 Okay, thank you very much. I'm going to talk a little bit about some of the aspects of decadal variability that are associated with the forcing functions that we're using. And because of the lead for questions, I should go quickly. So where have you got sources of decadal predictability? Okay, so if you think about decades in terms of multiple decades, the ongoing response to greenhouse gases is basically that gives you the prediction, that's a very large signal, and it's large compared to the noise. There are short-term variations in force things that are associated with volcanic responses. If another pinot tubo went off tomorrow, we would be able to predict very well what would be the evolution of global mean temperatures for the next two, three to four years. We've done that before we did it in 1992, and we are ready to do it again should such an event occur, when such an event occurs. We can also think about solar cycle variability. It's a solar cycle. It's quasi-11 years. I'll talk about some of the issues associated with that in a second. And there are short-term variations in aerosols, which are not generally caught up in the emissions data sets we're using in the models that we run. Some of those things are associated with legislation, clean air legislation in Europe and in the U.S., and accelerated economic growth and uncontrolled emissions in places like China. Those things are things that we could better predict, and I think there are some things that we could do much better there. And of course, the other source of decadal predictability that we've heard Vent mentioned already today is the initialization of ocean modes. If we can get the ocean set up in a way that we can predict what the ocean weather would be, there may be some ways of saying something about the oncoming climate after that. And I'll give a couple of examples of how that is not working very well. So as you're aware, IPCC made quite a big deal of this supposed discrepancy between the models and the observations in the last 10 years. So one of the issues that we have to deal with in terms of decadal variability is trying to understand what's been going on there. One of the reasons, in fact, there are many, many reasons. And a lot of the papers that have come out about it have been very tied to one explanation. It's this, it's the Pacific, it's that. You really can't think of it that way because the error bars that you have are such that it could be any number of a combination of things in any particular proportion and you're going to see something very similar. This is an assessment that Kate Marvell and I put together. The CMIT 5 ensemble for that period is, the range of trends is over there. Let me see if I can show you that. Okay, it's over here. And you can see that the range of trends doesn't seem to really correspond to the range of what we have from the observations. But the observations are not all one thing. There's uncertainty in the observations and there are perhaps systematic issues with the observations and the comparison that make this problematic. But putting that aside, what you can do is you can try and break down what's going on. Okay, so if you know what the forcing is and you know what the variability is and you know what the transient climate response is based on some median, this is the prediction that you would make for that period. Okay, now for instance if that doesn't include the right amount of inter-annual variability, then you could get something that was consistent just by increasing the amount of internal variability by a little bit. If you include the uncertainty in the transient climate response, again everything would line up nicely. If you include some kind of weird coupling between the modes of variability in the forcing, for instance greenhouse gases affecting the variance in ENSO, again you can make that line up nicely. And here's one of my favorite things is that if you update the forcing estimates to be more accurately responding to what actually happened as opposed to what we thought was going to happen, you can actually get a very, very clear connection between the ensemble and the observations. So it doesn't really leave you very much to explain. If you think about all the different forcings that you have, here are the greenhouse gases, here's the solar cycle, here are the big volcanoes. You'll notice that here the volcanoes go to zero. That's a mistake, not in this figure, but in our experimental design. And the solar forcing you can't really see on here is not very well predicted either. So this is the prediction for the solar cycle in 2008. At the bottom of the last solar minimum, this is the sunspot cycle, these are the predictions that people had for what the next solar cycle would be. In CMIT 5 what we did was we just said, well it's going to be exactly the same as that one. And so we just repeated that. But what really happened was this. So that's very different. The solar cycle turns out not to be very predictable despite the fact that it has cycle in the name. I know, sneaky, right? Okay, so do we really understand enough of the dynamics of the solar dynamo to be able to predict the solar cycle? Not quite yet. However, we can ask other questions. We can ask questions like, well, given the solar cycle, are there aspects of decadal variability that go with the solar cycle that can be predicted? And I'll go into that in just a second. If you take updates to the CMIT 5 forcings, including changes to the volcano estimates, so this is in CMIT 5, we had an overestimate of how big Pinatubo was and we had an underestimate of what the volcanoes were post-2000. Okay, we had an underestimate of what the solar variability was going to do because we got that solar cycle wrong and we've made an estimate of what the aerosol forcings would be, mainly