 Thanks Matt good every good morning everyone do something which one is yours Good morning everyone again Today, I'd like to talk about Observed changes in the Walker circulation over the past century and in particular changes in the last few decades and How well see me three and see me five models particularly see me five models are able to simulate the observed changes and Some of the specific issues that I'd like to address this morning are first of all, how has it how has the Walker circulation changed over those periods? Do the models simulate those of the observed trends since 1990 and if they do fantastic if they don't why don't they simulate those trends and We'll find that the models don't seem to be able to capture the observed strengthening of the Walker circulation Since 1980 and so we'll examine some of the reasons why that might be the case and then we'll briefly consider what implications That apparent deficiency might have for the confidence We have in projections of the Walker circulation over the coming century and then fine I'd like to finally finish by Outlining key challenges and opportunities that arise from this study So if you want more details on much of what I'm going to talk about it appeared earlier this year in the Journal of Climate and it was work. I did with Greg Kachuba and also like to acknowledge helpful discussions with Jinjia and Didmar The first thing you need to know to know about the Walker circulation in the observations is that a weakened during the 20th century and Just one illustration of that and there are many publications that have addressed this This is just one illustration where the Walker circulation is Tracked or the strength of the Walker circulation is tracked using the difference in mean sea level pressure Across the Pacific between these two boxes east minus west and then if you take a time series of that difference Calculate the trend over the night over the 20th century you find in the circled Ellipse that you get downward trends weakening trends from 1876 through to 1999 and from 1900 through to 1999 and also from 1958 to 1999 and you find that those first two the two longer trends Well the trends over the two long periods are both statistically significant So there was a very interesting weakening of the walk circulation during the 20th century and You can update this information using data more recent data And so this is a quite a complicated plot. It shows not only the observations there in pink or something close to pink on the left here So that's the trend in that index for the strength of the walk circulation over the period 1980 to 2012 and you can see that it's declined But you can tell by that from those brown rusty coloured whiskers that it wasn't statistically significant So that brown the brown whiskers are an estimate of the 95% confidence interval for the trend and then the other trends are From the models from the CMIT five models and you can see there's quite a mixture some go up some go down most of them go down and If you look at the multi-model mean of all of those results on the far right here you get a Negative weakening and that weakening is statistically significant. So the situation seems to be quite comforting that there's An externally forced weakening of the walk circulation over this period But there's also a huge amount of natural variability and so there does seem to be a good Consistency between what the models are saying should have happened an externally forced weakening Reinforced by internal natural variability and you get lots of variations because of that if that is what's actually going on among the models just as you would expect so After you looked at this you might And then we also had a previous study looking at the 20th century Which in which we concluded and others have concluded like gay becky and his colleagues that both external forcing and internal variability are needed to account for that observed weakening over the 20th century and the results that we've just looked at for the Longer period up to 2012 are consistent with those earlier conclusions and in this earlier paper We concluded that there was a high uncertainty about the relative importance of those two factors internal variability external forcing but we Estimated that the external Forcing account for something like 30 to 70 percent With internal natural variability making up the rest So the models and the observations seem consistent. We have a nice Physical framework for explaining what's happening the observed trend driven seems to be driven by a combination of external forcing and Internal variability and so the world seems very simple and climate science seems so easy So it's time to move on to something else but Let's now turn to what's happened in recent more recent times and specifically let's look at trends over the period 1980 to 2012 and Again the frameworks the same so this on the Left of screen is the observed Increase so there's been an increase in the strength of the walker circulation over this period in the observations This time the whiskers are above zero So it's been a statistically significant increase in the strength of the walker circulation over this period But then when you look at the models none of the models show statistically significant change and About half the model show a statistically sorry half model show an increase Around about half model show a decrease and if you look at the multi-model mean for this period 33-year period you get a negative number, but it's not statistically significant So it does seem to be a degree of inconsistency between what the observations did and what the models say should have happened Because you notice that see this rate this green line here is the The value of the observed change and you see that none of the models