 It is a pleasure to be here. I think this is my seventh time I think here. It is always nice to be here. I think you already heard a lot about sub-signal predictability and different ways of making sub-signal forecasts yesterday from Fredrik and Ajay Lois-Adrian. And today, together with my colleague Doisan from Jamstick, we are going to introduce something about the seasonal forecast. How we do seasonal forecast? The basis for the seasonal forecasts and also design some of the designs used for seasonal forecasts. I think each center has a different way of doing forecasts as we hear about that yesterday. ECMWF has very complicated way of making initializations and then do the forecasts both for sub-signal as well as seasonal. Our center we do it much simpler way. I think Doisan is going to introduce that later. Before that I will tell you something the basis on what we make the seasonal forecasts. I think this diagram already we saw several times. This shows different scales of processes from weather to climate change and as we as we know the predictability of weather systems is pretty high. The line here is showing you the predictability, skill of predictability and purposely I have not put the label here. It is quite qualitative. You can think like if this is pretty high the sub-signal is somewhere sitting here and seasonal is coming closer to weather predictions not at that stage yet. But we are doing quite well and mainly because we are doing well mainly because of some of the tropical climate phenomena and most important them is the Elino and the next one is the Indian Ocean Dipole and some other phenomena like here. Sub-tropical dipoles, the Atlantic Nino, the Marital Mode, NAO, AO and Southern Alginal Mode. As I said the most dominant one is the Elino Southern Oscillation and the predictability lies in the ocean as shown by this Birkenis feedback method. Actually his father started all those equations which is the basis for today's numerical weather prediction as well as climate predictions. He combined the Southern Oscillation which was discovered by Gilbert Walker trying to understand the droughts in India. He found that the sea oscillation between Tahiti and Darwin is linked to droughts and floods in India. Later Birkenis combined that with the already known fact that some years the Peruvian coast experience warmer than normal condition and then he linked that with the Southern Oscillation to say that this is part of the same phenomena coupled phenomena which can transit between two stages between Elino and La Nina through this normal condition. Normal condition and La Nina conditions are basically similar. La Nina is kind of strong of manifestation of the normal condition when you have strong easterlies and that can cause upwelling and cooling of the eastern tropical peak. The Elino is just the opposite of that and that happens because either easterlies become weak or we have strong easterlies from the west and that causes downwelling Kelvin waves, warm Kelvin waves which are coupled to those winds propagate eastward and call the Elino. The transition between these two phases are not clearly explained by Birkenis feedback process. Birkenis feedback process only says once we have a system like this, this can continue for some time and that gives us the predictability. Once we have some system like that, we know that after six months we will get the Elino and if we can observe that and put that into the model, we can clearly say when that transition phase will happen and when that will end into Elino. But the problem is then how we go back to La Nina and that's where the dynamical model comes to the picture and those can give us predictability for transition between Elino and La Nina. Why you want to learn Elino and La Nina? Because of the large influence we have from these tropical phenomenon like one is the Enso influence. You can see that Enso influences most part of eastern Pacific and also western Pacific, maritime continent, India, even Africa. And also we have teleconnexion to North America as well as South America. And because of that many people try to understand Elino and try to improve the predictability of Elino. Besides Elino we have also several basin scale phenomena. One of them is the Indian Ocean Dipole. Indian Ocean Dipole is kind of mirror image of Elino in which we have like warm anomaly happen on the western side of the basin. Usually eastern side is warm but some years the winds become weak and we have strong south easterlies and that can help upwelling on the eastern side and then downwelling Kelvin wave propagate westward and then we have warming on the western side to create this kind of dipole like structure in the in the SST anomaly is coupled to the atmosphere. Again because of those downwelling Kelvin waves and upwelling in eastern eastern Indian Ocean near Sumatra we can expect some kind of predictability because once we have a downwelling Kelvin wave we know that it takes some time to reach Sumali coast and that gives us the predictability and that's how the model gets the predictability. Again it has its influence. The western side is warm so that's why Kenya and East Africa gets lot of rain during positive Indian Ocean Dipole years. The maritime continent like Elino becomes dry and cool during positive Indian Ocean Dipole years. Opposite thing happens when you have a negative Indian Ocean Dipole. Of course also like Elino, the Indian Ocean Dipole also has teleconnexions going to the East Asia, Australia and South America. Good question. Usually the Elino as you know takes almost a year to transit from western Pacific to eastern Pacific and then coming back. The IOD is not so long it takes about three to six months. Yes, very much seasonal and very much locked to monsoon season actually. It starts somewhere between May and