Split OHC, temperature and salinity anomalies into separate tasks #294

xylar · 2018-01-03T14:00:08Z

Three new subtasks have been added to perform the 3 anomaly analyses, one for computing a moving average of an anomaly and two for plotting Hovmoller plots and depth-integrated time series
of the resulting anomaly. The anomaly data is now stored in the timeSeries output folder so it can be shared across plotting subtasks and for provenance.

New utility functions have been added to compute moving averages and anomalies.

In several tasks, regionIndicesToPlot has been replaced by regions, a list of the region names instead of the region indices. This should be more intuitive for a user to change.

Note: Plots should be qualitatively the same as they were before. However, I have seen visible differences in the OHC Hovmoller plot. I believe this is due to a slightly incorrect computation of the anomaly previously and that the current plot is correct. The previous computation assumed that only the layer temperature changed with time, whereas the current computation accounts for the fact that other fields (namely layer thickness) that contribute to the OHC can potentially change with time.

xylar · 2018-01-03T14:01:23Z

Testing

So far, I have tested this on my laptop using the QU240 test case. All plots except the OHC Hovmoller (see note above) look the same as before. I have made some very small formatting changes along the way.

xylar · 2018-01-03T14:03:21Z

@milenaveneziani, I would appreciate it if you could run a test or two on this new code to make sure nothing is broken. I would also appreciate you comparing the resulting 4 anomaly plots with the same from a previous run of the analysis to make sure you don't have any concerns about the resulting changes.

I'm happy to address any concerns you might have.

xylar · 2018-01-03T14:05:08Z

Note

I removed the ability to plot "original fields" (as opposed to anomalies) for now. These would be relatively easy to add back in if we thought they would be useful but since they ware off by default and we don't seem to be including them in our workflow, I decided to simplify my work and the default config file but leaving them out.

I will happily add them back in if someone thinks they are important.

milenaveneziani · 2018-01-04T18:07:45Z

mpas_analysis/config.default

-# plot S, T, and OHC fields themselves, in addition to their anomalies?
-plotOriginalFields = False
-# compare to observations?
-compareWithObservations = False


I see this has been removed: we were probably no longer using compareWithObservations anyway. At some point, though, we will want to process the OHC observations for transient/historical runs. This will only make sense for those runs, so we will need some sort of flag to turn on/off the comparison with obs.

milenaveneziani · 2018-01-04T18:39:20Z

@xylar, I was testing this on edison, using edison_acme_unified_2017.9.26, and got the following error:

Plotting depth-integrated time series of OHC Anomaly...
  Load ocean data...
  Read in depth...
  Load in OHC from preprocessed reference run...
analysis task timeSeriesOHCAnomaly_plotDepthIntegratedTimeSeriesGlobal failed during run 
Traceback (most recent call last):
  File "/global/u2/m/milena/MPAS-git-repositories/MPAS-Analysis/mpas_analysis/shared/analysis_task.py", line 325, in run
    self.run_task()
  File "/global/u2/m/milena/MPAS-git-repositories/MPAS-Analysis/mpas_analysis/ocean/plot_depth_integrated_time_series_subtask.py", line 364, in run_task
    field = compute_moving_avg(field, movingAveragePoints)
  File "/global/u2/m/milena/MPAS-git-repositories/MPAS-Analysis/mpas_analysis/shared/time_series/moving_average.py", line 32, in compute_moving_avg
    center=True).mean().dropna('Time')
  File "/global/project/projectdirs/acme/software/anaconda_envs/edison_acme_unified_2017.9.26/lib/python2.7/site-packages/xarray/core/rolling.py", line 242, in wrapped_func
    'Rolling window operation does not work with dask arrays')
NotImplementedError: Rolling window operation does not work with dask arrays

milenaveneziani · 2018-01-04T18:40:19Z

Both temp and salt anomaly trends went fine though. I guess that's because temperature and salinity are not xarray datasets?

xylar · 2018-01-04T19:39:04Z

Thanks, @milenavezeiani. I'll look into the error. I presumably need to disable dask somehow (it should be necessary for these fields anyway).

milenaveneziani · 2018-01-04T19:45:24Z

why did it work on your laptop though? different dask/xarray version?

xylar · 2018-01-04T21:57:47Z

I think it probably has to do with the size of the dataset but I also bet there's a trick for forcing xarray not to use dask. Otherwise rolling is a pretty useless feature...

xylar · 2018-01-05T09:16:37Z

@milenaveneziani, the answer seems to be that I'm using xarray 0.10.0 on my laptop, whereas we're still using 0.9.6 on supported machines. I'm going to see about deploying that newer version of xarray in e3sm-unified. I'll keep you posted as soon as I have something on Edison.

