tn10p expects to get a tasmin_per for each day of the year. Thus, tasmin_per should have a dayofyear dimension (instead of time). I am surprised how unclear this is in the docstring, I think there was a time when we wanted the "percentile" indicator to accept both the "dayofyear" version and the "flat" one (the one you want), but I guess the changes were not completed correctly ?

Anyway, so you can test on your end, this code for the tn10p computed over each dayofyear works for me:

import xarray as xr
import xclim as xc

# Get test data, 6-hourly air temp on North America
ds = xr.tutorial.open_dataset("air_temperature")
# Get a daily array with proper units
tasmin = ds.air.rename('tasmin').resample(time='D').mean().assign_attrs(units='K')
# The 10th percentile taken for each day of year, using a rolling 15-day window
tasmin_per = xc.core.calendar.percentile_doy(tasmin, window=15, per=10)

# The number of days in a year that are colder than the 10th percentile for that particular day of year
out = xc.atmos.tn10p(tasmin, tasmin_per)

There is another way to compute what you want. In most indicators, if you see an input annotated as "Quantified" in the docstring and signature, this means it will accept any of the following : a string representing a quantity with units (the most common case), a pint.Quantity object, or a xr.DataArray with units but without time. Thus, you could do:

# A much simpler 10th percentile over all days
tasmin_per = tasmin.quantile(0.1, 'time').assign_attrs(units='K')

# The number of days in a year that are colder than the thresh, i.e. then the all-time 10th percentile
xc.atmos.tn_days_below(tasmin=tasmin, thresh=tasmin_per)

Of course, the attributes here will be a bit more vague as the "10th percentile" information won't be noted anywhere, maybe you'll want to edit them after. But this looks like the computation you want ?

[seblehner]

Thank you for the quick answer!

With the limitation/intended use of only allowing it on a dayofyear basis it makes much more sense! Perhaps I was too naive in assuming it would just work when providing the extra temporal dimension as e.g. season (which was my actual needed variant) instead of dayofyear. I had thought so because it doesn't explicitly mention the dayofyear limitation in the description, or the tasmin_per description, but only at the end as the last sentence under the returns description. My bad!

Using tn_days_below instead is also what I ended up doing, thank you!