 Good afternoon. I wish to thank, to give a special thank to my colleagues Johanna and Elodie who has left the serfax several years ago. I wish also thank to the organiser for giving me the opportunity of doing this presentation. I especially to Anna for fighting every day to improve the food during the lunch time. Okay, so I just start with the introduction of this presentation. So the couple models are going to simulate a climate which is very different for those who have one. So if we want to make a prediction we are going to initialize a model for those observations at showing that this scheme. So the model is going to, no, this is not a good one, sorry. So the model is going to is going to tour its own climate which is also its equilibrium state of model attractor. So this is the model drift and can be defined as a sequence of physical processes occurring at different spatial time scales by which the model is going to reach its own climate. Okay, so one crucial question is does this drift affect our predictions? This is very difficult to show but we think yes. So the challenge is to minimize the drift by improving our initialization techniques. For forecast evaluation the drift are removed. We cannot compute the skill score where the drift are present in the predictions. So they are readily analyzed. However, this drift can give useful information about the processes involved in the development of model systematic bias. So this can be also helpful to give us the information about which are the mechanisms that we need to improve in our climate models. So the objectives of this work has been first to investigate a new initialization methodology to minimize the drift and second also to analyze some of the processes involved in a drift of a model where we perform decadal forecast. There are two regions here, the tropical Pacific and the North Atlantic. We have focused on these two regions. And you started with the initialization methodology. So this is a full initialization skin. But instead of starting from the raw analysis, which is the nemovary analysis in this case, we have performed a previous simulation in which the couple system, which is here, this is our model, the ocean component of the nemovary analysis. So in this case, we expect to obtain initial state where they are closer to the model. We have a tractor to try to minimize the drift and also technical issues we can get initial condition for other components, for example, CIS or land. So basically this nudging, I don't go into the details here. This is quite difficult to point out. This is the nudging of nemovary consist on a sea surface restoring on also 3D, in three-dimensional Newtonian dumping below the mixed layer. Following this, so we have construct two initial conditions, two different initial conditions. The first, I will call it the NOEC initial condition with the nudging to nemovary is applied everywhere, over the equator, which is living free. And the second initial conditions, I will call it no-trop initial condition, because in this case, the nudging is applied only in the mid-latitudes, leaving this tropical mount free, between 15 south and 15 north. So following our skin, we are expecting that this initial condition, no-tropical initial conditions are closer to the model climate than the NOEC initial conditions. Once we have initial conditions ready, we can perform the forecast. So for both initial conditions, we have round decadal forecast between the period of 1960 to 2000, but initializing every five years and doing 10-year simulation of 10 members each. And the second set of decadal forecast, and in this case, I will call it just tech, because these are the official ones that we published in the summit. So we have four-five database and two represent the model attractors represented by the historical simulation, the historical members, the non-initialized for the same period also with 10 members. I'm going to show some plots, because the model did, I'm going to estimate the model reads as the difference between deck minus east as a function of the lease time. So we have an age among all the initial dates and all the numbers. So let's study the first day tropical Pacific. So let's have a look to the initial conditions that we have generated and the differences respect to the model climate. So this picture shows the solar mean temperature in the Pacific, the difference between the two both initial conditions and the historical simulations. You can see here that in the mid-latitudes the initial conditions are colder. They are identical because according to the nitrogen protocol which is the same in the mid-latitudes. However, in the tropics there are warm bias respect to the model climate in the ocean surface which is a stronger in the case of noic initial conditions. So when we are going to initialize our model we are going to put these warm bias into the model. So how does the model react to this? For that we can track the evolution of the thermocline along the equator together with the surfer winds. So this picture shows the lead time. This is a hop-muller diagram. So you have the lead time here on the way axis on a seasonal basis. This is the longitude that the shading correspond to the depth of the 20 degrees is a term and the arrows are the winds. For both sets of the Kedel forecast, deck minus east at deck noic minus east. So it is clear here that in the first year of the forecast we have a shock on the winds with a weakening of the esterly winds. We are going to, in both simulations, trigger perturbation in the thermocline. We are going to travel from west to east and to deepen the thermocline on the east. This happens at year one of the forecasts. At the second year, which is more clearly seen here, this is the opposite effect. There is a strengthening on the westerly winds and the signal is propagating again toward the east with a shallow word of the thermocline. So this is a very well-known mechanism given warm cold conditions in the tropical Pacific, which is clearly the ENSO mechanism. So we see that the model to release all the heat that we have put during the initialization, she is doing something like an oscillation by the alternating NINU and NINU events during almost four years. We can see here, you don't see here, but I can tell you that after year four this mechanism is damped. This oscillation is damped. We can see also that with our technical