 Hello? Good morning. Can you hear me? Yes. Okay, so good morning. So today I give a different kind of lecture. Yesterday was more about foundations and general principles, and today is more about recent results. And so I will actually show you some results from recent experiments about sub-seasonal variability and interconnections that we have performed at ECMWF, and together with this colleague of mine, Sarah Jane Locke, is Robert Reddick, and Sennemann Reddick guitar. And some of them are experiments really on the sub-seasonal timescale, monthly, and some more predictions. And some of them are long runs, multi-decadal runs, with both just our atmospheric system, by our service team, or with our coupled system. And these are experiments that we have run for a new funded project called Primavera, coordinated by the UKMET Office and University of Reading. And it involves making long simulations for the historical period and the future up to 2050. ECMWF will only do the historical part. So, let's start from the sub-seasonal experiment where you've already seen a diagram like this many times. But, in fact, I will come back to show you this particular version of the Wheeler-Hendron phase diagram because it just describes one case of very strong Madeleine-Judian oscillation that occurred during the winter of 2007-2008. So, you see here in a very, very large amplitude of the EMJ cycle, especially in this part of the diagram, phase two and three, where the convection is located over the Indian Ocean. And so, basically, these experiments were considered with one particular aspect. And that is, you know, we always say, well, it's been said a number of times during the school, we have this signal that propagates into the extra tropics with a time scale of, say, 10 days. Most people would say the largest connection of Atlantic oscillation occurs 10 days after phase three. However, the EMJ-O is a propagating phenomenon. And therefore, because of that, when you do lag correlation, there's always some sort of ambiguity and uncertainty. For example, you could say, well, let's see what happens 15 days after phase two when the convection is in western Indian Ocean. And if you do that, and you make a composite of all the geopotential height anomalies at 500 kPa, 15 days after phase two, then you get a map like this with this wave number two that we have already seen and discussed yesterday. And you see a positive North Atlantic oscillation signal. Now, if you do the composite 10 days after phase three, you get a similar wave number two, perhaps this center might be a bit less defined, but now the southern part of the North Atlantic oscillation is stronger. So actually, you can put them together, and then you get a sort of smoother interconnection pattern, which, however, is not very different from any of the two. So the question is, where is the signal actually starting from, and how long does it take? Or maybe this is what we could say is that what we see in the North Atlantic is actually a combination of signals that come from different parts of the Indian Ocean but have different propagation time, and then they end up reinforcing each other. So these were the questions that we are trying to address. And of course, they have been addressed in the past. And one simple way to do that, and that is similar to what you are doing with speedy, when you fix the location of the forcing, is in fact to impose a diabetic heating in one part of the tropics in a general circulation model of the atmosphere and then see what is the response and how long it takes for the response to be established. And one paper which is often quoted on this is the one from Hylene Giverboune and a collaborator there published in 2010 where they actually took a dry primitive equation model with prescribed heating, and they added on top of the climatological heating some anomalous heating that somehow was mimicking the structure of the first two EOS of Euler variability. And they didn't get a lot of response from the first pattern, the monopole over the Maritime continent, but they got a pretty strong response to dipole here and the maps at the bottom are in fact the response in geopotential height at 500 after six to ten days after the start of the forcing and then you see that there you... Yeah, I hope you see the coastlines here in North America so they get some strong response in the Pacific, let's say from the six to ten and then after another five days, so 11 to 15, then they also get this response in the Atlantic that projects on to the North Atlantic oscillation. Contrary to that, there are some causes and reasons and I think it also seems that we get a huge response in the Pacific which I think is kind of intriguingly of the expected, but there is another one if you scale it so it's just one... Yeah, no, in fact, I'm not saying that this is... I'm not saying that this is a... a perfect explanation of what's happening, but it's a paper which is often quoted and it has a relatively simple methodology and also what they have done and it's quite nice, they've actually made a number of experiments again with the dipole structure but moving the longitude of the centers and then they look whether the signal actually basically projected on this pattern in a positive way or a negative way so in that particular experiment the center of the heating is about 90 degrees so basically this graph tells you that wherever you put the center of the heating west of 100 degrees longitude you will get a projection with this sign when