Open aekiss opened 5 years ago
The new 0.1 deg grid has removed the smallest cells near the tripoles (https://github.com/COSIMA/access-om2/issues/126) and looks like this (note the larger minimum size)
Here's what the 1 deg topography looks like on a shallow shelf (colourbar from 0-100m). Due to inheriting the 10m quantisation from GFDL50 (https://github.com/COSIMA/access-om2/issues/141) the minimum depth is 50m and there's 10m terracing, so we're missing the finesse that KDS50 could give
Here's the 0.25 deg topography - now the minimum depth is ~40m, and there's ~10m terraces to deeper water
...whereas this is the glorious 0.1 deg with none of those issues
It looks like the 1 degree and 0.25 degree topographies are still infected by the old dodgy OCCAM bathymetry, the 0.25 bathymetry in particular. The Laptev Sea is particularly bad and closer to home the Gulf of Carpentaria should not be 130m deep.
The gift that keeps on giving.
The East Siberian and Chukchi seas south of Wrangel Island are also bad at 1 and 0.25 deg. Is this 1deg topog.nc the same as what's used in ACCESS-CM2?
Yep, I brought this up several years ago and it never got fixed for some reason.
@russfiedler and I have generated new 1deg and 0.25deg topography from GEBCO 2014:
1 deg: /g/data/ik11/inputs/access-om2/input_rc/mom_1deg/topog.nc
0.25 deg: /g/data/ik11/inputs/access-om2/input_rc/mom_025deg/topog.nc
which might be considered as replacements for the existing ones (perhaps after a bit more tweaking).
They were created using /scratch/v45/aek156/bathymetry/tools/*_deg_test/make_topog.sh
.
They use a KDS50 vertical grid and set a minimum full depth of 11.8m (4 levels), with a minimum cell thickness of 1m, and eliminate terracing (fixing https://github.com/COSIMA/access-om2/issues/141). And starting from GEBCO eliminates the erroneous pits on shelves (https://github.com/COSIMA/access-om2/issues/158#issuecomment-629994561 and https://github.com/mom-ocean/MOM5/issues/172).
The 1deg uses the same land mask as before, but the 0.25deg slightly modifies the previous land mask in ocean_mask.nc
, adding 3 land points and 2 ocean points to deal with non-advective cells/edges (https://github.com/COSIMA/access-om2/issues/210). So far this is the only hand-editing that has been done (everything else was automatic, via make_topog.sh
).
The 1 deg model runs fine but I haven't successfully tested the 0.25deg as I'm still trying to generate remapping weights.
I've made some comparisons to the previous topography halfway down this notebook. On the whole it doesn't look like there are major differences. These are the differences that jumped out at me:
Of these, I think only Lombok, Timor and Denmark Straits could do with hand-editing at 1deg, I guess to make them match the old topo. Does that sound sensible?
Are there any other straits, pathways etc we should check?
FYI I have a fixed-up version of Alistair's GUI topography editing tool on this fork & branch: https://github.com/aekiss/MOM6-examples/blob/editTopo-update/ice_ocean_SIS2/OM4_025/preprocessing/editTopo.py that makes editing fairly straightforward.
Probably have a look at the outflow pathway from the Med. Maybe ask those with an interest in the Southern Ocean to have a look.
Ah yes, I'd meant to add Gibraltar. It's there now. It's a bit shallower at the west of the Gibraltar sill and a bit deeper to the east in both the new 1deg and 0.25deg.
In response to @russfiedler on the southern Ocean Simon added some extra Glacier Tongues to try and create more polynyas, but it does lead to ice build up behind them wich happens as fast ice in the real world at 1 deg back in 2007/8 . I have often wanted them relooked at though they are the source of a limited high salinity water they do have a latent heat signature to the atmosphere (which is realistic). There is also the issue of trapped embayments for ice where it cant move out velocity wise at 1 deg, I haven't checked the 0.25 deg for this but it might be a good time to do it whilst the grids are under discussion.
Shallowing at Bering St can lead to strong a flow depending on width so you might need to look at bottom drag.
Thanks @ofa001 we are using the previous Antarctic land mask at both resolutions, so if there were glacial tongues in the old topography they'll still be there in the new one.
There are no B-grid non-advective cells, but are you talking about ice trapping by some other process?
