 OK, welcome back. In this second lecture, I will basically develop some more thought about the connection between North Atlantic regimes and what happens in the tropics. There are a number of things I would like to talk, probably too many. So we may not have the time to cover the last point if you want coffee, which certainly would be nice to have. So we'll briefly review a comparison between cluster analysis over the Atlantic and Pacific to try to raise the question whether these regimes are connected or not. Then I will look at some modeling studies on the impact of tropical heating in the Indian and West Pacific Ocean on the Atlantic and Pacific flow. I will show you some results of some observational results on teleconnections within the Indo-Pacific rainfall that are from GPCP data and ECMWF analysis. And I will talk about the impact of Atlantic and Pacific regimes on surface heat fluxes. And if I have time, probably not. I will talk about the stratosphere anyway. You'll have the slides. So as I mentioned before, while in the Atlantic, you do a cluster analysis, different people have done it. And we all find basically four regimes. And they are more or less all the same. That's fine. In the Pacific, it's a bit different. And sometimes the domains overlap. And so a few years ago, there was an activity here at ICTP on flow regimes. And so I decided to actually rerun these cluster analysis on data from the era interim re-analysis. So I took five domains of geopotential height, 500 hectopascals in DJF. At the time, the latest year was 2012-13. And so I used the same method as Michelangelo et al. And also the one that we used with David Strauss, the K-means clustering. And I chose basically two domains, one covered in the North Atlantic, one covered in the North Pacific. And then a domain which is just the junction, just the same two together. So it's not a completely hemispheric regime, but basically covers from the Central Pacific to Eastern Europe. So the idea was to say, how do the regimes in individual sector compare with the ones over the larger domain? OK, so the Atlantic, well, no surprise, I find the same four regimes of everybody else. So these are the mean anomalies. We have already seen them. So in this particular sample, the most populated is the positive NEO. Then you have the blocking. Then you have the Atlantic region, very similar to the negative NEO. But OK, so far. For the larger domain? No, this is just for the Atlantic domain. So from here to here. Now, if you do that over the Pacific, and perhaps because my domain is a bit smaller than the one that was used by David Strauss, again I find four regimes. Some of them are similar to the two that we found in an earlier study, so the Pacific trough here and the Alaskan region, they seem to be quite the opposite of one another. The other two are also reasonably symmetric. You have something of an Arctic law and the negative P&A and the positive P&A pattern. And the difference being that in this case with the Alaskan region, the Pacific trough, the center of the anomaly is located a bit north. And then you have a much stronger meridional flow in the center of the Pacific. Now, if you do it on the extended domain, you get six clusters. I'm not going to talk about each of these six clusters, but it's interesting to compare the structure of the most populated regime in each domain with the structure of the leading empirical orthogonal function. So again, I did an empirical orthogonal function analysis to reduce the number of degrees of freedom, and I only used the dominant principal components. So if you look at the most populated cluster here and the leading EOF in the Atlantic, they look very similar, it's an EO pattern. You compare the Pacific trough regime with the leading EOF and again, very similar pattern. But if you look at the most populated cluster on the hemispheric scale and the first principal component, you get two different things. So in the linear part, you actually see a sign of anomaly, which is opposite between the North Atlantic and the North Pacific. But you actually find that the most populated cluster, although others have not two different frequencies, but the one which is most populated actually has laws over both the North Atlantic and the North Pacific. And it has a wave number of two structures that is not very different from the cold ocean wormland pattern that Mike Wallace introduced in the 90s. So the question is, yeah, another interesting thing is that you can say, oh well, but maybe this is just a random combination. So sometimes it just happens that in certain days you have the positive EO. In other days, you have the Pacific trough. These are completely independent. And sometimes it happens that they have the same sign. However, if this was the case, then the frequency of the cold regime should be just the product of the frequency of these two. So if you turn this into a fraction, this is 0.3. This is 0.3. So the product should give you 0.09. And instead, it's 0.19. So somehow this suggests that there is some mechanism that makes the combination of the positive EO and the Pacific