because of the increase in nitrates and secondary organic aerosols which weren't often included in the CMIT 5 ensemble. If you add that all up, you get a forcing error in the CMIT 5 runs of about 0.2 watts per meter squared by the end of that decade. When you put that into the ensemble and you make an estimate of how the ensemble would have responded to those changes, you get, instead of the gray bars, which is just basically the 5 to 95% envelope of the CMIT 5 ensemble, you would get a change assuming that things are roughly linear that follows these dotted lines. Okay, so these dotted lines are what the ensemble might have looked like if we'd updated the forcings. And you can see that these numbers, the colored lines are observational data sets for surface air temperature and there's a slight issue here with the fact that these are SST plus SAT over land and this is not quite that. And that gives you about another 0.05 degrees pulling down of the ensemble if you'd done that properly. Okay, so what it looks like now, instead of this running along the bottom of the ensemble, it's actually pretty much straight in the ensemble. If you look more closely, this is the GIST temp record here. And one of the neat things about being in charge of stuff is that you get to see sneak peaks of what's going to happen. So the 2015 estimated temperature for GIST temp, which is this line here, is all the way over here. The uncertainties are way higher than it was in 2014. I just calculated that it's 99.7% likely to be a record year and it'll be the first year in the GIST temp record that is over one degree warmer than the pre-industrial period. Yay. I don't know. People have a weird thing for round numbers. I'm not quite sure why. Anyway, so here is my modified ensemble envelope that I think we should have had had we used four things that were accurate. And you can see that the difference between this ensemble mean and this ensemble mean, the modified ensemble mean, is significant and in fact is on a par with some of the estimates that you get from changes because of ocean initialization. Okay, so we have already done decadal predictions and I'm just going to pick on one because Mojib and Noel are here in the audience. It's kind of coincidental. It's coincidental because the forecast period that they used actually ends in October 31st last week, two weeks ago. And so again, I've looked at the data up until last October, which will be released tomorrow and so I can say what the answers are going to be. This was the prediction. These are the decadals, the smooth decadal means. This is the underlying variation in the service air temperature. This was a prediction of absolute global cooling from that last decadal mean to the current decadal mean that just ended and this is actually what's happened. So that didn't work out very well. If you look at the spatial patterns, here is their free running model. So that's just the ensemble moving forward without any initialization of the ocean. This is what happens when you initialize the ocean. You've got this big cooling in the North Pacific, cooling relatively speaking in the North Atlantic as well. And drum roll please. Not really, drum roll, come on, come on. Okay, this is what actually happened. So their free running model without any initialization was spot on. The initialization just made it worse. This is actually an incredible match to this. I don't have the raw data here, so I can't calculate a special correlation, but by eye it's got to be at least 0.8.85. It's very impressive, but not quite the point they were trying to make. Okay, so can short term forcing predictions make a difference? So greenhouse gases, decadal variations in those are too small to worry about, so let's not worry about them. Somebody else is going to talk about volcanoes, they can talk about that. So let's talk about solar, because I've done a little bit of work on that. We do need better predictions of the solar activity going forward. We don't really have that full understanding of the solar cycle, so we don't have that yet. But we also need to do a lot more work in terms of modeling the full response to the solar cycle. And we have done that, so we can assess the second part, even though we haven't done the first part. I'll skip that. Okay, so solar stuff, how does that all happen? Well, you've got variations in the total solar irradiance. You have variations in UV, which affects the temperature in the stratosphere, which affects the ozone, which affects the QBO, which affects the surface, which affects stratospheric chemistry, solar, what do you call it, solar, electron, no, solar energetic protons or something, yes. We don't include that. That's a small term except in the mesosphere. But mainly you're changing the gradient of temperature. That changes your response in the AO and the NAO. And changes to the NAO in the North Atlantic do affect the circulation in the North Atlantic. And so you do have a potential for a connection that goes all the way from the stratosphere all the way down to the oceans. So we've been doing a lot of work on solar impacts. In CMIT 5, we used two sets of, well, four sets of models. We had models that were interactive with atmospheric chemistry and non-interactive with atmospheric chemistry. And we had models with two different ocean treatments. So very fundamentally, structurally different oceans. They have different modes of variability. They have different frequencies and the like. And we did a lot of runs which were all forcings, everything together. And we did a lot of solar only runs for the whole of the historical period. Okay, let's get that. Okay, so if you don't have interactive chemistry, you get a very strong temperature response to solar forcing, mainly in the upper stratosphere. And you get a response in the