The issue was addressed in this xarray PR:
pydata/xarray#1568

xylar · 2018-01-05T13:35:40Z

@gstreletz, I finally got around to splitting the work you did into several tasks and subtasks. I would appreciate it if you could take a look at what I've done and let me know if you have any concerns.

For the moment, I don't have tasks to plot the "original" fields (non-anomalies) as you will see above, but it wouldn't be hard at all to add back in support. I just haven't got to it yet (in part because we don't make those plots by default).

milenaveneziani · 2018-01-05T15:30:45Z

Re: original fields: I think it is OK to leave them for the future, because on the global scale and for the coupled system, we usually look at the trends of the anomalies.
Re: xarray issue: I suspected it was version-related. Let me know when you update e3sm-unified and I will test again.

What do you think about my comment on the comparisonWithObs flag? Should we ditch it for now and then put it back when it's necessary? That is unless there are already lines of the kind if comparisonWithObs, which we used to have, but I'm not sure whether they are still there.

xylar · 2018-01-05T16:19:12Z

@milenaveneziani

Re: xarray issue: I suspected it was version-related. Let me know when you update e3sm-unified and I will test again.

Yes, I think that's right. If you try e3sm-unified/1.1.2 on edison, that should now work. I tried it on a long beta2 run but it hadn't finished in batch mode (probably just because 153 years of MOC is too much) but the OHC plots look right:
http://portal.nersc.gov/project/acme/xylar/split_ohc/20170915.beta2.A_WCYCL1850S.ne30_oECv3_ICG.edison_batch/ocean/

What do you think about my comment on the comparisonWithObs flag? Should we ditch it for now and then put it back when it's necessary?

I don't think it's good to have flags in the config file for features that aren't yet supported. There is no reference to comparisonWithObs in the old code as far as I could tell so I think we re-introduce this option when we have observations to compare with. That won't be hard.

milenaveneziani · 2018-01-05T16:22:55Z

If you try e3sm-unified/1.1.2 on edison, that should now work

ok, I will try my old test with the new conda and see how things go.

There is no reference to comparisonWithObs in the old code

ok, that was my only concern. Agreed that we shouldn't have unused flags in the config file.

milenaveneziani · 2018-01-05T16:44:33Z

The new test passed.

I do not see any differences between the results that come out of this branch and the ones from develop, at least visually. Edit: the only difference I see seems to be due to a different aspect ratio of the new plots. You can check for yourself:
Results from develop: http://portal.nersc.gov/project/acme/milena/B_low_res_ice_shelves_1696_JWolfe_layout_EdisonTest1/ocean/index.html
Results from this branch: http://portal.nersc.gov/project/acme/milena/B_low_res_ice_shelves_1696_JWolfe_layout_EdisonTest2/ocean/index.html

xylar · 2018-01-06T17:44:59Z

I turned of the MOC and my test completed as expected:
http://portal.nersc.gov/project/acme/xylar/split_ohc/20170915.beta2.A_WCYCL1850S.ne30_oECv3_ICG.edison/

xylar · 2018-01-06T18:55:26Z

Waiting to merge this until I can rebase onto develop after #296 is merged.

xylar · 2018-01-06T18:58:11Z

@milenaveneziani, thanks very much for the testing.

milenaveneziani · 2018-01-10T18:24:49Z

configs/olcf/job_script.olcf.bash

 module unload python
-module use /ccs/proj/cli115/pwolfram/modulefiles/all
-module load python/anaconda-2.7-climate
+module use /ccs/proj/cli127/software/modulefiles/all


@xylar: what is cli127? I don't have access to it.

Hmm, it's the only group I have access to. Maybe our alcc project?