trough more likely than what you would just expect from a random combination of Pacific and Atlantic regimes. Now, this issue of the connection between the Atlantic and the Pacific, I just skip a bit to some local history. Because this is a problem that has actually caught the attention of Fred and myself for quite some time when we were working together here. And this basically came as a result of two papers published by the anchor group and collaborators, a group led by Jim Harrell. And what they did, they used the anchor atmospheric model, the time was CCM-3, forced by observed SSTs in the period from 1952 to 1999. So the study was published in 2004. As you well know, by the time you do a study and the time it's published, it takes some time. So this was basically almost as recent data as they could. And this was just after the big 97, 98 and Nino. So what they found is that they actually found a trend in their simulations. So they actually did also simulations with other models. So they had a CCM-3 tropical ocean and global atmosphere. And they had a multimodal global ocean and global atmosphere ensemble. And if they computed basically the linear trend in these 50 years, they came out with a pattern that had a strong wave number two component. And that was in reasonable agreement with the observation, at least as far as the phase was concerned. In terms of amplitude, it was slightly less than half of the observed signal. They found that this was associated to increase precipitation in the Indian Ocean and in the Central Pacific. And if they actually plotted the distribution of the NAO index in January, February, and March, so they had a control simulation where they just prescribed climatological SST. And this gave basically the distribution, which is the dotted line, which is basically centered around zero, so no impact on the NAO. But when they prescribed the observed SST, then the distribution was shifted towards positive values. So there was an indication that somehow the change in the NAO during the second half of the 20th century was somehow related to changes in SST. However, they also plotted the observed value of the change, and this was almost twice as large as the peak of the distribution, although you could see that it was in the tail. So the question remained whether the discrepancy was due to a modeling problem or simply to the fact that actually just happened that the NAO has changed to very large positive values towards the end of the century. So this is a question that Fred and I worked on for a while, other people did similar studies. And if I have to say that 15 years later we know the answer, unfortunately, it's not the case. But it's a very interesting problem. So some of the students want to find a challenging problem. This is a challenging problem. Now, in particular, they did also experiment when they prescribed just a linear trend of SST in specific regions, just the whole of the tropics, then just the eastern hemisphere and then only the Indian Ocean. And what they found was that the strongest signal in terms of NAO was actually found when they just prescribed the Indian Ocean. The question that came to my mind, that's why Fred and I tried to look more at what happened in the West Pacific, is that in this simulation, yeah, you have a strong positive NAO, but the signal over the Pacific is completely opposite to what happened in the full simulation. So you cannot say, oh, you can explain everything just by the Indian Ocean. You clearly need something else. Otherwise, you get the wrong sign in the Pacific. Now, let's move now to more recent times and to a different time scale. We go to the decadal to the MJO time scale. I've mentioned that you can look at the frequency of Atlantic regions, but you can also do a simpler thing, which is just make a composite of the geopotential hatch flow. For example, 15 days after the MJO is in phase 2 or 10 days after the MJO is in phase 3. So phase 3 usually comes five to seven days after phase 2. So roughly, this should give the same results in the extra tropics, and in fact, it does. You can combine this to have the overall effect of convection of the Indian Ocean during the MJO. And again, you get this wave number 2 with the same sign over the North Atlantic and the North Pacific. So somehow, it seems that the forcing that comes from the tropics seems to favor this connection between the Atlantic and Pacific regions in such a way that you have the same sign over both oceans. And again, people who try to model that, for example, this is a paper by Lien et al, published in 2010. They did the experiment prescribing some idealized heating in a dry primitive equation model. And this is the result in terms of geopotential height, six to 10 days after the start of the heating. This is 11 to 15 days. Now, you won't probably see very nicely the coastlines, but while in all my pictures, Europe is here, here there is North America, so in this case, Europe is around here. So 11 to 15 days after a dipole with the heating in the Indian Ocean and cooling in the West Pacific, you get some sort of positive energy signal. But