you move it to the east a design of the response will change and although as Fred pointed out the details of this response maybe are slightly different from reality the general features have been reproduced by many others One issue with this kind of technique is that in the MJO case the heating is not fixed it's actually propagating and so other studies have actually imposed the propagating heating and it's a very nice study by David and collaborators published recently in 2015 that a forcing that was actually diagnosed from actual data was imposed as a moving forcing and they studied the exotropical response so somehow when you are using speedy with this propagating forcing you are somehow doing something along this line now I showed you this graph before somehow these teleconnections are very similar to the ones that we also found on the seasonal time scale so if you actually look at the connection from a dipole in convection between the western Indian Ocean and the maritime continents then again you see this wave number two pattern with this low extension here north of the Caspian Sea so again this is a confirmation that if you have a heating dipole in that region that will produce this pattern one difference between this pattern and the previous one if you notice is the sign of the signal in the Pacific and that in fact that depends very much on the location because if you move this dipole too far to the east then the negative part will actually go into the central Pacific and then you will get a different kind of response a bit like the linear response that will give you a high over the Pacific so the sign of the Pacific response may also depend crucially on where the dipole is actually located I see that with the largest signal I see in the central Pacific so why do you claim that the signal is coming from the Indian Ocean? There's another figure in the paper where I removed the ENSO signal if you do it on the seasonal time scale you would... I don't have it here because I didn't want to put too many slides but if you autogonalize this signal so you can take an index like this you take the Nino 3.4 SSD and you autogonalize the signal so you remove the component then this blob goes away and my assumption is that you remove the from the Pacific you get a reduced amplitude and in the Pacific you get a reduced amplitude however it doesn't go away if you locate the convection enough to the west it still gives you a negative signal over the Pacific but yeah, the strength of this is actually related to this part as well it's not only... no, these are... this is an era interim geopotential height and GPCP these are observations so I wanted to try and use this simple approach also to look at different timescales effectively you know the sub-seasonal that we are concerned today with and the decadal timescale now for the decadal timescale I don't have time to talk about it here but this long link will give you access to a similar presentation I gave at the ECNWR seminar in September that also covers the decadal part so... we try to make experiments similar to the one of Haydn or the one you are doing but with a slightly different approach and the reason is that we didn't want to explicitly impose one particular structure so as we discussed then the response may depend exactly on what latitude and maybe the vertical structure of the heating whether you move it or you keep it fixed so what we did is that we choose one case of strong MJO and this is the one I showed you before so this one in winter 2007 and actually so we started ensemble experiments with the ECNWR ensemble prediction system started on the 10th of December so we are in this point in the MJO cycle and in the ECNWR ensemble we have what is called a stochastic physics that's a sort of abbreviated term basically so we put stochastic perturbation to the tendencies that are produced by the physical parameterizations this scheme is referred to as SPPT which stands for stochastic parameterization of physical tendencies and what it does is that it takes the integrated the total results of all the parameterizations and then it just multiplies these tendencies by a number which is one plus a small random number now in the operational ensemble this approach is applied all over the globe and in addition we put initial perturbations in the ensemble in this particular experiment we didn't put any perturbation in the initial condition but we did basically a control experiment where the stochastic physics is applied over the globe as in the operational ensemble so this will produce some spread of course smaller at the beginning but then we will go over the whole globe and then we looked at two particular regions and the first one is exactly the one that we have discussed before and the second one and you will understand why is this region covering South America and the tropical Atlantic so basically what we want to see is whether if we just originate some spread in the ensemble in this region or in this region would that be enough to modify substantially the variability in the North Atlantic and so somehow the reason why we have chosen a strong MGO case is that this forcing is stochastic so it doesn't have a preferred structure however so what we hope is that since this was a case when the convection was organizing itself this stochastic perturbation would somehow project on the existing MGO without making either stronger or weaker and so this is what happened so this is what happens in the experiment where you have this stochastic