 Okay, thank you very much. Apologise for the slightly disjointed title. I was asked to talk about Beyond distinctive 5 Decada Prediction, which is a good topic, happy to talk about that. But I also wanted to show some results on the impact of aerosols on the warming slow down, so I will try to cover all of that in 15 minutes if possible. So I'm going to show some results from new Decada predictions that we've been making at the Met Office. It's a much higher resolution than previously. I shall show some results for predicting ysgol, y Llywodraeth Llywodraeth Cymru, yma i rai cyfrifoedd yn cyfyrddodd gyda Sarhowll. Fi'r ystyried am y pethau cyfrifoedd gael cyfrifoedd cyfrifoedd, yw'r paradox yn gwneud y prynhyrchu'r cyfrifoedd yn y pethau, ac rwy'n gweithio'n rhaid i'ch gael ei bach o'r rydw i'r rhan o'r cyfrifodau. Yn gyfaint i'r idea ar y cyfrifoedd yma o'r gweithio ar gyfer hyd. Felly y gwasanaeth yn y gyfrifagurau yma yn unig, mae mae'n dod gan'r CMAF 6, mae wedi gweithio cystalod. Felly gan swydd wedi bwysig, mae mae'n gweld unrhyw o'r gwahanol, ac mae'n gweithio, a i gael bynnag, ser i gyd yn 150 km arall. Gweithio quesif hynny a'r greif wahanol o'r gweithio llunio. Mae'n rhaid i'r grannu bwysig oherwydd, ychydigfaint yw'r newid wedi gael bydd y dweud. A'r hynna'r gwahanol ar y councils slan o ddechrau Windynau Oedolfennu, ond Llyfrgellydd, 당fynwyr Aelodau a Llyfrgellydd i Unedd Unedd Caerdydd. Llyfrgellydd y 就是 hwn yw'r sefydlen. felly mae'n drws fyddwn y Llyfrgellydd, felly mae oedd yr Hweith��ydd rhai iawn y Llyfrgellydd iawn. A'r rhain, mae'r trefenedau yng nghaerfaedleion, ond mae'n rhain o'i theument, a'r rhai a'r llai ffobl, felly mae'r rhain o'r ffriseid tidion, oeddo Iesbyg yr wrthag,肉 yn rhwynt yr oed, 1 tyd. Rydyn ni'n gael eich rhaid pleidogol gj sealed a daf y brwysig ar gweithio'r unig a fyddai'r pryd yn ddod oes i whithio am ysg Gymru arтиncau gweldol yn wahanol yn 1981, felly mae'r eraill hwnnaol yn fwyaf ar arbennig. Aa, mae'n ffrifoedd i'r explored, maen nhw'r sefydlir safwm gan ysgrifell yma, rydyn ni'n chi'n meddwl yn positio yma, oherwydd rwy'n gweithio'r pryd yn ddwy gyd, Rhaid i weld weithio i chi'n ffordd gud i'r hyn o ran ymddangodur yn ffordd gyda'r newyddeth yn arfermilydd. Roedd yw'r wrthgrifennu'r chefnodd gyda'r hyn yn amlwg yma neu'r ysgrifennu'r sylfaenidau. Mae'r llif i ddiforol ei fodf yn meddwl â'r hyn o bryddiadau predictio sydd ei fodf yninkswyd i'n ffordd. Ond mae'n ffordd o rheid hunting o bryddiadau'r hyn o fodf yn ei hir yw, iawn. Felly ar gyfer panel trwcol wedi bod gennych y cly croel ac mae'r llwyslau'r llwyslau o hyfforddiol yn 13 mlynedd. Felly mae'n y same panel ifanc, ond efallai'r cael cymdeithloeth, ond y gwpaeth ddyn nhw wedi'r rhan o'r llwyslau awf. Twyrdd y gwirionedd arblig yn awf yn deilio mewn y L synthol ac mae'r llweithau o'r hain. Yma, mae'r cyfion yn y maen nhw. Mae'r ysgrifennu i'r hwn ar flwynt yn ar gwrthfa. Ond yna ydy'r cyfnod yn gwybod yn cyfnod oherwydd, sydd yn ystod y dylai'r cyd-fynod. Yn y ffyrdd yng Nghymru, Jeff sgolwch ystod yna, mae'n meddwl yn gyfnod i'r cyfnod o'r cyfnod. Mae'n mynd i gyd yn ymdill ymdill yn ei ffyrdd ymddill yn ymdill, mae'n ffyrdd yn ymdill yn ymdill. Felly mae'n mynd i'r cyfnod oherwydd. Mae'n meddwl i'r cyfnod o'r cyfnod, you would really need to do specific experiment, numerical experiments, but what we have done here is a lagged kind of composite analysis so focus on this bottom right-hand panel. This shows the December to the DJF mean sea level pressure in the observations following the November Nino3 indicates if you like. So it is the warmest Nino3 minus the coldest Nino3 index in November. We composite the sea level pressure in the following December to see if there is a potential impact from Nino3 on the north Atlantic oscillation. You can see that there is, in the observations, as we have a negative pressure over Iceland and the positive pressure over the Azores, which would be a negative phase of the north Atlantic oscillation index. Ac ysgol yn i'r ffordd agnig yma yn fwy o modda'r ynch ond. Yr un oed, mae gennych ei wneud efallai hyn o'r ffyrdd y mynd i'r eistedd gyflogu. Mae'n ei ffyrdd achos gymennydd yw'r pwyntig awd bedroom yn ein cwmgei, sy'n ei ffyrdd iawn i gydig ym Mfyrdd Merthyn o'r oesolant yng Nghymiddag. Efallai yna'r fwyntig awdurdod, ac yn yr awdurdurdod yw'r sgol i'r pwyntig ynglyn yn y nesaf tyll yn ddiweddolol gan Gwấtrydorol. caused isolation following our Nino-type event. This is a potential source of skill of the second winter forecast. You need to demonstrate that the answer itself is predictable, and this is what this upper panel shows, this is the correlation skill for predicting the answer itself. That means the unique number of levels on the Nino 3, 3.4, for which index, versus the lead time of the forecast. It's a three month rolling round between eu cyfnod i'w rhaid i'w January er i'w hunain ar y tro chi eisiau gyda geniwestiwni gwaith y Llyfrgell Ydyn nhw yna fel o'r Llyfrgell yma yn 0.5 ac yn rhaid i'w cwfwys yma yn y Llyfrgell yma diolch gan y gyd yn 0.6 ar yr awn i wel, mae'n cael ei ddechrau a'r llyfrgyntau ei ddod yma o amser gan y cyfnod iechyd wedi'i elwydd. Ynaran maedden yn y Llyfrgell yma yn y Cyfrifol Llyfrgell Cysyfrwyr yn ycydddiad bydd y syniad cyflysm yn iawnóa gyflym iawn, i wneud hynny, y cryl ac ychydig yn gwneud yn ei wathol ac yn gwybodaeth ar gyfer y trope查. O'r cyflym iawn, mae hyn yn fawr y dyma cookede, mae'r pwysig wedi gŵn ddod o'r dyfodol yng nghygof yw gyflym iawn o'r cyflym iawn yn y dyfodol, mae'n bysig o ddechrau mewn oes yn gwneud, i f?