 Okay, thank you. Thank you. I would like to start thanking the organizers for this nice workshop and also my co-authors Noelia Otero and Marco Gaetani who couldn't be here and What I want to talk today is about some decadal predictions of the Sahelian rainfall so in the last and the previous two talks we've been already looking at the West African monsoon and in particular to the Sahel rainfall because of this Great trends with the drought and the partial recovery and and the idea of our work is Could part of this decadal variability this decadal signal could could it have been predictive? so There've been already several works and put here just an example, but they've already been at least I think Three at least or four works looking at decadal predictions of the Sahelian rainfall This example I'm putting here Is showing just for one model the Canadian that it shows some some head that it could have been predictable Though this is quite model-dependent one of the things is that the the previous works were focused on rainfall direct outputs, but Typically General circulation models the the atmospheric part at least is known to to have some problems depending on the model with the precipitation outputs while Dynamical variables are typically better reproduced so here put an example from a paper of Nijade and collaborators in 2011 in which they are showing it's just an ameep type of simulation in which you have the atmospheric model Driven by observed SSDs and in black and black Bars you can see the skill these are for different models the skill in terrestrial variability in captioning the internal variability of the Sahel And it's quite model-dependent and Some models are skillful some others are not but when they look at dynamical variables. This is represented here by the white Bars and they typically show better performance So the idea of a work we was to try to predict Sahelion rainfall, but instead with the direct rainfall outputs looking at dynamic variables So that was the main aim starting the skill we focused on on the CME 5 models on summer And we also evaluated the additional skill coming from the initialization if it makes sense to do decadal predictions, so Regarding the data we used we looked at CME 5 Simulations to experiments the decadal handcast or decadal predictions with I guess this maybe belongs more to tomorrow But we've already had some talks about decadal predictions last yesterday. So in the CME 5 What they do is every five years they Issue a model prediction that last that runs from four Ten years so in 1960 they they launch to the the the model and it runs up to 1970 and 1965 they do the same and it runs up to 1975 and so forth. That's that's the core predictions And the important thing is that they have all four things but they are initialized This is in 1960 they were feed a fed with the actual state of the of the atmosphere and of the ocean mainly and Then just to compare if this makes sense or not to all this effort of modeling We we compare the results with historical simulations that have all four things but are not to initialize So they are typically coupled runs that started in Purdue in pre-industrial times and typically 1850 and they just run on And we compare with observations with precipitation observations and re-analysis data so as I was talking Same before we wanted to to Particularly look at the potential of using wind fields So for the West African monsoon it has been shown and there are many studies I got here one from Fontaine and collaborators in 1995 in which they show that there is quite a link of the strength of the monsoon with the Sonal jets in particularly I'm going to highlight here two of the jets the West Southerly wind monsoons and the tropical easterly yet. This is a plot of sonal winds and this is Precious levels, so this is up and this is down and this is latitude So in this paper of Fontaine they they show that when you have a beta or strong monsoon you have More rainfall over the of the Sahel and that typically comes with a stronger Tropical easterly yet and a stronger West westerly south monsoon So they define an index which is called West African monsoon index whammy and it's the modulus at low levels minus the wind sonal wind at high levels and This in definitions is independent from rainfall, but it's quite related to rainfall When you have a stronger tropical easterly monsoon jets and a strong westerly southwest winds you have more Sahelian rainfall and in our work we Chose the actual region where to do the averages of these quantities depending on the model and by applying and Combine you have analysis of high levels and low levels So these are just some results These are for the To re-analysis product we used the Low-level Sorry the low-level The The modulus of the winds sorry to couldn't the modulus of the winds had low levels From Edda an insect that go with the first combining uf pattern and here is For some of the models we analyze several we analyze 14 I just put here an example so you can see that the the stronger monsoons in the in the first Mode are depend a bit on on the they just said you analyze So we chose the area depending on this first mode and this first mode in the upper levels goes with increased tropical easterly jet and And and also in the in the models and I don't show it here, but it typically goes with increase Sahel rainfall in the observations so What about the skill of the models in this plot so that I guess we'll see more tomorrow I'm plotting one of the measures of skill which it's the anomaly correlation coefficient here in the y-axis And each color is for one model of the out of the 14 and we also did a multimodal Ensemble so now we have 15 and then in the X Axis I'm plotting different lead time so in the first column the first points are Showing what's the anomaly correlation coefficient for the One to four years lead time and the second is for the two to five years lead time Etc. And this panel is for the decal and handcast and this one is for the historical Simulations so in general we can see better scores for the decal simulations. That's that those that are in the slides Here I want to stress that I'm comparing the whammy the from the simulations with the observed Sahelian presentation index from CRU data In general that some models that stand out as statistically significant at least in some in some lead times which are Written here and some of them we also tested but and they were not they didn't give Skill when looking at their rainfall outputs instead of the whammy index so We also thought about this was an idea of one of the reviewers, which I think he suggested to look also at probabilistic scores The only thing is that we didn't have given the Experimental setup of see me five you don't really have that much data So this is just an average score taking into account the 14 models and all the lead times So typically we chose to follow a tercel category just trying to predict above normal normal and below normal categories and You typically want a big a strong heat rate This is you want to issue an alarm alarm rightly You want to predict it and you want a low false alarm rate. This is when when you issue This a lot of false alarm rate is a fraction of times that you issued wrongly an alarm so you typically plot both the heat rate Against the false alarm rate and you want to be over this this part of the you want to to have highest course of the heat rate Rather than of the false alarm rate So this is typically Built for different thresholds of probability and then you have your your your rock curve So we can see for the the canal that you typically for the above and below Categories you are above the diagonal. So you're having quite some skill You can also measure this as the as twice the area underneath the curves Between the curves and the diagonal and if you do this that you have this course here We we see the same information, but I can also show for the historical ones that you you typically have less less skill Yeah, so Sorry, what I showed before was testing our whammy index against observing Rainfall, but we also thought of testing our whammy ingress against the whammy in observation so we thought of using the Ensep and era 40 reanalysis. So when you do the same Scores that I showed before but with Ensep you typically find a decrease of skill But when you do it with era 40, there is a complete loss of skill So we looked a bit into it. Why would we have this loss of skill? What's happening here and To make a long story short the main problem was coming from the modulus at low level So here I'm plotting for in time the Anomalies the standardized anomalies of rainfall from CRU data, which is in in black and then the modulus of the wind coming from the Ensep in green and from the era 40 in Let's call it orange These are standardized anomalies as I was saying and it's the decal component So it was some low frequency filtering with a running mean and we can see that after Let's say 1997 era 40 modulus drops down and this gives you very This gives you a decrease of the skill that you can find when comparing with this data set and Last we were thinking of what are the perspectives because we have these decal predictions. So what is expectable from these in the near future Unfortunately, not all of the simulations or not all the models gave the 2010 initialized simulation, but what we had what we had I Plotted here the in different lines, but I want you to focus on this last part. It's a different subset of Multimodal ensemble means but they are both showing a slight increase with respect to the years before this would be the 2011 to 2015 Means so with this I conclude The first idea is that we could have predicted part of the decadal variability of Sahelian rainfall using initialized simulations the Because the skill of the initialized is generally better than the historical one. However, this is quite model dependent not all models are skillful and Some for some models using the dynamics instead of the rainfall outputs can give you Increased skills. So we really recommend a two-fold approach if possible looking at both rainfall and dynamic variables and We should be cautious when using a low-level winds in in era 44 decadal purposes and just that the respective for next coming years is a bit of a slightly recovery of Sahelian rainfall with respect to the last Once so think I think you all for your attention and I accept any comments and questions you may have