perturbation supplied everywhere so we computed an index of the North Atlantic oscillation just the difference between geopotential height in the region of Iceland and Lisbon and if you now do a regression against this index among the 51 members of the ensemble so in this case the variability is just it's not a variability in time it's a variability among the 51 members of the ensemble each of them has a different perturbation then you can see that there's quite a strong variability here in the North Atlantic you cannot see the maximum but it goes over 120 meters over Iceland so very strong in your variability and then I looked at precipitation along the equator so average between 10 North and 10 South so this map is actually an average for days 29 to 23 so basically it's a five day mean center on day 21 and then I looked at five day means of precipitation backwards and this red box just tells you that this stochastic physics is applied over the whole range of latitudes and so basically you look backwards at what is actually happened and if you look at here at the beginning of the experiment then you can see an increase in rainfall west of 90 degrees east and a decrease in rainfall to the east so somehow it tells you that you can actually trace some signal propagating to the east but you also see some signal here over the South America and North Atlantic region so this picture is a bit mixed but basically tells you that so you have some signal at the beginning of the experiment into region here the Indian Ocean and West Pacific and here the Atlantic somehow the signal propagates into the extra tropics and then you get this annual however because the perturbations are applied everywhere you cannot be absolutely sure that the spread in DNA was only caused by this perturbation however then we can look at what happened when we only put the perturbation in the Indian Ocean and the maritime continent and then somehow of course you do the same of molar diagram and so this is the area where the perturbations are put so you see that at the beginning the rainfall anomaly is only present in this area as you would expect again very large spread in the North Atlantic oscillation around the A20 and here now you can clearly you see more clearly the signal in precipitation propagating from the Western Indian Ocean to the east you see this green line moving from the beginning of the experiments up to the 20 so in this case we know that the source of the spread is only in that region there's no other source of spread in this ensemble apart from the changes in convection which are triggered by the stochastic physics now we can say how long does it take for this spread to be equivalent to the spread we have when we put perturbations everywhere so for example if you now compute the spread in the NaO index this is how the spread increases with time when we put perturbations everywhere of course you have perturbations already in the extra tropics so the NaO spread will go quite fast and after about 10 days it will almost reach its maximum value if you put just the perturbations in the tropics in the Indian Ocean and in Indonesia of course in the beginning there's hardly any spread in the NaO because the signal didn't have time to propagate but after you see after about 20 days the spread in the NaO is basically the same so this tells you that you put a perturbation in this case impose any structure we just perturb stochastically and let the model organize itself and basically after after 20 days you have a spread in the NaO as large as in the experiment where we had perturbations everywhere so this somehow is an indication that this is really a crucial region for the variability of the North Atlantic Oscillation but of course this is mainly showing up in the sub-seasonal range because of course in the first 10 days or so the signal is still somehow traveling we also put perturbations in the Eastern Pacific the problem of this experiment is that because it's random eventually it doesn't matter where you put perturbations if you wait long enough the spread of the NaO will be yes so it's just the middle of the time scale well we put perturbations in the Atlantic that's the second region and that is what you get so even after 20 days you don't get the same spread yeah you get some spread but if you actually look at the amplitude of what happens if we just put the perturbations here then you get something that again projects on the NaO but it's probably only 60% of what you get so yeah it's still that if you wait long enough everything will reach the same spread but here you have to go up to the 30 to have the same spread so if you go at the 20 which is around here in the other case the blue line was overlapping with the green one here we are about so of course in the long run any perturbation will spread to the whole globe but somehow this shows that in fact the Indian Ocean and West Pacific is a particularly effective source of perturbations for the North Atlantic oscillation now of course this is not a general methodology probably will not work in all the cases I think it worked because this was a case of strong MJO so this stochastic forcing actually was amplified by the strong organized convection and we'll have to repeat we want to repeat this experiment in other for other initial states to see what happens now and another way of seeing this is to instead start from the precipitation and to do the covariance with some leg that's the traditional approach for example if you do it using GPCP