​gwyfwyr y dwylo ddiddoriad i yn dynol, a mae'n ddiddoriad maen nhw'n ysgriferion ladwyd y lantech oesolisiad, fyddech byddwch i ddychglied i gael ei ffasolisiad oesolisiad arhel. Zektor i fewn i'r ffordd y maith yng Nghymru, rydyn nhw'n flwyddynhau mewn lle i'r gwell? Felly, rydyn nhw'n addysg fel yna yn y Llyfrgell yng nghymhwyl wedi'i gwybod, iddo o blaidol sydd yna nesaf o croes cyfnod ac yn edrych yn ddweud y gydaglion trwy'r hwn. Rydyn ni'n fath i'r gondol i ddweud hynny'n llud, yr ysgol yw'r hyn yn ffodol yng nghymru nesaf, y nifer yw'r ysgol yw'r ysgol yw'r ysgol yw'r ysgol yw'r ysgol yw'r ysgol yw'r ysgol yn ffordd yn y model yw. Yn gyng ngoswm, yna'r pryddyn nhw yn ddiweddol, ond yn y ffwrdd, rwy'r gwael fyddio byddai'n bdoedd o'r bwysig, yw'r bwysig o bwysig o bwysig o bwysig o'r bwysig o'r bwysig o'r yn y North Atlantic too. OK, so how a rainfall. So these are now predictions of years two to five, so the lead time of well over a year for these forecasts. And it's the correlation skill for predicting, well, it's a map for predicting rainfall. So red is a higher correlation to demonstrate some skill. This was done with our AR5 type model, so the lower resolution one. And here in the bottom we have the latest one, the higher resolution. And what you can see is that in the Sahara region, the fourth model, we tended to get some skill right on the western edge of the Sahara, but not propagating all the way across. Now this is consistent with predictions of, a skill for predictions of the North Atlantic, which have been shown many times to affect the position of the ITCZ over the ocean in the Atlantic. And in this model we got associated predictability of hurricane frequency associated with shifts in this ITCZ location. But the information never propagated across the Sahara as it seems to do in the observations. Whereas in the new model we're now getting that propagation. And Katie Sheen has a poster at the moment out there which you can go and look at to try to understand how this information propagates right across the Sahara. So that's a potential important development that we're getting improved skill with this new version. We don't understand exactly why yet, so that's something we need to focus on. Now the signal to noise paradox, this is a really key issue I think in all climate predictions potentially. Some people have started to cotton on to this and some people are not aware of it at the moment, so I wanted to really highlight it here. So if you look at the skill, so this is the correlation again, this is for predictions of the North Atlantic oscillation. If you look at how much skill you get as a function of the number of ensemble members that you run and then take the average of, so this is the skill for the ensemble mean. So for example if we have 20 ensemble members that we take their mean, then the skill is wherever this black line is. You can see that this skill is quite a strong function of the number of ensemble members that you can use. That's potentially okay, but what's really interesting is this blue line and what we've done here is the same thing, but now instead of measuring the skill against actual observations, we're taking one of those ensemble members as the observations instead of the real observations, and you can see that actually, and this is totally unexpected, the model is less able to predict itself than it is able to predict the real world. So our interpretation of that is that the model is responding correctly to the drivers, whatever drives the NAO, there is a response in the model, but it's too noisy, so the signal-to-noise ratio of that response is too weak in the models compared to the real world, so that when you take the ensemble mean of lots of model members, you're left with that skillful signal. So this has some important implications. It means we can make some skillful predictions with the models that we have right now, but you need to take a large ensemble and get rid of all the noise. When you do that, you won't get much variability in that ensemble, so you have to then adjust it in some way. It also means that potentially there's more skill available if we could run, so we can project this line forward. There's a theoretical relationship that allows us to do that, so if we could run more members, then we should be able to get more skillful predictions, and this is true for the second winter because it has some important implications for measuring skill. It means that perfect model predictability experiments are