 Thank you. I'd like to thank the the organizers for inviting me to come to this meeting and to talk about it some work that we've been doing over the last Several years and this is sort of a little bit of a summary of Perspectives towards and relevance for climate prediction And I'd just like to firstly acknowledge my close collaborators on this work. Jin bar Jenny making and a noradin Omrani So I was thinking I had to link up with the previous talks because the session was a We switched from a proxy of data to the middle of the ocean atmosphere interaction and so I thought I Put this slide up for two reasons one is this because I think it's very important that we should try to and I want to Strengthen the message of the previous talks. We need more data. We need really truth to understand what's going in the North Atlantic sector Or actually globally on these timescales So we see here a reconstruction That we were involved in and the other reason I wanted to put this up is because it's in a marine proxy And because we've used five records in a very similar way to what was just spoken about And we find the leading PC for tropical Atlantic SST Coral records which explains about 32% of the variance of the data gives you a very nice Multi-decadal signal and we have now essentially one more cycle from this data that we I could say we could trust But we didn't we're not brave enough yet to try to do what back another 700 years. We don't have the data So well that that that's sort of the baseline and we need more data Actually to understand things like the oscillatory nature of Atlantic multi-decadal variability So the things that I want to talk about today So I want to there's three simple messages that I want to convey the first one to talk about atmospheric forcing of the ocean variability And it's related to the talks we saw earlier in the day by Tom Delworth and Gokhan, and I don't try to say his surname sorry But the question I want to address here is how consistent is the slab ocean a GCM interpretation of AMV? so it's only three slides on this and The second part which is a Probably a more interest here is the is summarizing some work on the ocean forcing of atmospheric variability and in relation to any oh and the last part is on Maybe some ways forward in the decadal prediction community of how things that I see that might give us, you know Greater improvements in doing our prediction work So I just like to start with this slide. It's a very common slide It's been familiar to the talk it's presented in a slightly different way here So here it's a composite analysis and it's done for the winter sea surface temperatures It's a composite for Atlantic multi-decadal variability on the left You see the sea surface temperature on the right. You see the sea level pressure pattern So the first thing I'd like to point out is that the if you do it this way You don't really have this type of horseshoe pattern at least it's not so prominent It is much more like a monopole like structure and then we have a this NAO like signal negative NAO like signal that's associated with these changes that is now becoming quite well known and then so This structure obviously cannot reproduce and as a direct thermal forcing a horseshoe pattern Particularly in this type of region where the winds are of the wrong sign They should produce a cooling. That's the sort of things that were discussed in this paper from Clemens So I would argue firstly that the pattern doesn't really match the pattern in the paper The second thing I'd like to discuss is the the stochastic null hypothesis that was put forward That it should be essentially an AR-1 so slab ocean coupled to an AGCM. Now we know the ocean is actually dynamic And so we have this is a very similar set of work that was shown Earlier today by Tom Delworth except we're doing it in an uncoupled context and we instead of doing a you know Applying any old time series of particular periodicities. We've applied a stochastic time series with a thousand Nearly two thousand year length record And so here you just see the the spectrum of that and the wavelet spectrum of the things that we are going to drive this Half-degree ocean model a full ocean general circulation model And so you can see a white spectrum. So what do you expect as the response in the ocean to these things? So here you see two very common indices that are discussed throughout The the talks we see on the left the amok at 30 north and on the right you see the subpolar gyre strength spectrum Now neither of these is an AR-1 spectrum because they have a the integrate It's the ocean dynamics that are integrating the stochastic forcing So you don't no longer expect a simple AR-1 spectrum And in fact the a the amok spectrum the best fit here is an AR-7 and for the subpolar gyre it is an AR-5 and So they have quite different characteristics as you can see and what we have done from further analysis We can identify sort of two different regimes in this state in the the model one is on the much longer time scales sort of a hundred years or longer, which is a sort of a quasi equilibrium state just in sort of quasi equilibrium with the NAF forcing and involving a The Mahelan component and then we have a more heat flux driven component in the In the sort of a decadal shorter time scales, which is very similar