again, if you look at the Pacific, you get a high while we're seeing in the observational composites, we get a low. So there is this issue of why it's not easy to model these teleconnections. So if you model the Atlantic right, then you somehow get the Pacific wrong in these idealized studies. So you need to put all the various basings together maybe to get the right question. So we're talking about the Keter, we're talking about sub-seasonal. We can look in the middle and look at the seasonal. And a few years ago, with Tim Stockton and Frédéric Vitale, we looked at these teleconnections in the system 4, seasonal forecast system. And as a comparison, we also looked at observational data. So we're not talking about system 4 here, but we'll talk about the results we got on the observational data. So we did something very simple. We just averaged tropical precipitation anomalies over boxes in different regions. And the different regions were chosen according to the local correlation between SST and rainfall. So you have two regions which are the central and western Indian Ocean and the central Pacific, where the correlation between SST and precipitation is positive. So there's a clear indication that the SST force is part of the precipitation signal. While if you look in the region in the eastern Indian Ocean and around the maritime continent, then the correlation is much weaker. And in some areas, it even becomes negative, sign that the SSTs are actually forced by the radiative fluxes. So we choose these three areas, one here, one here, and then one in the center. We computed basically an index of the rainfall anomaly. And then we looked at the covariance of this index with geopotential height. These were seasonal means because we were concerned with seasonal forecast. So the data were for DJF. And the period was 81 to 2011. That was the span covered by the system for re-forecast at the time. So these are observational results. These are not modeling. So if you average the precipitation in the western and central Indian Ocean, you compute the covariance with geopotential height in the northern hemisphere. And you get, again, this wave number 2. So observationally, you may say, well, there is clearly a connection between the Indian Ocean and this wave number 2 pattern. However, if you look at this map, you can also see that the rainfall in this region is negatively correlated with rainfall over the maritime continent and is possibly correlated with rainfall in the central Pacific. So you may say, well, maybe it's not only the signal which is coming from here that produces this wave number 2. And in fact, you can play with this and modify the index. And for example, you can say, well, what about if it's not just a simple heating, but like in the MJO, you may have a heating here and the cooling there. So you can construct your index by taking the difference between these two regions. And if you do that, you again get wave number 2, but the Pacific part becomes more definitely stronger. So this is an indication that the suppression of convection here is important to explain the law over the Pacific region. Then you might say, OK, well, but maybe it's not this region that matters, big anomaly is here. However, if you do that, and then you compute the covariance from rainfall in this region, well, again, you have a tripolar part then in precipitation, then your Pacific signal becomes very strong. But the sign of the Atlantic just goes the other way. So it seems that it's quite a tricky problem from a modeling perspective, because it seems that you must have the right ratio between the anomalies in the Central Pacific and in the Indian Ocean. So you need this tripolar pattern. But if you have a very strong El Nino and that power basically dominates, then the sign of the NAO will actually flip from positive to negative. OK, now the results I showed were with era intermediary analysis. But you may be aware the DCM that we have now does so-called 20th century reanalysis. 20th century reanalysis are reanalysis that are done, again, the name it says, starting from 1900, with a very reduced observational set, basically only surface pressure data. And the reason is that if you do a full reanalysis using all your observation, your analysis may have trends, simply because in the last part you have all the satellite data, for example. So what people do in the US, similar studies have been done, in this analysis you only use data sources which you find basically throughout the period of your reanalysis. And you hope that your data simulation system and the model you use for your data simulation system is good enough to transport the signal from the surface to the rest of the atmosphere. The other thing you do is that you force the system with observed SSD. So the latest incarnation of this reanalysis is what is called CRA-20C, C stands for coupled. So it's a weekly coupled reanalysis covering the period from 1900 to 2010. One warning, you should not actually use it for changes in the deep ocean, because it's for practical reason the ocean has some discontinuities. But basically the upper ocean and