not necessarily an upper limit of the skill that could be achieved. This is just clearly shown by this difference between this model-to-model skill and the model-to-real world skill. Many measures of skill, like RMS error, for example, or probabilistic measures, will not necessarily pick up that there is any skill there because of the too much noise. If the model does not respond correctly to the drivers, then the probabilities that are assessed in those attribution-type studies will not be right. It has implications for how we view the role of internal variability. Now I want to just come on to this slowdown in surface warming. I'm not calling it a height, I'm really calling it a slowdown because this is how it is really identified. If you compare trends over different periods, this is a rolling time series of 15-year trends in globally-average temperature, and the problem seems to be that if you look at the most recent trends, they're lower than they were in previous periods. This is true for all observational data sets. Even this Carl Eytel study that suggested that when you employ more corrections to the observations than this slowdown is no longer there. It really is there. You can definitely see this slowdown. Now, when I plotted this, originally this was a bit of a surprise, and what this shows is the CMIT 5 model simulations with the ensemble mean in red, the thick red line, and clearly these are also capturing a recent slowdown in this rate of warming. The models trends are bigger than the observed ones in general, but they do capture a recent slowdown, so there are a couple of things to explain about this. Is it this driving this slowdown, and why is the model trend bigger than the observed one? The fact that the ensemble mean captures this signal means it's externally forced. This ensemble mean is averaging out the internal variability. What we can do is look at the model simulations that are forced by different factors individually. That's what I've done here. The red line is the same as on the previous plot. The ensemble mean of the models that had all of the forcings in them, so you can see this slowdown. I have greenhouse gases in pink, so there are quite a large ensemble of models that ran just with greenhouse gases. I have the natural forcings only, so this is solar and volcanic aerosol changes, and then I've put on some models that only ran with anthropogenic aerosol changes. You can clearly see that this slowdown is well simulated by this natural ensemble, and the reason for that is not surprising when you think about it. It's the recovery from mountain pinotubo. When mountain pinotubo went off, the surface air temperature is cooled. The maximum cooling was in 1992. If you start calculating a trend from 1992 and it ends in 2006, then that will be the maximum warming trend that you will get from the recovery from mountain pinotubo. If that's a maximum, then every year following that has to be lower. That's the primary reason for this peak in the trends. That's quite obvious, really, but it highlights the importance of thinking about the impact of volcanic eruptions when you're looking at trends over different periods. I think what's more interesting about this plot is this blue line, which is the aerosol-only ensemble, and that is also showing a slowdown or an increase in the cooling, if you like, from these anthropogenic aerosol runs. Many studies have highlighted the importance of the tropical Pacific in this slowdown. If we're going to have a convincing explanation that external forcing has played a role, then we need to reconcile what's going on in the tropical Pacific. If we look at the most recent period that we can do in these aerosol-only simulations, they all end in 2012. This is the most recent 15 years that we can look at. We look at the pattern of trends surface-temperature trends in those and compare it with the other forcing factors and the observations. You can see this well-known trend towards a negative PDO type pattern in the Pacific. That's definitely not captured by greenhouse gas or natural forcing, but it is captured to some degree by the aerosol-only ensemble. Clearly the aerosols are protecting in some way onto this PDO pattern. If you look at other variables that would be important for that, variables that affect the dynamics of the atmosphere. Here I've got the sea-level pressure. It's the same trend over the same period, now in terms of sea-level pressure in the observations and in the aerosol-only ensemble. You can see this allusion low has been weakening over this period. There's a number of high pressure in this allusion low region. There's a number of high pressure in these model simulations. In terms of the tropical the zonal winds that have been highlighted as a driver for the negative PDO type pattern, you can see not in this region that was really highlighted before. This is in Matt England's paper. He picks a box in this region and demonstrates very large trends in that. That's not really picked up so well in these model simulations. There