and Eri-interim in fact you put a dipole exactly when same region as we use for the seasonal forecast and after two weeks you'll see again number two in the northern extra tropics now if you do it with the experiment in fact with these perturbations only in the Indian Ocean then of course if you do the covariance across the ensemble member and like zero you hardly see anything because the signal has not yet reached the extra tropics then you start seeing something after one week again wave number two and this wave number two then becomes fully developed after two weeks now let me move to the second part so this is basically how we can trigger an AO variability in our sub-seasonal forecast now when you do sub-seasonal forecast of course we start initial conditions so the model still basically starts from the observed attractor but of course model have biases and they asymptote to their own climate and so we decided to take part into this EU funded project Primavera which is basically the goal is to diagnose the behavior of high resolution couple models and do basically help improvement in different aspects of this model this is a project that involves all the major modeling groups in Europe and will contribute to the so-called high res MIP part of CMIP6 so high res MIP is basically it's consistent set of simulations for the period 1950-2050 so this will be 100 year runs and you will have a control experiment with 1950 forcing you will have couple experiment and for the historical part you will have a MIP experiment now ECMWF is only doing the historical part so we are not doing the extension so our experiments are performed up to from 1950-2014 and since the purpose of the project is to explore the impact of resolution every institution runs these simulations with two resolutions in this case basically we are running with two versions of our couple model one with a sort of grid point resolution of 50 kilometers and one with a grid point resolution of 25 kilometers in one case the ocean model has a one degree resolution in the other case it's a quarter resolution the vertical resolution is identical 91 levels for the atmosphere 75 levels in the ocean and we use the link to CIS model in both cases so the first thing this was a new kind of experiment for us basically running our model for very long time to see what is actually the asymptotic climate of the model so I think one of the first thing one to see when you run a couple system is whether it's able to maintain and so variability and this was a nice surprise for us that the model actually does so these are actually tiny series of the linear 3.4 SST anomalies on the left you see what you get in the AME of course in the AME you prescribe the SST so these are actually had high SST too so in practice these are the observations of SST and these are the time series for four ensemble members run with the coupled high resolution run and with our nice surprise the system is able to maintain and so amplitude you can see that the things look a bit different between different ensemble members for example in this one you have pretty large amplitude with quite regular period but if you look at the other experiments then the variation is more regular perhaps closer to the observation and there will be each group will contribute one ensemble member that is the requirement this will be put in this CMOR format that is common to all CMIP6 experiment in this particular case you will get this ensemble member for it probably is good because I think it's somewhat in the middle between the other ensemble members so that's a nice experiment and not only you can do the covariance of SST in the top or rainfall at the bottom with the Nino in this case will be the Nino 4 SST anomaly because we wanted also to look at precipitation that peaks in the Nino 4 and you are on the left the covariance is computed from ERA interim and GPCP in the center the covariance is computed by Amy and here on the right the covariance is computed from the couple model and if you compare this covariance with the observance these are pretty good one aspect I'm actually coming back is that if you look at the covariance between the rainfall here and the rainfall in the Indian Ocean you see a pretty small signal in the observation so this tells you that the rainfall here and the rainfall here they have a positive correlation but it's relatively weak it gets a bit stronger in the Amy run and it gets even stronger when the rainfall runs and I will show you what effect this has on the teleconnections so these are now the teleconnections from the western Indian Ocean and the Nino 4 region the precipitation in this region on the geopotential height at 500 and these are actually seasonal means in this case we first look at the seasonal means top left what you get from the observation number two we have seen the number of times why so this is from the western Indian Ocean and you see if you only do the western Indian Ocean then you get this low over the north Pacific but it's much reduced and farther to the north so the big signal that you saw in the previous version is also because I included in the index the sinking over the maritime continent but the part over the north Atlantic oscillation you can get it just if you consider the rainfall over the western Indian Ocean if you look at the covariance with the Nino 4 precipitation then you get the traditional Nino teleconnection big signal here over the central and east Pacific the signal over the Atlantic where you have a big