PyCPT Error

[Benjamin Kwenda]

Greetings colleagues

I have tried to generate forecasts for other subseasons such as DJF and JFM. When I try to run this, I am getting an error from the Multi-Model Ensemble cell.

Anyone who has also encountered the same problem? How have you solved it?

Regards

--

Benjamin M Kwenda

Agriculture Meteorologist

Long Range Forecasting

Meteorological Services Department

Belvedere, Harare

(+263) 773504696

Skype: bkwenda_1

Scientists of today think deeply instead of clearly. One must be sane to think clearly, but one can think deeply and be quite insane!!!!- Nikola Tesla

[Celio Matuele]

Hi everyone!

Hey Benjamin, I am facing the same problem while running DJF and JFM!

The error is at the Multi-Model Ensemble cell.

Please see it in attached screenshots.

 

Best regards,

Celio.

--
You received this message because you are subscribed to the Google Groups "pycpt-esa" group.
To unsubscribe from this group and stop receiving emails from it, send an email to pycpt-esa+...@iri.columbia.edu.
To view this discussion on the web visit https://groups.google.com/a/iri.columbia.edu/d/msgid/pycpt-esa/CAEEVC%2BiKyyoQZndO8Z%3DGdfjifVS5%3DdcXzFq_TC6p0_jTDZVLtQ%40mail.gmail.com.
For more options, visit https://groups.google.com/a/iri.columbia.edu/d/optout.

[Andrew Robertson]

HI Benjamin & Celio,

Thanks for sharing! Please always attach the Jupyter Notebook for debugging. This will help others (hopefully) quickly isolate the problem. Make sure that the model hindcasts from all the models are available for the training period you’ve chosen - might solve it in this case.

Best,

Andrew

On Oct 25, 2022, at 05:05, 'Celio Matuele' via pycpt-esa <pycp...@iri.columbia.edu> wrote:

Hi everyone!

Hey Benjamin, I am facing the same problem while running DJF and JFM!

The error is at the Multi-Model Ensemble cell.

Please see it in attached screenshots.

<image.png> 
<image.png>

Best regards,

Celio.

To view this discussion on the web visit https://groups.google.com/a/iri.columbia.edu/d/msgid/pycpt-esa/CANyb4NwUR9H6gQtKLAMDSChvcacZ%3DQhMb1zYkM5qjX%3DbKskQZw%40mail.gmail.com.

[Benjamin Kwenda]

Dear Andy

Thank you for your email.

I have attached my notebook so you can check it out.

Regards

[Benjamin Kwenda]

This one as well is giving me a similar error for SST predictands

[Andrew Robertson]

Hi Kyle,

I looked into this and I believe there is indeed a bug: for a JFM target season, ‘first_year’ and ‘final_year’ for the obs download need to be incremented by 1 year because the hindcasts are initialized at the end of the previous year. 

Is this best handled in the Jupyter Notebook or in CPT-DL?

There’s probably a related problem with DJF and NDJ target seasons, although the use of seasonalAverage might just mean the training time series in shortened by 1 year.  Ben did find a problem w/DJF too.

Best,

Andy

<pycpt-operational- Benjy.ipynb>

[Andrew Robertson]

Thanks Kyle! We’ll try workarounds from within the notebook.

On Oct 26, 2022, at 13:40, Kyle Hall <hall...@gmail.com> wrote:

I have found this error during the course of my own work; it is a quirk of the Ingrid “seasonalAverage” function. It would be better to address in CPT-DL but you can write workarounds from within the notebook. 

Best

Kyle 

[Benjamin Kwenda]

Dear Andy

I had suspected that this could be the source of the error but I wasn't sure because I wasn't getting a similar error for NDJ. And I also discovered that the individual forecasts from the various predictands were not giving me such an error but got it when it came to the MME.

Regards

[Andrew Robertson]

Hi Ben,

Status update: I was able to solve the dates problem, but there is still an issue with missing data being generated in the deterministic hindcasts (though input datasets look fine). I’m still trying to track that down and will keep you posted.

Best,

Andy

[Benjamin Kwenda]

Great, thanks for the update! Will be waiting patiently.

Regards