Correctly interpolate seasons in Grouper

[saschahofmann]

Pull Request Checklist:

• This PR addresses an already opened issue (for bug fixes / features)
  • This PR fixes only part of Linear interpolation for seasonal bias adjustment leads to non-smooth distribution #2014
• Tests for the changes have been added (for bug fixes / features)
• CHANGELOG.rst has been updated (with summary of main changes)
  • Link to issue (:issue:number) and pull request (:pull:number) has been added

What kind of change does this PR introduce?

This PR adds a line to correctly interpolate seasonal values. It also changes the test_timeseries function that now accepts a calendar argument instead of cftime. Not providing it or providing None is equivalent to cftime=False and calendar='standard to the previous cftime=True. This allows for testing different calendar implementations e.g. 360_day calendars

[saschahofmann]

I just realised that the factor of 1/6 is assuming that all seasons have the same length which in gregorian calendars is not necessarily true but I am not sure it matters too much at least the function should be smooth.

[saschahofmann]

Just to prove that this leads to a smooth result, same input as in the issue:

[github-actions]

Warning

This Pull Request is coming from a fork and must be manually tagged approved in order to perform additional testing.

[saschahofmann]

Weirdly and contrary to what I showed yesterday, today I am still getting clear transitions as if there still wasn't any linear interpolation.

[Zeitsperre]

@saschahofmann We recently changed the layout of xclim to use a src structure. It might be worthwhile to try reinstalling the library.

[saschahofmann]

I reinstalled xclim but I am still getting very similar results to before the "fix". You have any advice on where else I could look?

[Zeitsperre]

I reinstalled xclim but I am still getting very similar results to before the "fix". You have any advice on where else I could look?

Could it be that you have obsolete __pycache__ folders still among your cloned folders? @coxipi is looking into recreating your example based on your branch for validation, but if the tests are working as intended on CI, then it's likely a caching/installation issue.

[coxipi]

I managed to install the environment, for some reason I only had the branch "main" when I cloned the fork yesterday

• I confirmed that the function has the appropriate modifications inside the notebook I'm using

import inspect
print(inspect.getsource(sdba.base.Grouper.get_index))

• I also find that the interpolation is wrong.

I'll try to have look later. Maybe the interp boolean condition is not triggered properly?

[saschahofmann]

I am pretty sure that the get_index function is updated in my notebook. Either I am wrong in expecting a smoother result (it seems to have changed slightly to what I got earlier) or there is something else going on. I will keep investigating

[coxipi]

It's simply interp which can't be "nearest", otherwise no interpolation takes place ... I think our only other option is linear.

from xclim import sdba
QM = sdba.EmpiricalQuantileMapping.train(
    ref, hist, nquantiles=15, group="time.season", kind="+"
)

scen = QM.adjust(sim, extrapolation="constant", interp="nearest")
scen_interp = QM.adjust(sim, extrapolation="constant", interp="linear")
outd = {
    "Reference":ref,
    "Model - biased":hist,
    "Model - adjusted - no interp":scen, 
    "Model - adjusted - linear interp":scen_interp, 
}
for k,da in outd.items(): 
    da.groupby("time.dayofyear").mean().plot(label=k)
plt.legend()

This doesn't reproduce your figure however. It seems your figure above was matching the reference very well, better than what I have even with the linear interpolation. But it does get rid of obvious discontinuities.

[coxipi]

There is clearly something wrong going on. Comparing
hist - scen_month
scen_time - scen_month
scen_season - scen_month
scen_month - scen_month

scen_season is way off

[saschahofmann]

@coxipi I think only mention this in the original issue: my analysis is done with QuantileDeltaMapping instead of EmpiricalQuantileMapping. Here the equivalent chart to yours for that:

season still seem kinda weird

[saschahofmann]

A similar trend becomes apparent when looking at the adjusted - historical (now for EmpiricalQuantileMapping)

[coxipi]

Yes, I have seen simlilar things by playing with the choice of how get_index. I feel this should not be this sensitive. Let me try and get this back

[saschahofmann]

@Zeitsperre I added these two lines to allow linear interpolation with dayofyear but I am wondering whether this combination makes sense?
https://github.com/Ouranosinc/xclim/pull/2019/files#diff-3fdadf9776d54afe05de332ab8810fc3d2939d75bd98273f3661f05d71f296a4R314-R315

It was failing in a test

[Zeitsperre]

I added these two lines to allow linear interpolation with dayofyear but I am wondering whether this combination makes sense? https://github.com/Ouranosinc/xclim/pull/2019/files#diff-3fdadf9776d54afe05de332ab8810fc3d2939d75bd98273f3661f05d71f296a4R314-R315

It was failing in a test

I'm going to defer to @coxipi or @aulemahal here. This is a bit out of my depth haha

[coxipi]

@Zeitsperre I added these two lines to allow linear interpolation with dayofyear but I am wondering whether this combination makes sense? https://github.com/Ouranosinc/xclim/pull/2019/files#diff-3fdadf9776d54afe05de332ab8810fc3d2939d75bd98273f3661f05d71f296a4R314-R315

It was failing in a test

I don't think there would be a point in doing linear interpolation for dayofyear. We have adjusting factors for each dayofyear, so we don't have in-between values where we need to interpolate the training data. I would say we probably need to change the failing test, let me see

I don't see the failing, can you point to a specific commit?