Sometimes it still gets a bit trapped and thick even if there is an adjective point, it depends on the local winds, In very long runs you might get it partially breaking out after 30-40 years if a wind change occurs. Having the glacial tongues and polynyas isn't too much of an issue in there are often icebergs also causing temporary blocks in the flow, and then polynyas behind as well, in observations. Be great to eventually try a fast ice parameterization at 0.1 degree resolution.
I've put an updated candidate 1 degree topography in /g/data/ik11/inputs/access-om2/input_rc/mom_1deg/topog.nc
.
This was created with https://github.com/COSIMA/make_1deg_topo/blob/b13ad6b/make_topog.sh
The bathymetry is mostly new, created from scratch from GEBCO 2014, but has 249 edits at important straits (see https://github.com/COSIMA/make_1deg_topo/blob/b13ad6b/topog_edits.txt) where the automatically generated GEBCO value was replaced by the value from the old topography /g/data/ik11/inputs/access-om2/input_20200530/mom_1deg/topog.nc
. These edited cells are marked by red circles in the plots below. They mostly deepened the GEBCO topography, but in the Gibraltar Strait it was a shallowing. Many of these locations appeared to have been hand-edited in the old topography, presumably for a reason, so I thought they should be retained.
How does that look? Is there anything I've missed?
If you want to compare the old and new topography you can get editTopo.py
from https://github.com/COSIMA/topogtools and do
module use /g/data/hh5/public/modules
module load conda/analysis3
./editTopo.py --ref /g/data/ik11/inputs/access-om2/input_20200530/mom_1deg/topog.nc --apply /g/data/ik11/inputs/access-om2/input_rc/mom_1deg/topog_edits.txt /g/data/ik11/inputs/access-om2/input_rc/mom_1deg/topog.nc
An update:
The 1 degree topography is awaiting discussion re. which of Simon's hand-edits in the old topography we should include in the new one. These edits are marked in red:
However I think the 0.25deg topography in /g/data/ik11/inputs/access-om2/input_rc/mom_025deg/topog.nc
is ready to use. Its history attribute contains a url showing the exact steps used to create it: https://github.com/COSIMA/make_025deg_topo/tree/37274b2
In addition to what is described here, this new version now fills in the Sulu Sea to the sill depth (~503m), copying what was done in the old topography (this is the only obvious hand-editing I could see in the old topography). This is a 4000-5000m deep enclosed basin with a ~500m sill all around so presumably this was done to avoid issues with the deep water being stagnant.
Here are some plots and stats calculated by https://github.com/COSIMA/topogtools/blob/8b2de2c/bathymetry.ipynb
The mean depth increase in the new topography is 30.6m or 1.9% and the median increases is 11.7m (0.5%). 90% of absolute depth increases are between -125.3m and 224.4m 90% of relative depth increases are between -24.5% and 12.9% Here are the histograms (log scale):
...and bivariate histograms (log scale; the gaps are due to the 20% minimum partial cell height). Note that the new topography extends to shallower depth (11.81m rather than 40.35m).
Here is the spatial distribution of changes (click to enlarge; see here for closeups on particular regions)
The new bathymetry fixes the gaps in the depth distribution (https://github.com/COSIMA/access-om2/issues/141), making better use of the vertical coordinate and partial cells, particularly in shallow water: closeup:
In the absence of any real provenance as to how the old topography was generated I'm prepared to believe that these differences are improvements in the new topography, since it is generated directly from GEBCO 2014 so it should be more realistic. Whether that will translate into improved simulations is a question to be resolved by test runs, but even if it doesn't, that would seem to indicate error cancellation when using the old topography.
Beautiful work @aekiss !
I think the new 1 degree topography in /g/data/ik11/inputs/access-om2/input_rc/mom_1deg/
is ready to use.
Its history attribute contains a url showing the exact steps used to create it: https://github.com/COSIMA/make_1deg_topo/tree/db7b546
It is based on GEBCO_2014 v20150318 but with these edits https://github.com/COSIMA/make_1deg_topo/blob/db7b546/topog_edits.txt marked in red, which are mostly where the old topography was copied to the new: This latest iteration has additional edits in the Red Sea, Indonesian Throughflow, Caribbean, Gibraltar Strait and Northwest Passage suggested by Simon Marsland.