Give you an observed change as large or larger than the observed change And in fact none of the models give you a decrease as large or larger in magnitude as the observed change depicted here by the Dashed line so there does seem to be an inconsistency I mean one way of reconcile this inconsistency is to say that there's been this very large very large Internal natural variation so that would give you consistency, but maybe that's not what's happening Maybe there's some deficiency in the models that it's making that's causing primarily causing this inconsistency You can't rule out the possibility that it's very large internal variability on based on this But you could park that candidate to one side and explore other candidates and that can't other can't the other candidate And that other candidate is there's something wrong with the model or something something wrong with the way the models are forced So let's explore that second option So just to summarize those results. We've seen and observed statistically significant increase in the strength of the walk circulation over this period none of the C-MIP five models exhibit a trend this large The C-MIP five models are roughly evenly split between increases and decreases and none of the models exhibit statistically significant trends over this period And the multi-model mean is is negative, but it's weak and it's not statistically significant And by the way, we've used this index to track the strength of the walk of circulation, but there's lots of Complementary supporting evidence to indicate that there has been a strengthening of the walk of circulation in recent times And here are some of the studies that have looked at this issue most of them on shorter time scales But they've looked at a variety of different variables SST wind for example so Based on the plot that we showed before there might be a problem and So in order to so what how can we reconcile what's happened in the observations which with what is supposed to have happened in the simulations So we've already discussed the possibility that it arises from unusually large and unusually large internal variation But I think a question is well how unusual must this internal variation have to be to account for the obs for the Results, so we've already pointed out that that observed trend was statistically significant So in that sense the trend is unusual in terms of the in terms of nature's own Variability, but another question you can ask is how unusual is that trend in magnitude relative to the variability that the models exhibit? And in order to address that question what we did was to analyze a very large number of pre-industrial runs and we looked at That same variable in those runs and we had over if you accumulate all of the information you get over 17,000 years of output And so we calculated all possible 33 year trends in that data and we found that Out of that 17,000 years you ended up getting 11 events There were 11 33 year trends that were as large or larger than the observed one and there were 15 There were as large in magnitude, but they're of opposite sign and if you add all that up work it out It's roughly roughly you get one and a half of those events every thousand years So yes according to the models you can get the observed trend just from internal natural variability, but it's a rare event so maybe one other possible explanation for What might give rise to this apparent inconsistency is maybe the models are underestimating the level of natural variability in this particular variable And so in order to Test that idea what we did was to take the time series of this mean sea level pressure difference along the equator And we calculated the standard deviation of the variance of the variability both in the historical runs and in the pre-industrial runs and The observed value is marked by the red dashed line. You can see the observations are up here and The the y-axis is a mystery variable which will come to in a moment So this is the standard deviation of the variability in that index that we've been talking about and you can see that for most of the models Exhibit variability that is larger than the observed variability So this is one of the world's most short-lived hypotheses that we we hypothesize that the variability in the models were too weak But the very first plot we show is that actually it's the reverse the models tend to be too vigorous But this shows the total variability What's more relevant to what we're talking about is the variability on decadal time scales So this is the same thing, but prior to calculating the standard deviation We calculate we run our 13 year running average through it And then we calculate the standard deviation of those pre-industrial runs in the historical runs And what you find in that case is that the vast majority of models Exhibit variability that is smaller than the observed estimate So if for some some reason if you look at all of the variability the models are too Energetic whereas if you just look at the decadal the models seem to be too weak Now how on earth can that be the case if the variability was just a white noise process? This wouldn't make sense I mean one way you get a difference between the left and right panel is through just sampling error But you know the vast majority of models are showing this behavior vigorous by hate to vigorous behavior on it for total variability Not enough on decadal well one way or the only way of reconciling this is to look at the details of the auto correlation function so that there's a strong connection between the right relationship between the standard deviation on the right hand panel to the standard deviation on the left hand panel To the properties of the auto correlation function and so we finally get to this mystery variable Which