 OK. Omedne ga prezeta taj poznato. Protoče, da bo dober narediti se vzore mtakov na ISMWF in tako da smo počeli v instalčke in vsega jaznivne organizacije tež pastryna. O ekipno. Protoče, stajte za nami ked vsim.hee, in dzi z tento obstru, which was a bit one on the basis of the idea of futuro accords on short term klimat, timescale. The idea is basically that oh, sorry. The idea is that if you have a climate variable kar da situacije vjela pred tem posebno. Teške ispodостصje seddanova idovina krat● gutago zašto se izageni, č clinicianji diesne verjevali producersia z воспite … ... v robili tudi je še basa po zres Arcticjkel? Nao več malo blocks… Do menej… koigeni Methodal? Kde zurnoziči nrstak. Kako prejdojamo in je prijemno od slog, da je veliko poznipit. Z blocks, kako je poznipit, nekaj se ga daje v tem zač ranchiti oselji, ali neko kot nekaj in je bilo način taj vokol... Prejdeš, da ne psii, na mali, na danes, after a while you lose predictability. If you put all together you put in the forcing and the correct initialization, you can hope to be able for the short term to follow the climate oscillations in principle to predict if there are some pozeses in the warming in if there are accelerations in the warming for a while. Basically, the idea was just to fill this gap between sasonal forecast and climate projections. So, in this way that these exercises of the Kedal prediction started Evno, da je lahko zazupak, da je to povrknice, zelo je to, da je to povrknice. Je to, da je to, da je to povrknice. Prejšel, da je to povrknice, da je to povrknice. in se nimi je vse zvonil na poslednju IPCC-R5. Na vse izgledno je načo predikčenja in projekcija. In se vse znači, da je dobro vkročno. I kaj ga se vse znači, da je vse odpravljena vse izglednji vsično izglednji vse vse vse vse vse vse vse. tudi in vse modelje zelo je vse vse. V nekaj pričo je to vse. Tudi je tudi model in tudi je tudi vse, in tudi je tudi in v svoj pomečen, vse je od tudi, in tudi je tudi, tudi je tudi in vse, vsebe vsebe, ki je nekaj spred, da se počusti, da se počusti, da je začal tudi vsebe. In zelo, da se tako nekaj počusti, vsebe je, da se tako način, tudi je vsebe in dobro vsebe, v zelo prejim ljudi, in vsebe vsebe vsebe nekaj, in ki se počusti, ali, kako so pričili, nekaj ne bomo vladi, zelo vse bo otvarjati kursovaj čas, ko priče inštrih konditions počusti, in še deforsi. Zelo, da ne iz vsej proprilim, ker se vse modeli so vseče vseče vseče, nešte vseče na vseče, na vseče, na vseče, na vseče, na vseče, na vseče, na vseče. Sreda toga, kako se oče, da smo počeli, in da smo počeli, da smo počeli, in je to oče, Kaj soHere we can see that comparing the for the mean global temperature and comparing basically the initialized predictions versus the projections there is sort of a marginal while there is a signal so basically the initialized predictions are able, are to be more consistent with the pose of the hiatus in the global warming with respect to the projections. And, for example, in the case of some specific indicator like the AMV, the Atlantic multi-decadal variability, actually one can see that in this area, the predictions are sort of doing a much better job than the projection. So despite of this bias, the model, there is some signal. Probably not as good as expected, but there is something. So the idea is it will be OK, maybe if we are able to decrease the bias in the model, then the model could initialize, they could do a better job on that. So what I'm going to present now is these experiments that have been done with the ECNWF couple model, which is basically the same as that has been described by Franco before, and in which in this case there is also the CIS component initialized, and it is a more recent cycle with respect to the actual, to the current seasonal system cycle. But basically, and the resolution, which is T255, so about 80 km for the atmosphere, is the same. So I used these systems, and I basically repeated the exercise of the CMIF5, but doing short and integration, so only three are long integrations. Every starting from November 88 to November 2000, so 22 starting dates every November, and five ensemble members. So they are actually multi-year, not decadent integrations. And this is our control integration, and they should be compared with higher solution integration for the atmosphere, so basically the same, exactly the same system, but in this case the resolution of the atmospheric component has been doubled, so basically T511, which is about 40 km in horizontal. And so done now, first I am going to show some results of this, and then I will talk a bit if there is time about these other experiments. So basically, the first thing is what about the bias. And here I am going to show the bias for the two meter temperature. Here we have the control integration, so basically the T255, 80 km in horizontal, and here