It uses a KDS50 vertical grid and a minimum full depth of 11.8m (4 levels), with a minimum cell thickness of 1m, and eliminates terracing (fixing #141). And starting from GEBCO eliminates the erroneous pits on shelves (#158 (comment) and mom-ocean/MOM5#172).
It uses the same land mask as before.
Here are some plots and stats calculated by https://github.com/COSIMA/topogtools/blob/b2f5ff3/bathymetry.ipynb
The mean depth increase in the new topography is 79.9m or 10.1% and the median increases is 31.2m (1.2%). 90% of absolute depth increases are between -130.8m and 443.6m 90% of relative depth increases are between -16.6% and 35.8% Here are the histograms (log scale):
...and bivariate histograms (log scale; the gaps are due to the 20% minimum partial cell height). Note that the new topography extends to shallower depth (11.81m rather than 45.1m).
Here is the spatial distribution of changes (click to enlarge; see here for closeups on particular regions)
The new bathymetry fixes the gaps in the depth distribution (https://github.com/COSIMA/access-om2/issues/141), making better use of the vertical coordinate and partial cells, particularly in shallow water:
closeup:
As for the 0.25deg topography, I'm prepared to believe that these differences are improvements in the new topography, since it is generated directly from GEBCO 2014 so it should be more realistic. Whether that will translate into improved simulations is a question to be resolved by test runs. It is possible that some more tweaking around critical straits will be required.
Thanks for all the work @aekiss, looks great!
Agreed, fantastic work, including the meticulous documentation
Impressive work!
Thanks - credit is shared with @russfiedler, Simon Marsland, @ofa001 and @adcroft for their advice and software.
We'll need to see if these new topog files have any unwanted effects on the circulation or water masses etc at 1 deg or 0.25 deg.
Most grid cell depths are set as the average of GEBCO data over that cell, which will underestimate the depth of poorly-resolved sills. At 1 deg we reused some of the key sill depths from the previous topography that were carefully chosen to allow the appropriate water masses to exchange. We didn't make these sill adjustments in the 0.25deg so that's worth keeping an eye on. This is an incomplete list of examples:
At 1 deg we also applied a 160m sill depth to the Red Sea outlet but retained the ~600m deep basin behind it, so hopefully that won't fill up with hypersaline water in a long run (in the old topography the Red Sea was set to 50m to avoid this, increasing to 80m at the mouth).
We also didn't do any hand-edits at the outlet of the Baltic, which is now 11.8m deep instead of 50m.
The GEBCO average also underestimates the height of sharp features like Macquarie Ridge, which is deeper in the new 1 deg topography.
Also Simon had deepened the previous 1 deg topography on parts of the Antarctic shelf in order to reduce a bias of excessive ice thickness. We haven't done that in the new topography because we don't seem to be having this problem. (We were getting excessive Antarctic ice until we changed to a turning angle of zero (rather than 16.26°) which I think is more justifiable since we resolve the Ekman layer better with a surface vertical resolution of 2.3m instead of 10m.)
It will also be interesting to see how well the EAC behaves at 1 deg without these shallow points sticking out from the coast:
Positive means new bathymetry is deeper than old?
yes
Closing - apart from the differing land masks between resolution (which we decided not to fix), all these issues have been addressed in /g/data/ik11/inputs/access-om2/input_20201022/
Summary from /g/data/ik11/inputs/access-om2/input_20201022/README.txt
:
@rmholmes @AndyHoggANU I've had a closer look at some of the key North Atlantic straits and it seems they were hand-edited in the old (GFDL) 0.25deg topography. This might explain the different AMOC you are getting with the new topography in the OMIP 0.25deg runs.
I could edit the 0.25 topography to copy the old GFDL depths in those key straits - let me know.
The Denmark Strait sill was apparently hand-deepened in the old (GFDL) topography to 619.9m and widened slightly at that depth. In the new topography the sill depth is 540m and more constricted. These 3 figures show (respectively) the new topog, old topog and difference. They were obtained by /g/data/ik11/inputs/access-om2/bin/editTopo.py --ref /g/data/ik11/inputs/access-om2/input_20200530/mom_025deg/topog.nc /g/data/ik11/inputs/access-om2/input_20201022/mom_025deg/topog.nc
The Faroe Bank channel was also deepened (sill depth increased from 540m to 837.7m) and widened at that depth in the old topography.