low roughly in the middle actually you can say that maybe a slight negative projection onto the north Atlantic oscillation because you have a positive anomaly here over Iceland but if you look for example over Lisbon it's around the zero line so it's a very weak with negative correlation now let's look at what the model does well there's a wave number 2 pattern in geopotential height not as clear not as strong as in the observation especially the southern part is much weaker but overall you could say it's not too bad if you look at the Nino 1 that also looks pretty good in the Pacific but now you see that there's a shift to the north of this of this particular pattern so now you tend to get perhaps a slightly positive connection with the Nino we still argue that the relevant flow pattern in Europe is still very similar only that one projection one the Nino the other a little bit less the flow in the Atlantic and in Europe is more or less similar now if you now look at the couple system then perhaps the yeah this particular interconnection here in the Atlantic it perhaps gets more similar now the position of the low is in this case it's actually farther to the south so in this case the projection is more clearly negative so look what happens from the Indian Ocean you hardly have any signal and it's interesting to see how that depends on the relationship between these two regions so we've seen in the observations they are not very much correlated so I've plotted here for the four ensemble measures the correlation of the Western Indian Ocean rainfall with the Nino 3.4 red dot the observation gives you 0.17 so the red line should be at the same level as the red number but actually these are the and these are the couple runs you see all the points are much higher now the second thing that you can plot is the correlation these are the green values of the Western Indian Ocean with the NAO index and if you look at the amy the green line the green dots are a bit below the 0.34 values that you get from the observation but they are not very far apart consistent with the map but if you actually look at the couple runs well they are very low or even in one case even negative now interestingly let's look at what happens to the correlation between the Nino 3.4 SSD and the NAO index we have seen that in the reality it's almost zero in fact minus 0. 0.05 so yeah negative but really very very small and more or less the amy does it right but as we have seen from the maps the couple models tends to do actually a stronger negative correlation so because these two regions actually spend to send opposite or at least orthogonal signal in the North Atlantic what happens in our couple system is that actually first of all from the Nino region we get a slightly negative NAO and then not only but the Nino region strongly correlated now it's rather strongly correlated with the western Indian ocean so somehow you correlate two regions then send an opposite signal on the North Atlantic and this signal basically cancel each other so this is what happens on the seasonal time scale then I try to look at the at the sub-seasonal time scale and I did a similar analysis as I showed you before in this case I used OLR just for a technical issue in the observations you get basically the same result this is actually a slide the longer period than I showed you before I used the full Iain Terim period 1980 to 2013 again covariance with the dipole in OLR in this case convection means negative OLR anomaly the subsidence is associated with positive OLR anomaly over the maritime continent and after two weeks again you see this in this case it's more connected but ok you have lows over this region after the varying straight lows negative over Iceland, positive over the Mediterranean so clearly positive NAO signal now in the EMI plan well again you get the wave number two and this negative extension towards the Caspian sea if you do it from the couple you would hardly see anything however as I said sometimes this correlation may depend critically on the location of the dipole so what I did is that this is probably more phase two of the NAO what happens if I shift the dipole to the east to be closer to phase three and in fact if you do that it's not that you get a fantastic signal but at least you start getting some projection onto a positive NAO now you can ask yourself why does it happen well one reason is maybe on the seasonal time scale we have seen that this stronger control of the ENSO region but in this case the ENSO signal is actually removed so all the signals are orthogonalized to ENSO so there may be another reason for that and what we actually found out is that we wanted to see how the OLR signal was actually propagating in this experiment so again we start from an OLR signal roughly looking like a phase two of the NAO and we want to see what happens after two weeks so then you would expect this signal to propagate to the east so this is what happens if you do it with the era interim OLR so you start from a phase two and in fact after two weeks now the convection has moved into the west sorry so this is the dry area so the dry area has moved into the west pacific and the negative OLR anomaly which represents convection has also moved from the western Indian Ocean to the eastern Indian Ocean so maybe something like a phase three or a phase four no no no in this case now these are sorry these are now weekly means these are now weekly means sorry weekly means question yeah and so you compute this index from a certain