to the Saravanan and McWilliams a mechanism or And it's a stochastic spatial resonance that's been discussed before Now if you look at the SST from the model again that has elements of both of these This is Atlantic multi Atlantic multi-cadal variability index and again it is not an AR-1 So I would argue that the the null hypothesis is not even though this is a stochastic climate model It's not necessarily correct to assume. It'll be an AR-1 process And the last this this was actually shown earlier today by Gokhan, but in a different representation So that region of the North Atlantic where you have in you would have expected a cooling based on the negative NEO like structure We actually see an outfluxing of heat and this is shown here by the reconstructions that Sergei Gulov has done And you see here the time series of a turbulent heat fluxes in a in green and the SST reconstructions in red for that band here and you see on these multi-decadal time scales They really covariate of each other and the sign is such that it's the flux going out of the ocean So it is not a region that is driven by the turbulent fluxes So I think that sort of addresses that were the null hypothesis And it shows that the pattern is not really consistent with a slab ocean model coupled to an AGCM Now so we saw the possibility that there's a flux coming out of the Gulfstream region All right So that has the potential to drive large-scale atmospheric circulation changes and there's two ways that That we can think about or revisit this question that has been addressed for many years and what people have been looking for So these days we have a high-resolution models and we have high-resolution data and we see things like the sharp SST fronts and Maybe the the frequency of the SST data. So if we use daily data rather than monthly data we potentially get a much stronger response than what was found before and Mojib talked about this earlier on Monday When he referred to a paper from theirs on the North Pacific So what I'm going to talk about is that the bottom part the role of the strategy or the potential role of the stratosphere troposphere interaction and the atmospheric response to the large-scale SST changes So here's a Schematic of the not a schematic sorry a figure for the 1960s warming warm period 1950-1960s warm period You see the SST anomalies on the left that were observed You see on the right the sea level the geopotential height anomalies at 1,000 hectopascal and at the bottom you see the response In atmospheric model driven by the observed SSTs So you see here the atmospheric model that resolves stratosphere troposphere interaction Well, it's able to reproduce the negative any O like structure with quite reasonable strength actually and The low-top model the mother does not resolve the stratosphere troposphere interaction is not able to We've looked into this into more detail and we see that it's really you can get you have a consistent weakening of the polar Vortex earlier in the winter and this is what is driving the the tropospheric signals now that was one case So how about It's difficult in the observed record to look for more cases So what we've done here is we looked in a high resolution a high-top coupled model simulation It is a more than 500 years of simulation Very similar analysis the left shows you the composite SST from the model It has a slightly different structures. It now has a hints of this horseshoe pattern The left is the geopotential height, but now at 500 hectopascal to put a 500 hectopascal You see a negative any O like structure going along with this SST Now if you take that SST and drive an atmospheric model the same atmospheric model You actually reproduce the structures very clearly, right? So we're able to reproduce those structures with the observed the model SSTs in the North Atlantic only and Here I show you the stratospheric response So here's the response in the coupled model or the associated patterns in the couple model of that SST And here's the response of the uncoupled model. So we see this is further confirms this mechanism Which can be sort of schematized in this figure here So we tried to bring together various elements here of what we think is going on in the North Atlantic So we have here a schematic of equated a pole and here we have put a warming in the middle attitudes The blue the blue here shows you the the westerly winds You can see at the troposphere and the stratus the troposphere and the stratosphere Now the first-order effect of the the warming in the high latitudes is to reduce the baroclinicity So this will lead to a synoptic scale variability and will actually also force a larger-scale planetary waves These can propagate into the stratosphere And if your model has a well-resolved stratosphere or it has no nudging in the stratosphere It can lead to a weakening of the polar vortex and a warming so we can have a warming up here in the high latitudes and a weakening of the winds which are Schematized by these red dashed lines and this signal that can then propagate back into the troposphere enhancing further the the meridional The low-level baroclinicity in the troposphere and leading further to a negative any or like response Yeah, so there was at the end I will come to some further studies that support this work From other groups. I thought