Nares Strait was also deepened to at least 397.5m thoughout the channel, whereas it is much more constricted in the new topography and has a sill depth of about 160m.
There don't appear to be any hand edits in Fram Strait.
There were also some edits for the Florida Current that I could copy while we're at it
Thanks @aekiss.
I'm not sure I have the expertise/experience here to comment. It would be nice to fix the AMOC issues (if this indeed would do it), but would require starting the OMIP simulation again.
I am almost finished with cycle 2. I will do a quick analysis on Monday of the AMOC in that cycle (and the other cycles of the 1-degree) to see if there are any hints of recovery...
The edits around Florida are interesting. I have not looked at any Gulf Stream metrics but they may show differences as well.
The AMOC does not appear to be recovering by the end of cycle 2 in the 1/4-degree (purple line below). It also doesn't look like it recovers even up to cycle 6 in the 1-degree (green line below), suggesting that the same may be true for the 1/4-degree:
Note the difference to the old topography (brown) which does not decay as much (although is still weak compared to obs).
This seems to me to be a relatively serious problem. @aekiss @AndyHoggANU should we put in the hand-edits and start again?
Thanks @rmholmes, for comparison here's what we had in Fig 8a of the GMD paper for the 5th cycle with the old topog at 1 deg and 0.25. So the 0.25 deg is definitely doing worse than with the old topog. I'm willing to make those edits if you want to try that. If this topog is to be used for a new 0.25 coupled climate model that's another reason to improve it.
At 1deg the new topog doesn't seem any worse than the old one (at least for AMOC), and presumably the hand-edits at Denmark Str, the Faroe Channel and perhaps Florida Current have helped. I didn't edit Nares Str in the new 1deg topograhy (see above) but that doesn't seem to have done any harm to the AMOC, though maybe it will have impacted the sea ice distribution. It would be good to check that with a comparison like that in fig 28 of the GMD paper (script here).
re. Nares Str - the old 1deg topog was edited to set the depth to 80m throughout most of the strait. It's actually deeper almost everywhere in the new 1deg topog compared to the old one, apart from the sill which is slightly shallower (65m instead of 80m).
Here are the new, old and new-old 1 deg topog (red is deeper in new), obtained by
/g/data/ik11/inputs/access-om2/bin/editTopo.py --ref /g/data/ik11/inputs/access-om2/input_20200530/mom_1deg/topog.nc /g/data/ik11/inputs/access-om2/input_20201022/mom_1deg/topog.nc
Thanks @aekiss. It looks to me like there is a clear motivation to try the edits in the 025deg. If you make those edits I can run a single cycle and see where it gets to?
At 1deg it's a bit less clear. I haven't got the ice data in the database (because Hakase's data also suffers from the daily ice output files problem). I'll sync some of those across to my directory and have a look...
@rmholmes, @AndyHoggANU I've updated the 0.25deg topography in /g/data/ik11/inputs/access-om2/input_20201102
- I've deepened Denmark Strait, Faroe Bank channel, Nares Strait and Caribbean Sea channels including the Florida Strait by copying data from previous topography /g/data/ik11/inputs/access-om2/input_20200530/mom_025deg/topog.nc
. The commit documenting exactly what was done is https://github.com/COSIMA/make_025deg_topo/tree/f262fbf (see topog_edits.txt
).
Here's an overview of the new edits (in addition to Sulu Sea - see above):
These closeups show the new topography, with the edit locations marked.
Denmark Strait:
Faroe Bank channel:
Nares Strait:
Caribbean Sea:
Thanks @aekiss. Those edits look good. I have started another cycle with this topography (seems to be running fine after 1.5 years) and will report back.
The first cycle with the updated topography is complete and the AMOC is definitely looking better (brown line below, where the new topography without these updates is the green line). In particular, the AMOC is not dropping off dramatically post-2000 as it does in the previous version. It actually remains marginally above the old GFDL topography (purple line).
I think this is promising enough that I will continue with these updates. I'm only 1.5 cycles behind, so it won't take too long to catch up.