you can see the higher solution integration. And this is for the first season after the initialization, so it's DJF. This is the average, they mean over the first year of integration, and this is the average over the three years. And basically one can see that there is, especially if one considers the last, the old periods, a decrease in the bias, and especially in the tropical pacific, which is quite consistent. But basically, so one could expect that this decrease in the bias could lead to somewhat better predictions. And so here you can see what happened in the first season. The first season is DJF, and then this is anomaly correlation for the air temperature, and seasonal, and this is the control, and this is the higher solution. And one can see that actually the higher solution, there is an impact especially, which is quite interesting over Africa, over this part and over Europe. Even if over Europe actually the signal is not significant, but is moderately positive, while there is no signal and even a bit negative in the control. If one look at the multi-year integration actually, one can see that basically for the first year there are no detectable differences, but if one look at the two years, the average year two and year three, actually one can see an improvement over the pacific and the same improvement over the region of the supolar gyre. So one can expect in a way to see something also, an improvement also in a way in the prediction of the hiatus. And here it's the prediction for the two to three years for the toss in the global. And this is the control, here you can see that actually I forgot to mention before that the volcanoes were not included in the forcing. So basically here one can see that what happened is that the model miscompletely this case, but which was quite expectable. And then for the control we don't have a very good prediction for this period of the hiatus for the year two to three, but actually the higher solution is not much, is not incredibly better, it is not much better. There is some signal here here, but if one put together all the ensemble members, basically they belong to the same distribution. Then just quickly I would like to show some results for basically an experiment in which one look at the sensitivity of the initialization of the initial condition in these experiments. And the idea was just to repeat the experiments for three starting dates, just shuffling, so switching and swapping the forcing and the initial condition. And this experiment was already done in a multimodal contest for only two initial conditions. And basically what it's interesting here is when you swap for example, I choose randomly three starting date, which are 88, 2002 and 94. And for example, if you do, we switch the forcing between 88 and 2002, even at the first year you can see a signal. Basic in blue is the control and the red is the integration in which the forcing, in this case is the forcing of 2002 and in this case is the forcing of 88. And basically there is, even in the first year, quite a change, which I didn't expect because in other experiments we didn't see that. So basically it seems that there is a very big sensitivity to the forcing in this case. And if one look at 94, the same with 2002 is more or less the same. So big, big sensitivity to the forcing. And this is not the same for, for example, switching different date, 88 in 94. For year one, if one look at the year, it's more or less the same. So there is a very big sensitivity to the forcing. Well, I can skip here because I show different basings, but we can skip to the conclusion, which I lost. So basically you can read the conclusion. I don't want to repeat the conclusion, but I wanted just briefly to say something that it was about the idea of a benchmark in which one could really try to understand the role of the initialization in predictions. And so far, all the times, the initialized prediction were compared to non-initialized predictions. But this is a bit comparing orange with apples because in one case, we are just operating over the attractor of the model. And in the other case, basically, we just pulled the model outside the attractor, even for full initialization, but also with anomaly initialization. So it's not, I think that it's very difficult to understand what is going on. And I thought that if really one wants to be pragmatic about trying to do predictions, the best way would be actually to compare predictions with the best possible initial conditions and the forcing with predictions with initial conditions that are taken from the real world, not from a model, and see how is the sensitivity. Because I would say that this is very important to understand if this exercise is really important. Well, if you can go on with the predictions, what are the potentiality of the methodology? OK, thank you very much.