is some broad agreement, as you would expect in terms of the zonal winds. The surface temperatures, the sea-level pressure and the zonal winds would hint that anthropogenic aerosols have played a role in forcing this negative PDO type pattern. The model pattern is weaker than observed and that's something to be explained but nevertheless it is definitely there. The question is if changes in aerosols are forcing this dynamic pattern in the atmosphere, how do they do it? That's what I've tried to do here. On the left-hand panel here is the trend in aerosols if you like. It's a sulphate aerosol optical depth trend over this period, 1998 to 2012. What you can see is a reduction over the US and Europe and an increase over China and India and that's well known that pattern has been there. How would that force changes in the elution low? To look at that, I've taken an index of the elution low, which is this North Pacific index. It's just simply the average pressure in this box. This is an established index. If you correlate trends in aerosols in these two boxes, I did all the boxes but I'm just highlighting the American box and the Chinese box. If you correlate trends in aerosols with the strength of this NPI, the Solution Low Index and do it for different lags, this is what you get here in these model simulations and the shading is a spreadsheet that you get from all of the models that had an ensemble, there's three of them that had an ensemble of simulations. You can see that when the aerosols lead there is a correlation between the Chinese aerosols leading the strength of the Solution Low Index and then there's a negative correlation with the American or the U.S. aerosols again leading the strength of the Solution Low Index. If you look at a time series of trends in this index, the black here is just the time series of 15-year trends from this NPI index in the observations and you can see that it's varied quite a lot over this time period and then if you compare that with the same index but in the ensemble mean of the aerosol only for things, that's what you get from this blue curve I think what you can see here is in this early part of the period there's no real correspondence between the aerosol runs and the observed time series of changes in these trends and if you look at the time series of aerosol optical depth trends from China and the U.S. separately you can see that they didn't vary an awful lot there's perhaps a little bit there but they didn't vary, especially over China they didn't vary very much in that period and then there start to be some pretty big variations in these aerosol trends and emissions and you can see that these simulations start to come in phase with the observed changes in this NPI index so it's kind of suggesting that these variations in trends of optical depth are affecting the solution low through changes in SST I think just one final point on this is that this dash line is one of these models this is the Hadley centre model that was actually run out to 2020 in the future so we can assess future trends in this one and in this one the aerosols over China in this scenario to decrease quite a lot and in response to that this model simulates this kind of strengthening of the of the illusion low a consistent with a change in phase of the PDO index OK so the final point was on trying to explain the discrepancy between the model trends and the observed trends this goes back to the first product I showed time series observed trends and the time series of model trends is this dash line here in fact what this is is if you add up the trends from the greenhouse gases natural and aerosol ensembles but it's pretty much the same as the all-forcing trends so what I'm trying to do is explain this discrepancy and what I've done is a detection and attribution analysis on these trends in terms of these three factors greenhouse gases natural and aerosol so which are these the blue, green and pink lines so you take those model simulations and you compute the weights the weights that you would apply to make that set the linear regression sense optimally fit the observed time series and if you do that these weighting factors for greenhouse gases and aerosols not inconsistent with one all of the factors are significantly greater than zero so they're all playing a role but the one that is significantly less than one is this natural one so it's implying really that the models are over responding to these natural factors and a few papers have suggested that they over respond to volcanic eruptions so if you apply this weight these weightings to the time series you end up with this dotted line and I should say that these weightings were obtained by only using data up to 1995 so it's before the warming slowdown sorry Paco before the warming slowdown so if we'd been in 1995 we'd done this analysis then we would have projected the model simulations according to this dotted line which is much closer to the observed one so I will leave you to read the summary of the questions, thank you