Thanks @rmholmes, that's good news. Fingers crossed it will keep looking ok for the remaining cycles...
I've set that latest /g/data/ik11/inputs/access-om2/input_20200530/mom_025deg/topog.nc
as the default in the 0.25 configurations. Hopefully that won't turn out to be premature.
The old GFDL-based 1 deg and 0.25 deg topog is available via /g/data/ik11/inputs/access-om2/input_20201022_old_topog
in case anyone prefers that (but be aware that the land mask is different at 0.25 deg - see https://github.com/COSIMA/access-om2/issues/210).
At 0.25 deg it looks like either of those is preferable to the previous new topog in /g/data/ik11/inputs/access-om2/input_20201022
.
Just a quick update on the 1/4-degree OMIIP. I've finished a second cycle and the AMOC is back up to just under 12Sv by the end of the cycle (lower green line below). This is a lot better than with the previous topography (blue lines). So I will continue and this should become our production OMIP run.
Great to hear it - thanks for the update @rmholmes. It's still below the 12-16Sv we got in the GMD paper though - hopefully it will continue to pick up...
Martin Dix reports that the Baltic is reaching zero salinity in ACCESS-CM2-025.
We aren't having this problem in ACCESS-OM2-025 because there is very strong salinity restoring. @AndyHoggANU says
both OM2 and CM2 have about 1.7e7 kg/s of freshwater going into the Baltic. This volume flux is balanced by a net outflow through the strait, with almost no inflow of salt. The difference between the models is that OM2 has about 1.1e5 kg/s of salt added through surface salt restoring! This is enough to keep the salinity at about 6 PSU! But there is no salinity restoring in the coupled model, for obvious reasons.
CM2-025 is using /g/data/ik11/inputs/access-om2/input_20201102/mom_025deg/topog.nc
which is the latest GEBCO-based topography from https://github.com/COSIMA/make_025deg_topo/commit/f262fbf60a072f2f1f92b76e009f2a28cc08e1d9
This has the main sill and many shelves at the minimum depth of 11.8m (4 levels):
In comparison, the old OCCAM topography /g/data/ik11/inputs/access-om2/input_20200530/mom_025deg/topog.nc
has 40m deep in these regions (presumably this was OCCAM's minimum depth, and probably also 4 levels):
This is the difference (blue is shallower in the current topography)
These files can be compared with
/g/data/ik11/inputs/access-om2/bin/editTopo.py --ref /g/data/ik11/inputs/access-om2/input_20200530/mom_025deg/topog.nc /g/data/ik11/inputs/access-om2/input_20201102/mom_025deg/topog.nc
We may need to deepen the straits around Denmark to fix this (preferably not widening, so land mask is retained).
We have similar shallowing of the Baltic outlet in the new topography at 1deg, used in ACCESS-OM2 but not ACCESS-CM2.
Here's the current /g/data/ik11/inputs/access-om2/input_20201102/mom_1deg/topog.nc
, which has 11.8m minimum depth through the outlet
and here's the old 1deg bathymetry /g/data/ik11/inputs/access-om2/input_20200530/mom_1deg/topog.nc
used in ACCESS-CM2 and (I believe) ACCESS-CM. This is also from OCCAM, but this time with a 50m min depth at the outlet and in most of the Baltic
This is the difference (blue is shallower in the current topography)
obtained by
/g/data/ik11/inputs/access-om2/bin/editTopo.py --ref /g/data/ik11/inputs/access-om2/input_20200530/mom_1deg/topog.nc /g/data/ik11/inputs/access-om2/input_20201102/mom_1deg/topog.nc
~Do we have a similar low-salinity problem in ACCESS-CM2? Or does the wider outlet save us at 1deg?~ edit: ACCESS-CM2 doesn't use this new bathymetry
It looks even more constricted at 0.1deg in /g/data/ik11/inputs/access-om2/input_20201102/mom_01deg/topog.nc
(very narrow channels, with sills at the minimum depth of 10.4m) so we should check the Baltic restoring and salinity in ACCESS-OM2-01 too.
yes @aekiss the wider 1 degree channel does prevent the problem and we may have had to test it at the time we were aware of it being a trouble spot as it was also one in the CSIRO MK3 models runs, where the salinity went negative after 80 years there so almost the same time scale as this latest CM2-025 run.
