 Hello. From the UK Met Office, I'm going to carry on this specific theme that we had yesterday. It's been quite a long week so far so to wake you all up, a hands up who thinks the Pacific Decadal Oscillation is actually ... Can you hear me now? A hands up who thinks the Pacific Decadal Oscillation is actually predictable? Anyone? A hands up who thinks it might be predictable in the future? Gwlad am iawn? Mae'n gelych yn amlwyddo. Felly, ac mae'r acredigau aeroler. Felly, yn fyddiwch ar gwahagol. A lluniau. Dyma. Rwy'n gweithio. Rydyn ni'n fawr, mae'n fwy o bach na'r cerddol cerddol a'r fawr. Mae'n cyflym o'r gweithgiad. Mynd yn y teulu cywrch â gwrs. First I'm going to think about the Pacific Decadal Oscillation as an EOF on temperatures with the global mean temperature removed. So this is the sort of standard observed EOF pattern that we have seen a lot of recently. This is an average of the pattern that I get from the initialised Decadal hindcast from the CMIT V dataset. So the pattern is looking pretty good. As Matt Newman might point out, it could be a PDO-like pattern rather than a natural PDO pattern but it's getting in the same kind of variability going on in the North Pacific and you've got the tropical node down here. As some people have pointed out already, this node does go a little bit too far west but there's something going on in the model that's also possibly going on in the observations. There's been a nice study recently by Dong Et Al doing a much more simple definition for Pacific variability. Here they've just broken it down into two simple box averages. We've got a North Pacific box and you take this away from the tropical box. So I'm going to look at both of these in the models, in the same five models. So now I've got time series. The shaded dark line is the observations. We've got the EOF definition up here and then this box average PDV index definition down here. The red lines are the initialised CMF5 decadal hind casts and the blue lines are a similar set of uninitialised decadal hind casts. Now, for both of these time series I'm actually getting some skill for the PDO. The correlation is looking at about 0.5 so we kind of thought, oh, this is quite interesting. I would like to add a small caveat that I'm going to come back to later on that as Neil pointed out, linearly detrending can cause issues and this is the type of detrending that we've actually used in this so I will come back to this. Initially we kind of thought, oh cool, we've got a correlation of 0.5. Breaking down the box average definition you can kind of have a look at how much influence the North Pacific box is having and how much influence the tropical Pacific box is having. So now we've got these time series for the North Pacific box at the top and the tropical Pacific box at the bottom remembering that we have linearly detrended these. You can see straight away that the North Pacific box is driving the whole of this sort of PDO pattern. We've got a correlation of 0.8 whereas with the tropical Pacific box we've only got a correlation of 0.0. I should have said this is five year means we're looking at kind of long term changes. So all the model skill is coming from the North Pacific so it's not really a proper PDO thing. Now we come back to this point that detrending, linearly detrending can actually cause some issues in our assessment of skill indicator models. So here on the top panel we've got the original plot, the linearly detrended North Pacific region and you've got this nice sort of periodic kind of thing going on. But if you look at just the anomalies we've now got a correlation of just 0.6 and you can see that perhaps although the observations you've still got this kind of nice pattern showing in the model now we're not getting that at all. We're not getting this positive phase here. We're just getting this rising up from around the 1980s. So it looks like we might actually be putting in some artificial scale into our model. This is particularly because we've been removing a linear trend on quite a short period. The hindcasts only go from 1960 onwards and this happens to be mainly looking at the rising part of this sort of PDO type shape. I should point out here as well that the uninitialised hindcasts all have very similar skill. This is quite important that it doesn't feel like on these timescales that initialisation is actually really helping. You can also see a similar problem in the fields of temperature. So now we've got the correlation for five-year mean temperatures over the whole of the globe. The left plot is the linearly detrended plot. The central plot is the sort of raw anomalies and then on the right we've got the linearly detrended minus the anomalies. You can see normally we expect that linearly detrending is going to reduce our skill. We kind of use it as a way of showing how well our models are doing around the global mean trend because it's very easy to get high correlations if everything's warming. It doesn't matter if the model's warming more than the observations, you're still going to get a good correlation. So to try and get an idea of how the model is doing around this trend, we tend to use linearly detrending as a technique to assess