week and you correlate this with the week the anomaly is one week after or two weeks after so so basically let's so this is what you get after two weeks if you do it from the reanalysis this is what you get if you look at the amic and if you look at the amic you see a signal which is as strong as the observation but you can hardly see any propagation the signal is almost standing so there's a bit of propagation but very low and that's probably the effect of the imposition of prescribed SSD and actually you don't allow the feedback with the ocean which are known to be contributing to the propagation of the mother and junior oscillation so basically the covariance map with the signal after one week sorry no at week zero and then after two weeks shows very little sign of propagation so the signal remains standing or it's weakly propagating and you can read this nice paper by Pian Kajarov and David where they actually show that actually this low episode are actually the ones that give the stronger teleconnections in the North Atlantic so somehow it seems that also our amic brands tend to keep the forcing in place for a longer time and this is probably again another factor that makes the teleconnection stronger they have the variance the problem is that we haven't yet we have to do actually the lag correlation to see what is the speed of propagation this would suggest that the speed is extremely low now if you look at the coupled model maybe you can say that now the phase is the right one but look how weak is the signal so this would actually suggest that in our coupled system there's not really a coherent propagation because although the variance if you do these things in other position you get the same result the variance is very comparable with the observation but the lag so when you do a correlation at zero lag but when you look at what happens after two weeks then you almost lose any signal so this seems to indicate that the propagation is much less coherent in the model than it is in the observation therefore while in the EMI PRAN you have this signal that remains in the same place for quite a long time in the couple run there's actually very little coherence in the propagation of this anomaly and so somehow the signal probably interfere with each other in a destructive way it's not bad in the sense that if you look at the location of the anomalies is roughly the right one if you look at the amplitude because this is actually a nonlinear scale you probably have something like half of the amplitude so for example here you have a contour of three in the same area you have a contour of six and again here this three this contour of three get this signal here you have a contour of one so in this case so it seems that overall the average propagation speed is probably the right one the fact that you lose so much signal to me it indicates that in the propagation it's not that everything dies out because if I move if I move these boxes the variance at like zero is still high but it's when you do this lag correlation then you lose a lot of signal so to me this indicates that on average the system has the right propagation but with a lot of spread so you don't have a coherent movement of this nice convection in the couple anyway these are really just very new results that have just been delivered and therefore hopefully you will see many more results from the primavera rounds in the future so we will have the other two years to analyze these ones so a brief summary so we have seen that we can actually basically reproduce the propagation of the signal from the western Indian Ocean and the maritime continents in sub-seasonal experiment just by basically adding stochastic perturbations to the physics and we have seen that in fact the Indian Ocean and the maritime continent are very effective in generating the spread propagation time is two weeks to ten days so it's very relevant in terms of time scale perhaps less so for the early-medium range we have looked now for the first time in very long runs with our couple system and in parallel with amyp system so there is a mixture of good news and bad news good news is that we have a realistic simulation of answer variability and the teleconnections but we have problems reproducing the teleconnections from the Indian Ocean on both the seasonal and the sub-seasonal time scales and there are two reasons one is a stronger dependence of Indian Ocean rainfall on the ENSO SSDs and the other one it seems that in our couple system the propagation speed is less well defined somehow than in observation in the observation you get a more coherent propagation in the couple model on average it's right but probably with a lot of variability and therefore the average signal gets weakened so so it seems that there are a lot of ingredients that you have to get right to get this the connection you need to have the convection in the right place you need to have the right amplitude of course but you also need to have the right propagation and the right connection with ENSO so it's not a very simple there's not a very simple recipe to get these teleconnections and a number of things have to get right and so when you see for example a fairing with our computer maps of teleconnections for many S2S models and you might be a bit depressing in saying that all the models get a weaker teleconnection than in the reality but as you have seen you have to get a lot of things right to get this teleconnection so still a lot of work to do on the modeling side so thank you