Forecast for November 2023

[scotthosking]

@JimCircadian - could you produce for me the sea ice conditions around Rothera in Novermber 2023.

Even 3-4 months from now would be useful

It would also be useful to know whether we expect more or less sea ice on average in this region, compared to the last 5 years (but this can come later if too much more work)

[JimCircadian]

I'll run a 93-day forecast off now (as that's what the model is trained / structured for.) Tom and I have collectively discussed and want to get a monthly run out at some point, but that will take longer.

I'll get that forecast to you by tomorrow, then we can up the priority of training other timescales (or time resolutions, likely monthyl) for a November prediction, if we think that's a good egg to crack!!! It's very easy to configure a 180-day model, for example, so maybe we should give it a whirl...? 😄

[JimCircadian]

Tasklist:

• Get southern hemisphere data updated to latest 2023
• Run 2021,2022,2023 "operational" forecast on dh23 (93-day model in use at BAS) and new JASMIN SH model
• Notebook for comparison against averaged SIC from previous years
• Demo a 180-day forecast model
• Review the priority of monthly forecast test

[JimCircadian]

export ROTHERA_START="2023-03-14,2022-03-14,2021-03-14"
export ROTHERA_END=$ROTHERA_START
./run_prediction.sh rothera dh23 south rothera current 2>&1 | tee logs/rothera.out
./produce_op_assets.sh rothera_south 100,216,180,296 2>&1 | tee logs/rothera.assets.log

Running now at BAS

./run_prediction.sh rothera dh23 south rothera current 2>&1 | tee logs/rothera.out
First model file: results/networks/dh23/dh23.network_current_north.22.current_south.22.16.h5
Dataset model was trained on: current_north.22.current_south.22
Dates from ENVS: ROTHERA_START and ROTHERA_END
Prediction start arg: 2023-03-14,2022-03-14,2021-03-14
Prediction end arg: 2023-03-14,2022-03-14,2021-03-14

JASMIN slurm memory killers got rid of downloaders, so trying again.

[JimCircadian]

Running a 2017 forecast (already have the SEAS data) for analysis about how the predictability might be going into the austral summer, which should be useful. Will run 1, 7, 14 and 30 analysis using the side by side tool as well. @scotthosking

(icenet) [jambyr@bslws03 pipeline]$ export ROTHERA_START="2017-9-1"
(icenet) [jambyr@bslws03 pipeline]$ export ROTHERA_END="$ROTHERA_START"
(icenet) [jambyr@bslws03 pipeline]$ ./run_prediction.sh rothera17w dh23 south rothera current 2>&1 | tee logs/rothera.out
First model file: results/networks/dh23/dh23.network_current_north.22.current_south.22.16.h5
Dataset model was trained on: current_north.22.current_south.22
Dates from ENVS: ROTHERA_START and ROTHERA_END
Prediction start arg: 2017-9-1
Prediction end arg: 2017-9-1

[JimCircadian]

Attempting run in JASMIN. This is likely to fail due to the state of the models and data, which are struggling for various reasons (see #46).

./run_prediction.sh rothera20s jasmin_dev_0123_dual south rothera dev16 2>&1 | tee logs/rothera.out