 All right. Thanks a lot. So I'm going to continue talking a bit about the North Atlantic and bring back a bit about the Arctic and a lot of this is is rooted in some of the Canonical view. Where do I point this? Okay, good. The canonical view about how the North Atlantic circulation relates to sea surface temperature in the North Atlantic. So Gokhan Dena Bosoglu gave a very good introduction to this and right? And so the the view is that when you have a strengthening you have this AMOC in the North Atlantic and it varies in strength on time scales of years to decades and when it strengthens you would expect a strengthening of poleward heat transport and with that a warming of North Atlantic sea surface temperature and then over time there would be feedbacks that would that would arrest that and reverse the sign of that correlation. So the way that's typically portrayed is in the form of this lag correlation like I have in the right panel here and what that says basically you have lag in years on the x-axis and positive correlation at lag 2 indicates that when there's a strengthening of the AMOC about a couple years later you have warming of SSTs and you jump ahead maybe 20 years the sign of that correlation reverse reverses their feedbacks that kick in to arrest that. Okay, so in the decadal prediction community we like that because that gives us hope that you know you could detect some changes in the AMOC maybe a couple years later you'll be able to say something about what SSTs are going to do. The problem is that if you attempt such an analysis in across the C-MIT 5 models you might get something like this and this was a paper from Zhang and Wang a couple folks in Miami and they did an analysis of the historical simulations of C-MIT 5 and they found that the correlations between AMOC and NASST were all over the place and some models like the ones I boxed here show a picture kind of like the one I showed in the last slide where AMOC indices are leading North Atlantic sea surface temperature index by a couple years and then you see others like these guys where the correlation is negative or negligibly positive over most of the lag range. So the question is what's going on here? So that's the key question for us is why do models seem to show so much disparate behavior in the relationship between AMOC and North Atlantic sea surface temperature? Right and our answer is going to be is that a lot of this isn't really intermodal differences per se as I might look at first blush. It's a lot of it is an interference that's occurring between forced and unforced variability. Right so those calculations were performed over historical simulations where there's both forced and internal variability and those forced variations can happen on time scales that are short enough so that they can actually act in opposition to the relationship you would get from unforced variability and that's even after you linearly detrend that interference can occur. We're going to show that. Okay, so we're from Canada. We also care about the Arctic and polar bears and so we're also going to ask what are the implications of such interference for what might be going on in the Arctic and our answer is that there could be something there's not so much of this AMOC large scale ocean circulation connection, but when you look more regionally you can see manifestations of this interference between forced and unforced variability and we're going to highlight an example of that with the East Atlantic pattern and its relationship with sea ice concentration. Okay, so to just I think a larger aim of this talk is actually to hopefully convince you that you shouldn't be afraid of interference between forced and unforced variability. It's actually it's not something you should just grudgingly accept when you've run out of other options. You should be okay with it and and it should make you smile and sleep well at night. And and so this is kind of my bedtime story version of of how this interference can occur between forced and unforced variability and with the closed loop that you see here that's kind of a simplified version, a simplified portrayal of internal variability in the North Atlantic, right? So along the lines of what Goken and others have been been talking about you could just have a strengthening of the AMOC that comes about for any number of reasons and that will lead to a strengthening of poleward heat transport, a warming of sea surface temperature, and then some sort of feedback would kick in either an increase in the stratification, change in salinity gradients, then you would get a weakening of the AMOC and the cycle would close eventually. Okay, so that on its own you would expect a positive correlation between the AMOC and sea surface temperature with maybe the AMOC leading by a few years. Now if you add external forcing to the mix you get something like what you see in the bold arrows, right? External forcing like an increase in CO2 can lead to a warming of sea surface temperature and as a number of studies have shown that in turn can lead to a weakening of the AMOC which is believed to be due to an increase in the ocean stratification. Right, so if that pathway would push the system towards a negative correlation between sea surface temperature and the strength of the AMOC, right? So you can imagine attempting to compute a correlation between these indices where both of these are acting together it'd be unclear whether you should get a positive correlation or a negative correlation, right? So to see this in action we've done a calculation with the CESM one large ensemble, right? So the the big motivation for these large ensembles is that you have a ton of realizations and the ensemble mean that you get from this from this ensemble. All the internal variability should have averaged out and the ensemble mean should be a pretty darn good estimate of just the forced component, the forced variations. Okay, so this was a 30 member ensemble of or a 29 member ensemble of CESM one and when you see in the black spaghetti or all the individual realizations, the red is the ensemble mean, the blue is an observational SST product and the pink is the multi-model mean from CMIT 5 and in the top row what I'm showing are just the original time series of the North Atlantic sea surface temperature index in the left column and the Amock index in the right column. Okay, and then in the second row what I'm showing is what happens after you remove the linear trend and surprisingly not much actually happens in terms of anomalies because there's not much of a linear trend, right? And what and if linear detrending were sufficient for removing the forced variability that red line in the second row should be pretty much flat and it isn't at all. It has there are many variations in that red line that are not actually well separated from the internal variability and you can imagine computing correlations between the detrended time series, those forced variations will very well imprint themselves on the correlations that result. All right, so what instead will do what we'll consider the better approach in this context is subtracting the ensemble mean from each realization and then computing the correlations. So that's what we're calling the demeaned time series in the bottom row. So by construction the red line is completely flat in those time series indicating the fact that the forced variations have been completely removed from those time series. All right, so just by eyeballing this you should be able to see that if you computed persistence of North Atlantic sea surface temperature in the second row versus the third row, right? The persistence in the detrended time series is actually quite a bit higher than it is when you remove the ensemble mean and we've shown that in our paper, but what I'm going