the atmosphere you can think about as a continuous system. But the advantage is that you can go back. And so the first thing is what about if you use this new data set to look again at the 1981 to 2010 period. And if you do that, this is the map you get for rainfall, from the Indian Ocean. This is the map you get from geopotential height. And that is the map that you get for temperature at 850 millimeter. So the first thing that you can say is that, well, good. If you use this particular data set, you get results which are very similar than using GPCP and era entry. So you see this tripolar pattern. You see the wave number two. And interestingly, you see a warming over the northern continents. And again, this fits with the idea that this pattern is somehow the equivalent of the Karl pattern suggested by Mike Wallace. So we have negative anomaly over the northern ocean in temperature and positive anomaly over the continents. So the continents are relatively warm. And the oceans are cool. But now you can go backwards, because you have more data. And since I'm involved in a EU project called Primavera, which is about simulations from 1950 onwards, I actually looked at a period from 1951 to 2010. So this is a 30-year period. This is a 60-year period. And these 30 years are the last 30 years in this data set. You do your maps, and well, it's nice. You get almost the same pattern than in the last 30 years, if you put 60 years together. And the geopotential high pattern is also not very different. So again, the main two troughs here, so positive an AO trough over the North Pacific, positive anomalies over the northern continents. Here, again, I've used seasonal means. So you can think about this map as representing inter-annual variability. However, because this particular data set spans 60 years, there's also potentially a decadal signal in these maps. And so I wanted to see whether the decadal variability played a role, again, thinking back of these simulations by the Jim Harrell group. And so I just plotted the difference between the first 30 years and the last 30 years in the record. So this is the inter-annual teleconnection that you've seen before. And this is the difference between the last 30 years and the first 30 years. This is the plot for rainfall and geopotential height. And I found it quite amazing how similar these are. This basically tells you that in the 60 years, there's not only a contribution from the inter-annual variability, but there's also quite a strong contribution from the decadal variability. And again, if you look at the geopotential height, it's remarkably similar. And in fact, it's very similar to the maps that were published by Jim Harrell and so on. So for some reason, on this decadal time scale, there is this combination of two positive rainfall anomalies in the Indian Ocean, in the Central Pacific, suppression of convection in between. But again, the tricky thing is that these two anomalies have to be comparable. Because if the Nino one dominates, then you get something very different in the Atlantic. So the challenging question is why it happens like? Why are you separating 80, 81? Or because it's just half, the first half of the record, second half of the record. And then it just happened that 1981 was the period that we used before because this was the period of the system refocus. But in this case, it's just last 30 years minus first 30 years. Now interesting, if you now look at the difference in temperature, now this map is mostly warm. Well, there is global warming, so you would expect that the difference is mostly positive. And in fact, it is mostly positive. But if you actually look at the regional differences, then again, they are very much related to this particular pattern. So it's true that the old hemisphere, apart from some regions over the northern ocean, is warm. But in some region, it has warm much, much more than in others. And basically, this teleconnection pattern tells you which are the regions where the warming has been stronger. This is again only the warming in winter. We are not talking about the whole year. We are not talking about the whole globe. But so if you have to create some climate services, which are very popular in these days, and you want to know, for example, whether here in this region, in the Alps, you will have snow accumulation during winter, which is very important for agriculture. Because here in the Mediterranean, the winter season is the rainy season. So if you have a cold winter, a lot of the precipitation will be snow and will accumulate and then will melt and basically give you back the water during spring. If the winter is warm, the rain will fall immediately. So it may be interesting to know whether over one particular region in winter, the global warming trend is stronger or weaker, or may even be reversed. And in fact, you probably all know that if you actually look at shorter periods, and particularly what happened after 1998, the global warming trend slowed down quite substantially. And people have thought about possible reasons. And there was a very famous paper by Kozaka Enchi. And they did a