I’ve got a quick question about how the JRA-55 repeat year forcing was constructed and any subtilities that I was hoping you would be able to answer.
@john_reilly and I were about to kick-off some EAC runs looking at a repeat El-Nino(s) year and La-Nina year (similar to the repeat SAM year project from a few years ago). As far as I can tell, this appears pretty straightforward - if I was to run (say) a repeat 1997-1998 El-Nino case, I would simply:
take the JRA-55 do data for those years;
concatenate it together to make a continuous 365 year dataset running from 1st May to 31th April;
roll the time-series so that it runs from 1st january to 31st december (although this step seems unnecessary);
run the model;
However, at our last meeting, John asked about how you handel the transistion from one year to another to ensure a smooth continuation. To that end, I read both the normal RYF paper and the Yeager and Large CORE forcing paper.
Is my understanding correct that the RYF dataset simply transitions from one year to the next at May the 1st, and hence may be discontinuous? Is any detrending performed?
Boreal Spring (MAM) is considered to be the most quiet time for ENSO (ie. SST anomalies reach their minimum and there is a loss of memory from prior conditions. May 1st should be ok as a transition date, but I might need to do some experiments to see if there are any influences using an earlier transition date.
Yep, there’s no smoothing across the transition region. There aren’t too many subtilities about it. Kial’s paper has the details as to why this time of year was chosen.
The code to produce the RYF forcing files for ACCESS-OM2 is here. It should be very easy to modify to use different months/different years.
Aidan wrote the script to generate the forcing files (in the repo Ryan linked). This can make multiple RYF files at the same time. We already have the following RYF (v1.3) files made: 8485, 9091, 9192, 9899, 0203, 0304, 1011, 1516, 1617.
There is no 31st of April; take the timeseries that runs from 0000UTC on 1st May to 2359 on 30th April. If you happen to pick a year that has a Feb 29th, go to the 29th April so as to keep 365 days per year.
Starting the forcing in January lines up with the initial conditions of the model- these are the January WOA hydrography rather than the annual average WOA hydrography (only makes a difference above 1500m in the ocean, and maybe with the initial sea ice field?).
We do not do any smoothing at the transition; so yes, it discontinuous. But shifting this discontinuity away from the peak NADW production time and peak SAM time in early January, to the more climatologically benign time of May means we don’t interrupt these processes or cycles. If you have a reason to shift this transition day (eg. if your dynamic of interest has a cycle that peaks or builds in April-May), you may want to consider picking another day.
Attempting to implement a smoothing transition operator is impossible without disrupting the fidelity and self-consistency of the various forcing fields. So we make no attempt to switch off/on storms, etc, or mesh weather systems into one another.
Thanks @kial and @rmholmes that’s clear now (just making sure that I don’t do anything ridiculous before I push the button).
1 Like
Aidan
(Aidan Heerdegen, ACCESS-NRI Release Team Lead)
5
It’s a fairly venerable script now. I am sure there are probably better ways to do it, or improve it. Feel free to make a PR to improve it, or add another updated script.