[saschahofmann]

Yes. Its this test TestExtremeValues.test_real_data: https://github.com/Ouranosinc/xclim/actions/runs/12985440090/job/36210393696

I fixed it by allowing again linear interp of dayofyear. I think it was allowed previously but I slightly restructured the code so now by default it was disabled.

[saschahofmann]

I reverted the change that allows linear interp with dayofyear grouping and instead changed the test to use nearest

[saschahofmann]

@coxipi I think maybe reenabling linear interpolation from dayofyear for now is a better solution. Since otherwise it represents a breaking change, e.g. right now the tutorial does some similar (hence the breaking test in sdba-advanced.ipynb) where we do EQM with dayofyer with linear interp. Instead I could add a deprecation warning for maybe 0.57.0?

[review-notebook-app]

Check out this pull request on 

See visual diffs & provide feedback on Jupyter Notebooks.

---

Powered by ReviewNB

[coveralls]

coverage: 89.955% (+0.009%) from 89.946%
when pulling f518971 on saschahofmann:fix-#2014
into b0b2193 on Ouranosinc:main.

[saschahofmann]

@aulemahal or @coxipi . Would love to get this over the line I think if you agree I add a DeprecationWarning and then ready to merge?

[coxipi]

Hi @saschahofmann sorry for the late response, we were busy with a seminar last week and I missed your message

One thing I'm starting to realize is that the linear interpolation can act on two fronts, on the grouping dimension (e.g. season, month) but also on the quantiles, I forgot to take this into account in my last comment. We really want to leave open the possibility of linear interpolation on quantiles I think.

If you want to leave things as they were in the most economical way possible (if you need to deactivate a test, go for it), I really think this is beyond the scope of your PR anyways, you fixed many things indepently of this. I can take care of this in another PR. No need for deprecation warning, we will just fix this

@aulemahal, I think that in utils, we don't use the 2D interp function for prop=="group". We should not use it for prop=="dayofyear" either, agreed? interp should still be possible to use with quantiles, it's just implicit that dayofyear grouping will also be 1-dimensional

def interp_on_quantiles(
...
    if prop in ["group", "dayofyear"]:
        if prop =="group":
            if "group" in xq.dims:
                xq = xq.squeeze("group", drop=True)
            if "group" in yq.dims:
                yq = yq.squeeze("group", drop=True)
        out = xr.apply_ufunc(
            _interp_on_quantiles_1D,
            ...

[saschahofmann]

I guess we can merge then 😬

[coxipi]

just commenting so I can commit my suggestion

[coxipi]

Feel free to merge when it becomes possible again, we're just waiting on tests

[saschahofmann]

I dont have write access so cant merge

[coxipi]

I dont have write access so cant merge

Ah, right.

Thanks again for your nice contribution!

[sylvainmarchi]

Hello,
I’m having trouble understanding this thread and the one titled "Correctly interpolate seasons in Grouper #2019." Could someone clarify the role of the interp argument, particularly in the context of MBCn as shown below? Does it apply to the group dimension, the quantiles, or both?
Thank you,
Sylvain

group = sdba.Grouper("time.dayofyear", window=31)

ADJ = sdba.MBCn.train(
	ref,
	hist,
	base_kws={"nquantiles": 50, "group": group},
	adj_kws={"interp": "linear", "extrapolation": "constant"}, 
	n_iter=20,  # perform 20 iteration
	n_escore=1000,  # only send 1000 points to the escore metric
)

[coxipi]

Hi Sylvain,

MBCn with sdba.Grouper("time.dayofyear", window=31) and interp="linear" is not concerned by this, there is only interpolation on quantiles

In this case of sdba.Grouper("time.dayofyear", window=31), a linear interpolation will only be performed on the quantiles. Consider the adjustment on dayofyear-1. I will label the subset with rolling windows as ref1w31, hist1w31. We want to adjust the simulation on dayofyear-1 with this, sim1. You find the values of 50 equally spaced quantiles:

{ref1w31_q01, ref1w31_q03, ... , ref1w31_q99}
{hist1w31_q01, hist1w31_q03, ... , hist1w31_q99}

Compute adjustment factors from those
{af1_q01, af1_q03, ..., af1_q99}

Then the linear interpolation gives you a continuous function of the adjustment factors:
af1(q). You want to apply these adjustment factors on the ranks of sim1.

In the particular case of day of year grouping, no more interpolation is needed, we have adjustment factors af1 for sim1, af2 for sim2, af3 for sim3, etc.

When working with seasons, we have less adjustments factors, instead of af1,af2, ... af365, we only have afDJF, afMAM, afJJA, afSON. But we would still like to have distinct adjustment factors for each dayofyear, so apply different adjustment to sim1, sim2, ... sim365. To achieve this, we interpolate the results between seasons {afDJF, afMAM, afJJA, afSON}. So in this sense, there is also an interpolation on the grouping dimension, not only on quantiles. My description of what is done is not exact, because the two-fold interpolation on quantiles and season is done in one step, but I think it still schematizes well what is happening.

Hope this clears this up!

Éric

[sylvainmarchi]

Hello Éric,
Thank you for your very clear explanation.
Sylvain