Overlapping SEAS5 and IceNet forecast init dates

[EllieBowler]

Hi IceNetters! Thought i'd try using the new discussions page... but sorry if isn't the best place! I just had a question about the compute_metrics_leadtime_avg function. I am using it to compare average IceNet and SEAS5 forecasts over a range of initialisation dates. I have a forecast file with 36 init dates and all corresponding SEAS5 dates in the data store. But i noticed the code misses the first one out (strictly > start date but <= end date) here, so i only get 35 comparisons. I was wondering if there is a reason for this, or if the code should be changed to => for the start as well? Ta!

[bnubald]

Hi @EllieBowler

Thanks for posting! I think this is the perfect place for it.

I've just checked some output netCDF files, and based on that, it should include the first date as you've suggested (i.e., for a prediction initialised on 2024-06-30, the first forecast day is 2024-07-01, and so, dates >= fc_start_date instead of dates > fc_start_date). It might be a bug, but would just like to confirm the versions used for (1) training, (2) prediction and (3) outputting this plot before moving forwards.

Related topic I find of use, and should definitely be documented by me...: #256 (comment)

@EllieBowler, also, for posterity, could I ask for the following please:

• Version of IceNet.
• Command you are running.
• Location of run.
• Range of dates of SEAS data downloaded.
• Paste the output (and for ease, length) of the forecast_date variable in the forecast_file netCDF file being passed to the compute_metrics_leadtime_avg function.

Note: This refers to following code:

icenet/icenet/plotting/forecast.py

Line 560 in f128536

dates = dates[(dates > fc_start_date) & (dates <= fc_end_date)]

[EllieBowler]

Thanks Bryn! Yes sure:

IceNet version: 0.2.8
Command: icenet_plot_leadtime_avg north "paper-predict/ellie_paper_north.nc" -o "plot/test.png" -m mae -dp "metrics/test.csv" -ao all -s -sm 1 -e -b -r 114,242,131,260
Location: BAS HPC
SEAS downloaded: All currently listed in this folder: /data/hpcdata/users/jambyr/icenet/other.data/data/mars.seas/north/siconca
Info on forecast_date from forecast_file:

[bnubald]

Thanks @EllieBowler

Nice find, I've just gone through this in a bit more detail, and can confirm your suspicion on it being a bug and your suggested solution.

i.e.:

dates >= fc_start_date instead of dates > fc_start_date

[bnubald]

@EllieBowler I've created corresponding issue #289