pacemaker experiment where they used a couple of models and they basically relaxed the couple of models towards geoservices in the central and eastern Pacific. And they showed that by doing that, they could reproduce quite a lot of features in this slowdown, including its seasonal cycle. So the slowdown was stronger during the northern winter. So this basically, the interpretation was that you cool the Pacific, so you decrease basically the heat transport from the tropical Pacific to the global atmosphere. And therefore, you decrease the heating into the atmosphere. And these specific trends somehow slowed down, partially compensated the global warming trend. However, if you ask yourself, where are the regions where there is the strongest variability in the heat fluxes on the inter-annual time scale? Well, you definitely see the linear region here. So this is the region of the Kozaka Enchi paper. But actually, a much larger variability is found in the region of the Western boundary currents. Also somehow, in some regions of the southern hemisphere as well, but definitely in the area of the northern boundary current. So you may think about, what about the variability of those regions? So can the variability of the heat fluxes in those regions affect the decadal trends? And there have been papers published on this. For example, these are two GRL papers that have actually looked at the CM5 models and have actually pointed out that if you want to somehow to characterize the decadal variability in the warming trends in the five simulations, the single index that seemed to explain more was not the variability in the heat flux in the tropics, but the variability of the heat flux at the high latitudes over the ocean. So we may want to look at this. And again, with Fred and other colleagues here at ICTP, we somehow tried to find an index that was actually inspired by earlier works on thermal equilibration of planetary waves and John and one of them, that basically was trying to relate the variability in planetary waves to the heating and interaction with the surface. So having basically different equilibria, different flow regimes associated with different mid-latitudes of the heating. So since one of the predictions of this theory is that in one case you would have a much stronger zonal flow and stronger heat fluxes, we constructed an index that was basically the difference between the net surface heat fluxes, basically over the north and ocean minus the fluxes over the continents. And we looked about how the circulation was changed in different phases of this index. So when we first did that, first of all, one interesting thing that we found is that in the period when there was the slowdown, this index was mostly negative. And the other thing is that if you compute the covariance of this particular index with your potential height, then again, you'll find the wave number two pattern that we have seen before. And the reason for that is quite simple is because this particular pattern gives you north-western flow in the regions of the western boundary current. So you are back cold there, and also, you increase the zonal wind speed in this region. Because this pattern is actually in phase with the time mean circulation. So if you have an anomaly of this sign, you increase the strength of the surface flow. And you are back cold there over the warm ocean. And so in this case, this map is for downward surfaces, heat flux. So blue means that you actually have a transfer from the ocean into the atmosphere. So you can have this variability of the heat flux. It's also on inter-annual timescales related to this wave number two. And so the question is, well, if there is decadent variability in this index, and again, maybe this is influenced by the tropics. So I showed you before. This was the teleconnection on the inter-annual timescale with this index of tropical convection. And what I showed in the previous slides is this pattern that comes from the teleconnection with the surface heat fluxes. And you see that your potential heights are relatively similar. So you can say, oh, what about if the correlation is 0.71? It's not 0. It's not 100%. You have to admit. Then you can say, OK, so then you can relate the surface heat fluxes with the topical precipitation. And you get something like this. So of course, the connection is weaker. But again, you have this tripolar pattern. So positive or normal in the Indian Ocean, in the central Pacific, and negative in the center. So basically, what this tells you is that somehow there is observational evidence that the topical forcing can favor this association of the Pacific Gulf regime and the positive or the opposite phases. And if so, from a practical point of view, if you are a local forecaster, well, you may say, well, I don't care about the other half of the hemisphere. I just care about the Atlantic regimes. But if you are interested in this, for example, the variability of the heat fluxes and how they affect the mean temperature over the continents, northern continents, then the importance of having the same phase in the Atlantic and in the Pacific is