there was here the slide. So now What are what sort of insights that this give us to the prediction problem? I mean it seems that we may have a coupled coupled a Coupled the type of variability in the North Atlantic and in general Climate prediction problem is a couple a coupled issue so Just to start the discussion. I put up a schematic here or conceptual view of the prediction problem And the each of these circles represents a differing Dynamics different different Different representations of the predictable dynamics. So we have the observed very good the observed rep. Well the truth here and the observer observed climate We have a model climate. So these are the free-running models These obviously differ from each other been talking about that a lot and then when we do a prediction We actually introduce observed data and By because we are now even in the most optimal way that we might try to do this with advanced data assimilation techniques We will introduce terms into the dynamics that will actually lead to a different representation of the climate at least at the initial state so the You could say that the most the reliable part of this diagram is that in this center so to improve prediction basically we have various strategies, right to get these circles to overlap more closely and to Expand this blue region so we can get a better idea of the mechanisms We can also use data assimilation to reduce to combine these circles We can also use data assimilation methods as I will argue to improve our understanding of mechanisms and together by Collapsing these circles we can get a better idea of predictability So that's sort of a very conceptual view. So the prediction problem. I would argue Rises if we start with a starting point is the climate predictive Climate predictability arises from the interaction of the slowly varying components of the climate system with the atmosphere So then I would argue that is actually the misrepresentation of these interactions That is probably the major course of error in our climate predictions at the moment So these lead to errors in the mean state and the variability of the models in their free-running climates When we introduce observed data into the models it leads to strong forecast drifts that they create the grade the forecast skills and the these render the models actually a poor estimates of Providing us poor estimates of climate predictability or the limits of predictability So I think that couple data assimilation actually provides a pathway forward in this issue because you can You can consistently treat the dynamics in the different components subcomponents of the system and so you can You can reduce errors I think also in the mean but in particular in the in the forecast drift at the initial stage and it's not it's I think you have to take There are clever ways to do this in a sort of a brute force couple data assimilation I think if you if you take into account the underlying predictable dynamics Then you you have a better chance of dealing with this problem And I just want to show that the power of this type of method from some work that we've been doing in Bergen and this is a We're working towards couple data assimilation so far We have been doing a data assimilation only for the ocean, but here's a data using the ensemble common Filter data assimilation method. We've been assimilating only sea surface temperature anomalies into the Norwegian earth system model and here you see how Only if observed sea surface temperature We can propagate Information into the the deep row ocean and actually able to constrain that the the Subpolar Jaya strength in the North Atlantic So the the blue is our free running 30 member ensemble The this pinkish line is after we have done the data assimilation of observed SST and The black line is actually the observed sea surface height index of subpolar Jaya strength So actually with only a sea surface temperature. We can propagate information through the coverances of from the ensemble Into the deep ocean and constrain Yeah, a large part of the variability that you might not expect and I'm I can see that this is a very powerful method to Extend, you know information from say sea surface temperature or the ocean to the sea ice or looking at the atmosphere land interactions I think this has a really a great potential and we have to explore it And so just in summary I Think the the slab ocean a GCM Interpretation of the AMV is not really consistent of our current understanding. I think we probably largely agree of that in this group and I talked a little bit about Observations and models and how that they indicate that the warm AMV tends to drive a negative NEO And the stratosphere plays a key role in this I didn't discuss this but in our experiments We actually have quite some difficulty in reproducing the response to cold AMV conditions We able to reproduce the response to warm We think that this is related actually to the background state of the winds and the westerly winds in our model are too strong And it's quite sensitive of this. The last thing I would like to point out is there are now a growing body of work from other Groups that are also supporting this that are North Atlantic SST are actually able to drive a negative Influence the low frequency variability of the NEO Because I thank you