humidity, mainly near the tropophores, where as the tropophores warms, you get an influx of water going in through the cold trap. Once you start to include interactions, you get temperature changes that aim much larger because the ozone is also changing as a function of the temperature. And you get temperature changes that go all the way down to the surface. The changes you get in humidity in the stratosphere are much larger because the warming at the tropophores is much larger. And actually in the upper stratosphere is negative because of the chemical effects of photoelectric chemistry directly on water vapor in the upper stratosphere. The ozone changes that you get, this is an up-to-date estimate of what the ozone changes are with respect to a solar cycle. So you get an increase of ozone kind of around here. This is about one to 10 millibars in the stratosphere and an increase above. You get that in the model. The magnitude is slightly off, but it's the right pattern. You see chemistry changes. Here this is a change in the oxidation of methane, which changes water vapor. And here this is a change in the photoelectric chemistry of water vapor itself. This is the change because of the change in oxidation of methane. So these set up larger term variants in the stratosphere that make a larger signal. If you look at the surface predictions, so this is in like lag 1 with respect to the solar cycle, this is the non-interactive case, this is the interactive case. You can see that you get a much larger signal in the tropics. In fact, a larger signal in the Arctic as well, that's not quite as significant. In fact, in this model, the total global mean regression almost increases by a factor of two. You see as well interesting patterns in how the annular modes respond to the solar forcing. This is in two different models. This is the HICOM ocean model, this is the Russell ocean model. And you can see this going through in time. So the solid is the response in the same year as the solar peak, and then that kind of becomes an opposite response after about three or four years. In this model, it's the same pattern and it's more significant in this model. And you'll notice that though it looks like it's similar, the actual response in time is a little bit different. And if you look at just the Atlantic change, if you just look at the Atlantic change, you see in this particular model, you see a very large response immediately which is towards a positive NaO and then a negative NaO, subsequent after three or four years, which is something you also see in the, you can infer from the data. Though in this model, it wasn't quite as significant. So does that impact the underlying ocean circulation? What you're seeing here is animations of the significance of changes in the ocean, in the North Atlantic. So everywhere where it goes above one, so anywhere you see an orange point, that is highly significant based on whatever statistics I was using there. And so what you see is this, so you can see, okay, so here that's a negative and then it's going to be positive at about a lag one and here there's a positive at about a lag five, okay? That's a very different response. And when we've looked into this in more detail, what we see is that the different models have a different coupled variability between the wind patterns in the North Atlantic and the response in the oceans. And I thought initially, oh well, it's going to be dominated by the top-down variations, but even though they have the same top-down variations in stratospheric temperature gradients and the like and changes in ozone, by the time it actually works its way out and it has a response in the ocean, the timing and the modes that it interacts with vary as a function of the structural variation in our ocean model. The magnitude of the changes that you're seeing here are around 0.5 sur-drops in the maximum over-turning stream function with different lags to TSI in the different models. So it's not clear to me whether we have an observing network that would allow us to observe a solar cycle impact of 0.5 sur-drops on a solar cycle basis. I don't think we have it right now. It's possible that we might see it at some point. But it does indicate that there is a potential for a small solar impact even on ocean variability. So let me quickly conclude. Forcing changes clearly influence decadal variability and sub-decadal variability and can inform predictions. I think that's very important to remember. Ocean initialization experiments do have some way to go. And we really need to think about the uncertainty in future forcings. This is non-negligible. And when we design an ensemble that is supposed to be something that we can actually say, well, to the people, look, this is the envelope of stuff that might happen. We do need to be including the uncertainty in the forcings as a function of the uncertainty in that ensemble. We can't simply be saying, you know, we all used the same forcings and it was wrong, and okay, well, we did something different to what actually is going to happen. So I'm very keen on having variations in future forcings be part of those designs. Solar forcings has very detectable changes in the stratosphere. There are changes in the annular modes, however where you cut it, once you include enough interaction to fully capture, not fully capture, but to capture enough of the solar variation that it makes that difference. And the evolution of the North Atlantic coupled system is really, it's a coupled problem, right? It is not a top-down problem. It is not a bottom-up problem. It is a coupled problem. And we really need to be looking very, very closely at how the different coupled variations in the control runs or in the internal variability of these runs are impacting how they respond to the forcing. Okay, thank you very much.