to focus on here is the relationship between the AMOC and SSTs. So what I have here then are the lag correlations computed between detrended indices and then for comparison in the bottom between the demeaned time series where all the forced variations have been removed. Right, so in the top, in the top panel what you see first off is that near lag zero depending on the realization you choose the correlation could be either positive or negative. Right, so this links back to some of the original plots I was showing where you might think it was just intermodal differences that were leading to the change in sign in the correlation. What this shows is that even with a single model you choose a different realization you can change the sign of the correlation. But if you remove all the forced variations they all show that the correlation is positive around lag zero. Furthermore, this appears to influence the lag around 10 years to 15 years, right, where you see that a number of realizations are showing positive correlation, significant positive correlation by a t-test which would indicate that the AMOC is providing a lot of significant predictive power for sea surface temperature. But if you subtract the forced variations from that then you get the second row, a lot of that apparent predictive power seems to go away. So this brings us to a key point of this which I think is relevant for decadal prediction, is that you can't, if you want the AMOC to provide information about what sea surface temperature is going to do and it might not actually do that the way you expect unless you're careful about how about how to break down the forced and unforced contributions, and I think that might actually have implications for the way you initialize models. If you might do a good job of initializing to the ocean circulation, but you that ocean circulation might not correspond in a realistic way to what the forcing is doing and that could lead to poor skill in your prediction. Okay, so we've seen behavior like this in a number of models including KAN ESM and the CMIT-5 models. Okay, so then we're going to think about about possible implications of such interference for the Arctic, you know. I'll highlight a bit of a negative result as we would see it. Rong Zhang brought up this paper earlier. It's showing the relationship between strengthening of the AMOC and sea ice concentration, and we've attempted such correlations ourselves, and the correlations come up pretty weak, which would seem to indicate to us that, you know, we don't have as much affinity for viewing the AMOC as driving these things. We do acknowledge that there's a there's a contribution from poleward heat transport to what might be happening in the Arctic, and further down the chain, the AMOC could be contributing to poleward heat transport in the North Atlantic, but when you attempt to link the AMOC directly to sea ice concentration, it becomes a pretty noisy calculation. The signals become pretty weak, and so we would argue that instead of looking at the large-scale AMOC influence on this, you end up seeing a clearer signal if you think more regionally. So what I'm going to do here is consider the relationship with the East Atlantic pattern. So the East Atlantic pattern being the second EOF of sea level pressure over the North Atlantic, the first EOF being the well-known North Atlantic oscillation, the second EOF is the East Atlantic pattern, and the positive phase of this would correspond to a cyclonic pattern over the Labrador Sea. So there are a couple possible relationships you could envision for the way this East Atlantic pattern relates to Arctic sea ice, and one is the subpolar gyre of what we'll call the subpolar gyre effect. So a strengthening or a positive phase of the EAP would lead to a strengthening of the subpolar gyre, which would in turn lead to a warming of the Labrador Sea and a cooling of the Arctic, and then sea ice changes that would correspond to that. So that pathway has actually been well studied by a number of folks in Bergen, in the Bergen group, so Hatun and Langahog have looked at that and established that relationship quite nicely. Now there's another effect that might come into play that would arise from forcing, so you can imagine if a force change in SSTs would lead to a melting of sea ice in the Arctic and then an equator-ward shift of the North Atlantic eddy-driven jet, and then corresponding with that a positive phase of the EAP. So the last link in that chain, I don't know if anyone has explicitly shown that, I'd be interested to know if anyone has, but everything up to that last link has been shown and has been suggested in earlier studies by like Clara Desser and others where they've forced a loss of sea ice in the Arctic, and they've seen an equator-ward shift of the eddy-driven jet associated with that. So I'm not claiming that we've totally pinned down that these are the two possible pathways occurring in this model, but just that they're plausible mechanisms at work. And what you see that, depending on the mechanism that's at play, you could either get a positive correlation between the EAP or a negative correlation between the EAP and Arctic sea ice. So the underlined quantities there, as I've indicated with the color coding. So to see some evidence that there's such interference occurring, I've done some calculations with the KANI SM2 model. So what I'm showing in the top row is, while in all panels is a correlation between the EAP index and sea ice concentration at each grid point for February, March, April averages. So what I have in the top row are the calculations from different time chunks of a pre-industrial control simulation. And in the bottom row I'm showing four different realizations of the historical period. So I've chosen time chunks from the pre-industrial control just to allay any concern that this is purely a sampling issue. And what you see in the top row is that the correlation patterns in the pre-industrial control all agree with each other regardless of the time chunk that you choose. But in the bottom row the patterns can be quite different depending on the realization that you choose. So in the top row it's showing very much the sub-polar gyro effect as earlier studies would have suggested it, where a positive phase of the EAP would lead to loss of sea ice in the Labrador Sea, gain of sea ice in the Arctic. But then other, but in the historical simulations it's showing evidence that a more of a forced signal is coming into play there. So that's just two cases I wanted to highlight of where this forced and unforced variability could be interfering with each other. So I first showed that this could be at play in our understanding of North Atlantic Sea surface temperature, and then I showed an example of this possibly coming into play in the relationship between the atmospheric circulation and Arctic sea ice. And so I think the key point coming out of this is that you really have to be careful if you attempt to predict changes in sea ice or temperature based on changes in the ocean or atmospheric circulation. And I think large ensembles are actually an extremely useful tool for dissecting this and for separating those forced and unforced contributions. I would like to think that you could go all the way with observations like others have suggested, but I've found that even with global mean surface temperature there's a lot of internal variability that can come up in such a time series. So it's not necessarily a guarantee that you're removing forced variations when you regress on global mean surface temperature. And I definitely don't think linear detrending is a solution in this context, and I think many of us in this room would have suspected that already. So with that I'll thank you for your attention and I'll take any questions.