this. But you can see that in general this is a fairly okay method. The correlation has decreased over much of the globe but in this particular north Pacific box we're actually seeing an increase in skill. This is partially due to trends in this region over the hindcast period. There's very little trend in the observations whereas there's quite a large, there's still some trend in the model. Also just as a reminder, you can also see in both these plots that there's less skill on decadal timescales in the tropical Pacific which is something that's been pointed out yesterday quite a bit. So detrending, it might be adding artificial skill but we've sort of shown that the initialised hindcast are doing similarly well to the uninitialised hindcast. So now I've extended my analysis to look back over the whole of the transient run historical ensemble set from the CMIT 5. And you can kind of see the problem again that if you linearly detrend on this short time period you're getting this nice U-shape, this nice part of the oscillation. But if you use the whole going back to 1880 of the historical runs and detrend over the whole period you can straight away see that you're not getting this. A positive phase any more because it's just due to detrending over such a short time period that you've added an artificial cooling in the model that isn't really there. Also you can see that the correlation drops so when the initialised runs over this period we've got a correlation of 0.6 whereas once you extend it back all the way to 1880 it's only 0.2. So if we go back to the idea of the PDO index now if we do it on the perhaps more standard method of taking out the global mean temperature rather than linearly detrending this is perhaps a more realistic idea of the skill level with now correlations of just 0.4 and just 0.3 when you use this box average method. So we could say there is some skill but it's a very modest, very kind of low skill and as I've said this is all really coming from the North Pacific box still and it's also coming from the initialised models as much as it is coming from the uninitialised models and I know Paco you've done some work with and Leona paper a couple of years ago that this could be picking up on the second DOF rather than the first DOF in temperature that there is some skill on longer time scales from what they call the North Pacific dry oscillation. So if there is some skill coming from the initialised models then it's kind of we've got a good set of data at the moment from the CMF5 data set to have a little look at which part of the external forcings might be driving some of this pattern. So I've looked quickly at the aerosol only runs and the natural only runs and these are first of all correlation plots for the aerosol only and the natural only. Over the hindcast period this is 5 year means and you can see that both the aerosol and the natural forcings seem to be inducing some sort of correlation in the North Pacific but they're not at all influencing the tropical Pacific and you can also see it in terms of the time series. So here we've got the historical runs going all the way back to 1880. If you look at just the recent hindcast period for the aerosol only runs the correlation is around 0.5, we're getting quite a high correlation in this period and the grey bar is meant to be showing the kind of residual noise that you'd expect just from ensemble averaging for the amount of members you've got in this ensemble and it is coming out. There is slightly more variance in the ensemble means that there could be some real signal in this and similarly for the natural but the correlation is only 0.3 but then once you go back all the way to 1880 this correlation drops to sort of a negligible 0.2 and 0.1. This kind of fits in with a lot of stuff that people have been saying this week including in Doug's talk earlier on today that there seems to be something special going on in the most recent period and aerosols seem to potentially be having an impact because they come into action in around the 1950s whereas if you go back on longer time scales they're not so apparent. So if this is a real result then we need to know more about what aerosols are doing in the recent past and also what they're going to be doing in the future. So in summary models seem to capture the Pacific variability OF patterns very well. Okay maybe this is a PDA like patterns but all the models are sort of showing some signs of getting the right variability. There's moderate skill for the Pacific index but this is amplified artificially if you linearly trend the temperature feels it's important to find another method to look at this and the skill is mainly coming from the North Pacific so it's not a true kind of Pacific decayed oscillation that we're getting in the model it's very much a North Pacific driven variability. There seems to be some signal from external force things and we're guessing that this looks like anthropogenic aerosols could be important but also natural factors and likely volcanoes. So an important future point is to try and get really realistic projections aerosols for the future if we're going to make robust projections for the future. So thank you any questions?