that basically, you will get the signal in terms of heat fluxes from the two oceans in phase. So when you get more heat out of the Atlantic, you will also get more heat out of the Pacific. And so these two contributions add up. And so you may have actually a stronger impact on the hemispheric mean temperature, because then you have if the Atlantic and Pacific regimes are totally uncorrelated. OK, maybe five minutes for the last topic, and then we go for coffee. Now, so well, so far so good. Observation is everything seems to fit, but we have seen before, when you try to model these things, it's not so easy. And usually, you don't get the right association. Sometimes the model seems to like more the combination of a ridge in the Pacific and the travel in the Atlantic or vice versa. So why is it difficult to get the right combination? Well, one possibility is that you need to have the right intensity of the anomalies in both the Indian Ocean and the Pacific. So one possible reason is that you don't get the right amplitude. But another reason that has been advocated is that to explain the impact over the Atlantic, you need to look at the effect over the stratosphere. And it just happens that if you just make an EOF analysis of temperature in the lower stratosphere, this is T100 during winter. So you see a very zonal-symmetric leading EOF. Then you compute the covariance with geopotential height at lower levels. And OK, you might expect this is a zonal-symmetric pattern. You will also find a zonal-symmetric pattern. Instead, no. You get something which is a much stronger signal over the Atlantic than over the Pacific. Now, why is that? Well, we will talk maybe a bit after the coffee. But observation in this result is there. So it basically tells us that knowing about what happens in the lower stratospheric polar vortex matters much more for the NAO than for the PNA, or whatever you want to call it. And this was actually, this topic was explained on the seasonal timescale, but a series of paper by the Toronto School, Fletcher, Kushner, and recently one paper also with Christophe Cassaud. And what they did is that they looked at separate impact of SSTs anomalies in the Indian Ocean and in the tropical Pacific. And they actually looked at the impact on the zonal mean flow up to the stratosphere. And by doing an experiment when they put the anomalies separately, what they found is that if you put them together, you get some impact on the zonal mean flow, which is not too large. However, if you do the two anomalies separately, so if you only put the heating in the tropical Indian Ocean, or if you put the heating in the tropical Pacific, you get a much stronger signal, but with opposite sign. So basically, if you put the heating in the Indian Ocean, you have a low at the North Pole, therefore a positive Arctic oscillation. And as we have seen, this will also lead to a positive NAO. If you look at the result of the Pacific, it's the opposite. You put them together and they cancel each other. So I think that is a possible reason why it is so tricky to get the right result over the Atlantic if this theory is true. Because the net result is a combination of two much larger terms. So if you have a relatively small error in one of the terms, then your total result may be quite different. And what they showed is that the result basically comes to the convergence of heat flux by planetary waves in the lower stratosphere and the upper troposphere. So basically, the signal induced by, so these are various experiment. And there is another blue bars are for the heat flux convergence from the tropical Indian Ocean, the red ones from the tropical Pacific, and the green from the combination. And in their experiment, you see a cooling induced by the Indian Ocean, a warming in the polar vortex induced by the Pacific. And in most of their combination is the Pacific that wins. And you get a sort of negative Arctic oscillation response. In fact, the study was mostly focused on the latter part of the winter, where in fact, usually the answer gives you a negative Arctic oscillation. But in the early part of the winter, sometimes the balance is actually reversed. And so basically, it seems that to get this positive and your signal, you must make this blue bar bigger than the red bar. So, yeah, I think that's all what I want to say. Again, there are no certainties, and there are theories. But what I hope that I've done is to show the students that this is a very challenging problem. It has a lot of practical relevance both for forecasting but also for studies of decadent variability and modulation of climate change. It involves the interactions with the ocean. And it's a very tricky problem from a modeling point of view because you have these two heat sources that somehow give signals that partially cancel each other and so somehow the net results over the Atlantic is due to this basically balance between the effects of these two sources. And unless you get the balance right, you will not get the North Atlantic oscillation right. And with that, I think we've all earned a good coffee. Unless there are questions.