Hi @c.j.r.williams just to let you know that my run carried on past 1/1/1854 so I can confirm that the error you saw wasn’t due to your model setup; i.e. good news!
All the best.
Jonny
Okay, so do I just try to retrigger the failed coupled task? Or do I need to properly restart?
Charlie
It’s probably easier just to restart from the beginning to be honest since it’s close to the start. Restarting from a later point is certainly possible and not too onerous but it’s up to you 
.
Cheers.
Jonny
Hi Jonny,
Hope all is well.
So I have now resubmitted number as you suggested. In the meantime, while waiting for it, I thought I would do a little bit of basic analysis on what I had from previously i.e. just the 3 years. The results are puzzling me mightily.
In the images below, PI = u-dw696 (i.e. no modification to the T & S) whereas SST Test 1 = u-dw704 (i.e. with an anomaly added to the T & S). Both images show the differences i.e. Test 1 - PI. The anomaly was added during the first year, months 1-12. The anomaly should make the Test warmer than the PI. My expectation was that the anomaly would only have an impact for the first year, maybe several years if I was lucky, but otherwise the internal dynamics of NEMO would take over and any initial anomaly would be washed away (pun intended).
But as you can see, the results are almost the opposite. The results are absolutely identical for the first 11 months of the run, with any difference whatsoever only appearing month 12. Then we see a difference. This is true in the global mean and spatially. I thought I was going mad at first and doing something wrong with my analysis, but I have doublechecked the raw data (using xconv) and every month in the first year until December is absolutely identical between the runs.
Can you explain this? It is also somewhat different to your figure above, although admittedly you were looking at daily data whereas I am looking at monthly data. But you were still seeing a clear difference in the global mean within the first 60 days, whereas I am now seeing no difference at all until the end of the first year.
The only possible reason I can think of - is there some sort of lag in NEMO, whereby the initial T & S conditions are only applied towards the end of the first year of the run i.e. it ignores 1850, and only begins in December of 1850? I know that the climate meaning of the atmosphere component does something similar i.e. if you wanted a December-February seasonal average from the beginning of your run, you wouldn’t get an average of January-February 1850 but you would your first average would be December 1850-February 1851. Is NEMO maybe similar in the way the initial conditions are applied?
Charlie
Hi Charlie.
I’m not able to see your data on JASMIN since I’m not a member of the ‘pliod’ group workspace. I’ve applied for access though since this may well come in useful in the future. Fundamentally I can’t explain what’s going on without access to the raw data itself I’m afraid.
One thing to note is that the runs differ only in their initial conditions, i.e. at 1/1/1850. After that the model integration continues with what it has been given at t=0. Therefore I don’t think there should be any ‘lag’; indeed I didn’t see any.
In my figure that I showed earlier (see below) I only showed days 30-60 since there was (if I remember rightly) a large positive spike at the beginning of the run which sounds like this is exactly what you expected. I don’t have this data on disk any more but these are the Iris cubes that I used in my plot…
cube1 = iris.load_cube(
"dw696o_1d_18500101_18501230_grid_T.nc", "sea_surface_temperature"
)
cube2 = iris.load_cube(
"dw704o_1d_18500101_18501230_grid_T.nc", "sea_surface_temperature"
)
All the best.
Jonny
Thanks very much indeed, I have just granted you access to that GWS (it’s kind of helpful that I happen to be in the GWS manager for it).
Charlie
Hi Charlie
Handy!
I’ve plotted the daily and monthly mean sea_surface_temperature field from the nemo*1m*1850*grid-T*.nc and nemo*1d*18500101-18510101_grid-T.nc files from your space on JASMIN and they look fine to me (see below). I’m happy to share the plotting code if you want.
All the best.
Jonny
Sorry for a quick reply, away from my laptop and about to go into a meeting.
But that’s really weird. You are looking at the raw output, yes? In contrast I was looking at my regridded data, which I regridded onto a standard 360,180 grid using CDO. All I can think is that this has gone wrong somewhere. My regridded files are in the same directory path, but a couple of levels up. Can’t remember exactly, but something like /processed_output/. Could you possibly take a look in there to confirm what I said? Because otherwise I’m going mad.
Charlie
Hi Charlie.
I do indeed get the same results as you using the data in /gws/ssde/j25b/pliod/cwilliams2011/plioda.d/hadgem3.d/processed_output.d…
It looks like you created your ‘processed’ output using results from an ‘old’ run, as can be seen in the `history’ attribute of the processed files…
cdo -L -remapnn,global_lonlat_1deg -selname,tos
[snip...]/18500101T0000Z_old/nemo_dw696o_1m_18500101-18500201_grid-T.nc ...
The results from your current run look fine though.
Cheers.
Jonny
Hi Jonny,
Sorry for the delay, I was teaching all of yesterday.
Okay, I have now resolved this issue thanks to your last comment, and maybe unsurprisingly it was a schoolboy error. Basically (if you’re interested!), I had saved the first year from the previou runs I did (i.e. 18500101T0000Z_old in both dw696 and dw704), just in case it didn’t work again. I renamed this using the suffix _old in case it did work. But then when I did my regridding, it sequentially reads through all directories. In this case, it would have read through 1850 from the new run, but then overwritten the output with the files in the old run. And as we know, the previous runs I did were not correctly using the initial T&S conditions, but rather were starting from the same restart dump. So they should indeed be identical, which is why the results were!
I have now corrected this, and now my output is indeed different from month 1 onwards.
Apologies for the error, and thank you very much for your help.
Charlie
Hey Charlie.
Great news that’s all sorted now!
Jonny