is the lag one auto correlation so you can see that the observed value is Up here point three and you see that the majority of models Exhibit too little persistence and in fact there's a large chunk of model a large number of models Exhibit a lag one auto correlation value, which is negative So what that means is even if the internal if the inter-annual variability is large It's going to make it harder for those models to generate multi-year decadal Variability and indeed it's going to make it harder for those models to generate 33 year trends in a multi-decadal trends That's just a one. We also looked at a two and lo and behold. We also find there's a tends to be deficiencies in a two as well in the observations the lag to auto correlation value is pretty close to zero minus point oh nine whereas the multi-model mean value is minus point three and It turns out there is that you can go to textbooks and work out what the spectrum looks like for an AR2 process and if you plug in the numbers and you put in this bias you've lo and behold you find that The weighting or the spectra at decadal and longer time scales will tend to be too weak because of this particular defect So you've got the problems at a one with a one you got the problems with a two So you've got the problems with the inter-annual variability Giving rise to problems in the ability of the models to simulate decadal Multi-decadal trends, so you might think that you can just put problems with ENSO to one side when you start talk about 33 year Trends, but this illustrates that you can't necessarily do that So yes, the answer is model decadal variability does seem to be too weak And this will certainly contribute to the apparent inconsistency that we showed in one of the earlier plots But of course There are other possible candidates that might also be contributing to the inconsistency We've already talked about that it does seem to have been a large natural Occuring internal variation in the observations that's going to help to produce apparent inconsistency, but of course you could also have Problems associated with forcing being omitted or misrepresented in the models or There might be a problem with the model response to the forcing in post And I think the point is that we've learned something about and we've established that it does seem to be an apparent inconsistency You can reconcile it with if you suppose that the internal variation is very very large the models seem to under do the Natural internal variability which which would promote inconsistency, but we don't fully understand the cause of that inconsistency And in particular we don't fully know the answer of the magnitude of these effects that I've got here in the black dot points All their relative importance in this issue So thanks you so now we turn to that's all looking at the past What happens? What are the models project will happen in the future? Well, there's been for several generations of models a Consensus among the models that there'll be a weakening of the walker circulation in response to business as usual increases in greenhouse gas concentrations over the 21st century and the C-Mip five models are no exception So this shows the trends over the 21st century in the strength of that index the strength of the walker circulation And you can see in the majority of models. It's projected to decline and the multimodal mean value again is on the left It's down here, and there's the value that runs along there, and it's statistically significant and it's It's the same very similar results. You get the same or very similar results using C-Mip three models so this agreement this tendency for agreement among models Both within C-Mip five and C-Mip five relative to C-Mip three increases confidence that the walker circulation is going to Decline in strength over the coming century. However Our confidence in those projections on the other hand is weakened by the fact that we aren't we can't fully account for this apparent inconsistency in the in between the observed and model trends in the last few decades So we need to temper our enthusiasm for for for the projections that are taken from C-Mip five models and in particular to To outstanding issues out do the models Overestimate the magnitude of the externally forced weakening or if you want it to be even bolder and more Want to generate a bit of discussion you might even say How confident should we be that there is actually going to be a weakening in response to the external forcing? Can we dismiss the possibility that the models or multimodal mean projections that actually has the wrong sign? and so The final slide is to present just some of the challenges and opportunities that arise from this study and It really stemmed the challenge really and the opportunity stems from the fact that the walker circulation is really one of the world's most important atmospheric wind systems and it's exhibit this very marked strengthening over a fairly long period over 30 years and So this can constitutes a very major event in the recent history of the earth's climate system and yet We don't fully understand why that's the case nor do we fully understand why the models and the observations seem to be inconsistent maybe the The explanation that I provided earlier in the talk is the full explanation, but we don't know for sure And so a challenge and opportunity is to redress those two issues and to come to a better understanding of those two points and to determine the relative importance of all the other all the plausible candidates for what might be causing that inconsistency and Hopefully that would by investigating that issue more carefully That could prove to be a route to provide significant advances in our understanding of the civic climate variability and climate change. Thank you