initialization this spurious oscillation has been drastically reduced in this simulation. Okay, so the ENSO is the most important signal in the globe. So we expect to have some teleconnection. So this signal is not confined into the Pacific. It's going to travel outside. So if we look for this snapshot, this is the first winter of the second year, we can have a look at the predictions, differences as well, decayed or minors east. So you have here the atmospheric circulation given by the geopotential head of the 500 hectopascal and the precipitation by the shading. So we recognize very well the precipitation pattern associated to the NINU. And also you can see that this signal, which is more prominent in the case where this effect is stronger, you can see that there are anomalies in the atmospheric circulation spread out into the mid-latitudes which are coordinated with the positive phase of the pineal and the negative phase of the North Atlantic oscillation. Okay, so this is concerning the drift in the tropical Pacific and now let's travel to the North Atlantic. For the North Atlantic rig, which is happening in our different time scale, let's focus on the atmosphere first. So this picture shows again the initial conditions. In this case, I'm just going to show you the decadal initialized from the no-drop initial conditions because both drifts are quite similar. So the no-tropical initial conditions minus the in-sip re-analysis. So as I remind you that in this simulation we have restored the surface of the ocean so this can be seen as the bias of the atmospheric, the standalone atmospheric component of the model. So you can see that this bias is a positive bias of sea level pressure, respect to onset. We project on the negative phase of the North Atlantic oscillation. If we look at the predictions for different 10 scale in the lead time, so the first 10 scale, 1 to 4 and the second one, 5 to 10, we can see that this bias is still at the beginning of the forecast so the atmosphere is going to adjust very quickly to its own attractor. However, this bias is simplified, obviously by the coupling. So we have the negative phase of the North Atlantic oscillation with this acting at the beginning of the forecast. So maybe this is going to force the ocean. Let's look at what happens in the ocean circulation in this case. So the top panel are again the drifts of the differences with respect to onset for the two different lead times and the bottom panel shows the evolution of the MOC. So these are the differences between the cattle forecast and the initial conditions. At the beginning of the forecast for the first lead time video, there is almost no differences in the MOC with respect to the initial conditions. However, at the end of the forecast, we can see here clearly that the MOC has drastically reduced, which can be seen as a response to this now minus forcing because there is a reduction in the westerly winds together with a warming on the subpolar areas which are going to reduce the deep water formation. So what happened with the... This is for the vertical ocean circulation. What happened with the recent circulation in the subpolar areas? Sorry. Let's focus now on this area. This is the subpolar area. So we have made the difference between the east and the western and the east of the subpolar area. So we have performed indices. These are the predictions. And the initial conditions are in blue. The cattle forecast in red and the attractor of the model is in black. So let's focus on the west. What happens here is that from the initial condition, the model is going to drift and it's going to reach the attractor around year 405. There is a slackening of the GR circulation coiling with this now minus forcing of the atmosphere, coming from the atmosphere. In the case of the east, it's quite curious because there is also a slackening of the GR. However, the model is going to drift away from the initial conditions. So what happens, fortunately, we save 30 GR forecasts to date so we can see what happens if we run for a longer. So this is very funny. The GR circulation is going to weaken until year 11 or 12. And then again it's going to strengthen to reach the attractor. So this is quite end with two time scales of the response of the subpolar area. The first time scale is what is represented here, but this is diagram. The cinema. So this is the T.S. diagram in the east of the subpolar area. I consider only the first 700 meters. So at the beginning of the response, there's going to be a warming, clear warming. This is the, sorry, the color represent the lead time. You have 10 years here. So at the beginning there is a warming of the subpolar area, subpolar gyre coedent with this now minus forcing on the weakening of the westerly winds. However, at the, from year four or five, there is also a desalination and a cooling of the gyre, which are coedent with the cold advection coming from the north, coedent with the normal loose atmospheric circulation, which is also associated with a sea ice formation, which you can see here in the KDAL forecast, comparing to east as a function of the lead time. This is for the air, sea ice formation in the gridland in Regium C. Okay, so there is two time scale, the first time scale of the response is the warming on the gyre, and the second time scale is the cooling of the warming, the strengthening of the circulation. So I pass here to my colleague who is very important to study the drift, because here we have shown several mechanisms that help us to understand our model's behavior. There are many papers, well, not many, but several papers emerging on this topic, and in particular this paper and Tonya saw and was studying the drift in the Tropical Atlantic, they show that even in the mean bias of the model are the same, the mechanism in the Tropical Pacific with cutting a shorter time scale in which the model is going to use the Enso phenomena to release the energy, the energy, the exit of heat given during the initialization. This spurious signal is, which is important, is propagated outside the tropics. In the case of the North Atlantic, our mechanisms seem to be associated with the bias in the atmosphere, who is going to drive the ocean and to reduce, which this bias is projected on the negative phase of the NIO, who is going to lead to changes in the ocean circulation at longer time scale. That's all. Thank you very much.