This plot from Martin's email shows that ACCESS-CM2 uses the old topography, so we don't know whether the new 1 deg topography would cause problems without salt restoring. .
This is what it looks like in GEBCO: really shallow on average but with a thin channel about 60m deep, so changing it to 40m in the model seems justifiable in terms of the density that is accessible to the Baltic over the sill.
Yes @aekiss all those large vessels in those ports along the German, Polish, Swedish, Finnish, and Baltic states would have difficulty getting through the Westernmost of those channels at 12m never mind the submarines. I think 40m seems a more a reasonable depth to go for. Its only 2 tracer point and one velocity point wide at the narrowest point both your versions so perhaps we need to make it at least 2 or even velocity points, to see what flow we get.
Yes I think our values are too low because it takes the mean of the GEBCO depths unless we hand-edit it.
Martin also finds increasing salinity in the Mediterranean in ACCESS-CM2-025.
Gibraltar is shallower than in the old OCCAM bathymetry at 0.25° so maybe that needs digging out too.
There are a lot of comparisons here that may also be of interest: https://github.com/COSIMA/access-om2/issues/158#issuecomment-711438128
Hi @aekiss yes that strait of Gibraltar now looks very shallow will go and check the actual sill depths and the velocities before suggesting adjustments but it may need adjusting, I knew we got worse in the 1 degree runs but not as bad ad Martin's recent graph I must look at the PI control and compare it with the PD run he showed.
Mediterranean average SSS in ACCESS-CM2-025 grows in the same way as in the 1 degree run ( blue line is 1 degree PD, orange is 0.25 PD). The 1 degree PD and PI runs are very similar. .
Hi Martin yes the sill at the Mediterranean inflow is to shallow though this time we do have freshwater coming in from the Black Sea which was missing in the 1 degree run, will check it against ACCESS1-0/10-3 runs which got saline but not as bad as this later today.
Siobhan
From: Martin Dix @.> Sent: Friday, 22 October 2021 10:43 AM To: COSIMA/access-om2 @.> Cc: O'Farrell, Siobhan (O&A, Aspendale) @.>; Mention @.> Subject: Re: [COSIMA/access-om2] Improve topography at 1deg and 025deg? (#158)
Mediterranean average SSS in ACCESS-CM2-025 grows in the same way as in the 1 degree run ( blue line is 1 degree PD, orange is 0.25 PD). The 1 degree PD and PI runs are very similar. [med_sss]https://user-images.githubusercontent.com/1627802/138371247-d8cd1139-6b1c-4bd9-ab28-9201a861bb5d.png .
— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHubhttps://github.com/COSIMA/access-om2/issues/158#issuecomment-949081303, or unsubscribehttps://github.com/notifications/unsubscribe-auth/ADNNDAXIFZWEKEKL5OZU6HDUICQRPANCNFSM4IL237RA. Triage notifications on the go with GitHub Mobile for iOShttps://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675 or Androidhttps://play.google.com/store/apps/details?id=com.github.android&referrer=utm_campaign%3Dnotification-email%26utm_medium%3Demail%26utm_source%3Dgithub.
Just trying to keep up with this thread. This is really useful to help us tune the 0.25° OM, as well as the CM! But the question I have is what process to follow. Here, I am thinking we should:
sfc_salt_flux_restore
between the original and the new topography. For this, we will need a ~10 year run, I think(??)Are we happy with that? If yes, I am happy to do the test run, as long as it is after next week (my semester of teaching finishes Thursday afternoon!!)
At some point should we consider creating new topography for 1 and 0.25deg which is more consistent with the 0.1 deg topo?
The minimum depth is probably too large at 1deg and 025deg, now that we have finer surface resolution with KDS50: 45.11m (10 levels) in ACCESS-OM2, 40.36m (9 levels) in ACCESS-OM2-025, and 10.43m (7 levels) in ACCESS-OM2-01
The land masks are inconsistent, particularly near the tripoles:
There are also gaps in the depth distribution at 1 and 0.25 deg: https://github.com/COSIMA/access-om2/issues/141 and other problems at 1deg: https://github.com/mom-ocean/MOM5/issues/172 and non-advective points at 0.25 deg: https://github.com/COSIMA/access-om2/issues/210