I don't have access to cli115. Is that the E3SM group on olcf?

yep, it's the main e3sm project at OLCF. You should have access to it. Could you ask for it? it should be fairly quick: usual form and Mark Taylor is the PI.

Okay, so I'll move this to cli115 as soon as I have permission.

milenaveneziani · 2018-01-11T17:28:59Z

as expected, tests passed on edison when the new e3sm-unified conda is used.
I will test on titan as soon as you can move e3sm-unified to cli115, @xylar.

xylar · 2018-01-12T15:31:22Z

I now have rhea access so I'll set up e3sm-unified soon.

xylar · 2018-01-12T16:34:52Z

/ccs/proj/cli115 appears to be full so I'm not able to install e3sm-unified there. I believe an acme_diags test is at least partly to blame. I have posted a confluence note for @chengzhuzhang.

milenaveneziani · 2018-01-12T18:19:20Z

/ccs/proj/cli115 appears to be full

really.. I also have a lot of stuff in there that I can't really delete (v0 processing).. I didn't get any notice that the space was full or had any problems recently.
Anyways, let me know if you get to install e3sm-unified and I'll test.

Edit: that's the shared space that is full. OK, I have nothing in there.

If an MPAS dataset is saved back to disk, it will have a Time variable rather than xtime (or its variants). Support has been added for reading back such files.

These new functions compute the moving average of a data set over a window of samples in time, and the anomaly of that moving average from the beginning of an MPAS simulation.

The user would now supply a list of regions to plot and wouldn't need to know which indices they correspond to. This is more intuitive for a user to change in the future.

Sometimes, a moving average is not needed and other times the moving average has already been performed on the data before plotting.

The temperature and salinity tasks are now separate from the OHC task. Also, three new subtasks have been added to perform the 3 anomaly analyses, one for computing a moving average of an anomaly and two for plotting Hovmoller plots and depth-integrated time series of the resulting anomaly.

This is needed because we now require xarray>=0.10.0

xylar · 2018-01-25T00:31:21Z

@milenaveneziani, I now have e3sm-unified/1.1.2 everywhere, which includes both the version of xarray needed by this PR and the version of nco needed by #300. Could you retest on titan or rhea using the instructions on the Supported Machines confluence page?

milenaveneziani

I have tested on titan after loading the new e3sm-unified conda and all worked fine.

xylar · 2018-01-25T05:13:17Z

@milenaveneziani, great! Thanks so much for testing so late.

Merges PRs #282, #285, #287, #289, #290, #291, #293, #295, #288, #296, #237, #297, #294, #300 and #292

xylar added the clean up label Jan 3, 2018

xylar self-assigned this Jan 3, 2018

xylar requested a review from milenaveneziani January 3, 2018 14:00

xylar added the priority label Jan 3, 2018

milenaveneziani reviewed Jan 4, 2018

View reviewed changes

xylar added the in progress label Jan 5, 2018

xylar force-pushed the split_timeseries_anomalies branch from ffdf131 to 63e7afc Compare January 5, 2018 13:03

xylar requested a review from gstreletz January 5, 2018 13:32

xylar added bug and removed in progress bug labels Jan 6, 2018

xylar mentioned this pull request Jan 6, 2018

Check for missing variables in MPAS time series #296

Merged

xylar added the in progress label Jan 6, 2018

xylar mentioned this pull request Jan 9, 2018

Support comparison with a reference run #292

Merged

xylar force-pushed the split_timeseries_anomalies branch from 66a8dc0 to 38e1c1b Compare January 9, 2018 20:55

milenaveneziani reviewed Jan 10, 2018

View reviewed changes

xylar mentioned this pull request Jan 14, 2018

Creating v0.7 tag #299

Closed

15 tasks

xylar force-pushed the split_timeseries_anomalies branch from 554e881 to 8f9f644 Compare January 17, 2018 04:26

milenaveneziani mentioned this pull request Jan 23, 2018

Read meridional heat transport lat bins from either of 2 streams #297

Merged

xylar force-pushed the split_timeseries_anomalies branch 2 times, most recently from 08a7d62 to 9304c77 Compare January 24, 2018 19:44

xylar added 9 commits January 24, 2018 17:28

Add support for reading Time in open_mpas_dataset

1b07534

If an MPAS dataset is saved back to disk, it will have a Time variable rather than xtime (or its variants). Support has been added for reading back such files.

