This post is a follow-up on discussions at the 2025 ACCESS-NRI workshop with @Matt_Woodhouse and @rmholmes.
One of the long-running challenges of interpreting experiments using an ocean-sea ice model is the over-strong constraint that the surface boundary conditions place on surface variability, especially surface air temperature, since the input conditions can’t evolve with the surface ocean conditions. This definitely impacts sea ice (which is very slavish to to input SAT) and some processes in the tropics.
A potential solution (or at least test comparison) is to run the coupled model but ‘nudging’ the atmosphere at some level above the boundary layer to reanalysis conditions (e.g. winds) - so you can get an IAF run with realistic historical conditions but freely evolving boundary conditions. @Matt_Woodhouse reckons this is pretty achievable with the current model set up.
So, this is a call-out to the community (especially COSIMA) to see:
would you be interested in such an experiment?
thoughts on experiment design - what forcing data (ERA5 vs JRA55-do), what level to nudge (850hPa, 500 hPa etc)
Thanks @willrhobbs. I think this would be a cool thing to do. Besides sea ice, there are applications in the tropics in terms of freeing up intrinsic ocean variability (e.g. I used an atmospheric boundary layer model to free-up intrinsic ocean variability for an ENSO study in this paper). That said, personally I obviously don’t have any students or any projects that would benefit at the moment.
Others who might be interested, or other applications I can think of:
@jemmajeffree - could go someway to addressing some of the issues we discussed last week
@navidcy - I wonder if you’d see some increases in the variability you looked at in this paper by removing some of the atmospheric damping (although maybe it’s not relevant for the decadal time-scale)?
@ellepola, @Bela Conde - I wonder if this kind of approach could also help with your studies - e.g. by freeing up the ocean SST to react more to changes in/biases in winds.
Hi All, I agree this is a nice idea. The one catch with using atmospheric nudging is that it would come with a significant computational cost - because I think you would have to run the whole UM. And it might have to be at high resolution if you want decent coupling to a high-res ocean …
An alternative is to have an embedded mixed layer which can do a similar thing but with lower cost. The one example of this that I know about is CheapAML:
Were you thinking of restoring heat and winds etc, or just winds? I suspect we might need to run a few short runs with nudging at different heights to see how strongly they constrain the ocean (which could be interesting in itself, really) to pick a good level for a longer run.
Re Andy’s embedded mixed layer idea, I don’t think that would suit my purposes because I want large-scale atmosphere behaviour (especially zonal winds) to respond to the ocean surface. I’d be fine with a coarse resolution UM shoved on top of a high-resolution ocean though. Also, I was under the impression that with ACCESS models the ocean is by far the most expensive bit. ie, OM3 takes nearly the same SU as CM3?
@AndyHoggANU in the paper I linked above I built an updated version of CheapAML and added it to ROMS. I agree this alternative approach of running a full coupled model would be more expensive, but the advantage is it requires no coding to get going…which lowers the bar to actually doing something (although maybe with NuOPC in the new models an ABLM could be really easy?).
@jemmajeffree - I actually thought it was the other way around, and that the atmosphere is more expensive because of the requirement for a smaller time step. Would be great to get some clarity on that from someone who knows? Also - it sounds like for your application (zonal winds responding to the ocean) - you just want to run the full coupled model with no nudging?
Echoing what @rmholmes stated, I agree that an ‘emulated’ boundary layer is probably the long-term solution, but this is a nice stop-gap and we can perhaps use it to estimate just how ‘bad’ the surface air temperature constraint is. With NCMAS grants coming I’m minded to put in for an allocation from AAPP to do this as a sea ice-specific project, and of course make the run open to everyone.
That’s a good question, and I’m hoping to ahve a bit of discussion about the best way forward. Personally I’d like to set the clouds and/or radiation as well since that’s obviously important for the surface energy budget, but this outside my expertise so happy to defer to others (@Matt_Woodhouse ) on what a sensible experiment protocol would be.
U, V and potential temperature are the typically nudged parameters, though potential temperature is optional. I don’t know of any facility to nudge clouds and radiation - that’s the domain of full data assimilation and well beyond what nudging is intended to do
CM3 is 30 kSU/year for the 25 km version. I remember OM3 being twenty-something kSU/year? But I’ve been kicked out of the NRI internal chat so I can’t verify the second one
CM2 is 12.5 kSU/yr, OM2 is 7.2 kSU/yr if both at quarter degree ocean and CM uses an N96 atmosphere (reference: Adele’s NCMAS application)
It sounds like I shouldn’t reply to exciting hive posts while still in holiday-brain-mode
I don’t think glueing on an atmospheric mixed layer would produce an ENSO? Or am I misunderstanding?
I think what I was trying to get at is a way of quantifying the spread of ocean model ensemble members if the “directly responds to the ocean” bit of atmosphere is allowed to respond, but higher altitude atmosphere is still constrained. Maybe this would just constrain Walker Circulation and thus zonal winds, maybe not. Anyway, I’ve only got a vague idea of I’d want to do with such experiments, and could probably find something useful whichever way we go
Nope it won’t. There were two parts to the atmosphere in my Climate Dynamics study; 1) an ABLM which removes some of the damping of intrinsic ocean variability by the fixed atmospheric air temperature/humidity (the point I brought up during your talk 2 weeks ago which would afflict the OM2 simulations), 2) a simple statistical coupling which modifies the wind forcing based on SST anomalies (e.g. if the eastern Pacific is warm, the trade winds are reduced, based on an SVD). Only 2 will give you an ENSO. I think both are important to properly quantify the impact of ocean intrinsic variability on ENSO, but in this case we are only talking about 1.
H’m I noted today that you’re a “guest”. DM me if there’s a particular channel you’d like getting access back to (that’s the way guest accounts work, need to opt back in).