Add subset_variables to io module (for convenience)

b7caceb

Add util funs for anomalies and moving avgs

93879d3

These new functions compute the moving average of a data set over a window of samples in time, and the anomaly of that moving average from the beginning of an MPAS simulation.

Switch from regionIndicesToPlot to just region(s)

e1698a4

The user would now supply a list of regions to plot and wouldn't need to know which indices they correspond to. This is more intuitive for a user to change in the future.

Support time series plots without moving averages

bd0a120

Sometimes, a moving average is not needed and other times the moving average has already been performed on the data before plotting.

Update xarray version in README

d507361

Update documentation

75eb12f

Update paths to e3sm-unified/1.1.2

9c42d4b

This is needed because we now require xarray>=0.10.0

xylar force-pushed the split_timeseries_anomalies branch from 9304c77 to 9c42d4b Compare January 25, 2018 00:28

xylar removed the in progress label Jan 25, 2018

milenaveneziani approved these changes Jan 25, 2018

View reviewed changes

xylar merged commit 2b27771 into MPAS-Dev:develop Jan 25, 2018

xylar deleted the split_timeseries_anomalies branch January 25, 2018 05:13

xylar mentioned this pull request Jan 26, 2018

Merge develop to master #302

Merged

xylar added a commit that referenced this pull request Jan 26, 2018

Merge develop to master

e56da52

Merges PRs #282, #285, #287, #289, #290, #291, #293, #295, #288, #296, #237, #297, #294, #300 and #292

Split OHC, temperature and salinity anomalies into separate tasks #294

Split OHC, temperature and salinity anomalies into separate tasks #294

Uh oh!

Conversation

xylar commented Jan 3, 2018

Uh oh!

xylar commented Jan 3, 2018

Testing

Uh oh!

xylar commented Jan 3, 2018 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

xylar commented Jan 3, 2018

Note

Uh oh!

milenaveneziani Jan 4, 2018 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

milenaveneziani commented Jan 4, 2018

Uh oh!

milenaveneziani commented Jan 4, 2018 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

xylar commented Jan 4, 2018

Uh oh!

milenaveneziani commented Jan 4, 2018

Uh oh!

xylar commented Jan 4, 2018

Uh oh!

xylar commented Jan 5, 2018 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

xylar commented Jan 5, 2018

Uh oh!

milenaveneziani commented Jan 5, 2018

Uh oh!

xylar commented Jan 5, 2018

Uh oh!

milenaveneziani commented Jan 5, 2018

Uh oh!

milenaveneziani commented Jan 5, 2018 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

xylar commented Jan 6, 2018

Uh oh!

xylar commented Jan 6, 2018

Uh oh!

xylar commented Jan 6, 2018

Uh oh!

milenaveneziani Jan 10, 2018

Choose a reason for hiding this comment

Uh oh!

xylar Jan 10, 2018

Choose a reason for hiding this comment

Uh oh!

xylar Jan 10, 2018

Choose a reason for hiding this comment

Uh oh!

milenaveneziani Jan 10, 2018 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

xylar Jan 10, 2018

Choose a reason for hiding this comment

Uh oh!

milenaveneziani commented Jan 11, 2018

Uh oh!

xylar commented Jan 12, 2018

Uh oh!

xylar commented Jan 12, 2018

Uh oh!

milenaveneziani commented Jan 12, 2018 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

xylar commented Jan 25, 2018

Uh oh!

milenaveneziani left a comment

Choose a reason for hiding this comment

Uh oh!

xylar commented Jan 3, 2018 •

edited

Loading

milenaveneziani Jan 4, 2018 •

edited

Loading

milenaveneziani commented Jan 4, 2018 •

edited

Loading

xylar commented Jan 5, 2018 •

edited

Loading

milenaveneziani commented Jan 5, 2018 •

edited

Loading

milenaveneziani Jan 10, 2018 •

edited

Loading

milenaveneziani commented Jan 12, 2018 •

edited

Loading