June when you have transition to some Indian summer monsoon and then it continues through the summer monsoon and through the fall and then towards November, December when the transition goes to the maritime continent monsoon the IOD gets terminated. Mostly coupled to the maritime continent winds. The southeast alleys are very important here. Like to maintain this upwelling near Sumatra we need these strong southeast alleys rates and that can continue until we have the winter monsoon. Once winter monsoon sets in the northerly comes in so that's why southeast alleys dies and the phenomenon terminates. Monsoon and this yes yes yes it is. And then later we identified this is our boss from Jamstek. So together with him we worked and identified this new not so new but we call it Elino modoki. Modoki is a Japanese word which means something which look like but not quite like the Elino. So it's a kind of fake Elino in other words. It looks like Elino. If you are on the western side like if you're in Japan and maritime continent this is kind of Elino. The east side is warm the west side is cold but if you are on the South American coast this is not Elino because during Elino year South American coast will become very warm and you expect a lot of fish mortality and many other problems. Of course we get a lot of rain but during Elino modoki years when central Pacific is warm eastern Pacific is not warm it's like a lanina condition. So that's why for people in South America and even North America it's like lanina but people in maritime continent and East Asia maybe it's like Elino. That's why you call it as Elino modoki. What happens is like on western side we have westerlies which kicks in the Kelvin waves, warm Kelvin waves but they don't propagate all the way to the eastern Pacific because on the eastern side the studies are still strong so they don't allow the warm Kelvin wave to go all the way to the eastern side. So that's why we have warming in the central Pacific. It's kind of Elino but doesn't go all the way to the eastern Pacific and this also has its own life cycle like three to six months not one year like Elino and its influence is different compared to Elino. The opposite phase is lanina modoki. This one? So this means we still have downwelling on both sides upwelling in the central Pacific and this means upwelling on eastern side western side but downwelling in the central Pacific. The thermocline gets depressed in the central Pacific not so much on the eastern side or western side. It's more like lanina condition but central Pacific is warm and we are interested to know about Elino modoki and its predictability because the influence is quite different particularly on the west coast of US. You can see here this is the climatological rain for Japan, China and US. This one is these three plots are related to Elino modoki and the bottom three plots are related to Elino. You can see that during Elino modoki years the west coast become very dry simply because it's like lanina. Eastern side is still cold so they don't get so much of rain whereas during Elino year because the eastern side is quite warm they get a lot of rain like Peruvian coast. So that's why it's very important to know whether it's Elino or Elino modoki for those regions and also southern Japan is very dry during Elino modoki years not so dry during Elino years and Angjidiver valley here gets a lot of rain during Elino modoki years not so much during Elino year. Besides those basin scale phenomena we also have some regional scale phenomenon like the Bengal and Elino for southern Africa many people know what is Bengal and Elino. Similarly recently we found a coastal phenomenon which we call the Ningalu Elino. Ningalu is a Ningalu reef here of west coast of Australia and in some years we observe a lot of coral bleaching and fish mortality and those are the year when you have warm anomalies here and those we call as Ningalu Nino years. Opposite thing happens you also have Ningalu Nino when you have strong cooling and good fish productions in west coast of Australia and that kind of regional phenomenon also very important for regional rainfall variability as well as fisheries and other societal activities. So that's why as you can see through all those climate phenomena it's very important to understand the air sea interaction, ocean atmosphere interaction and that is the basis for the predictability of seasonal climate forecasts. If you don't have dynamics ocean wave dynamics coupled to the atmosphere we don't get the potential predictability source from the observations to get into the model and predict them on seasonal time scales. This is what was done in early 80s to try to predict Elino. Now Elino is a kind of linear phenomenon in tropical Pacific where the heat content in the operation is coupled to the atmospheric wind and that coupled process moved from west to east to give you Elino and then return back to give you Lanina. For example if you for some reason you have Westerlies, strong Westerliebrost in western Pacific that can cause like the the downwelling Kelvin wave to propagate from west to east at the same time because of the wind stress call we get negative anomalies on both sides of the equator. The warm anomalies Kelvin wave anomalies go eastward at the same time cold anomalies move westward and these cold anomalies after some time gets reflected on the western boundaries and come back to to eastern coast to initiate the Lanina condition. That is the potential source of predictability in the Elino phenomena and that was exploited in 80s by simple coupled model and one of the well-known coupled model is Kenjabek model where one layer of the atmosphere was coupled to two layer of the ocean and it was very good in providing predictions on several seasons ahead to two years ahead which was then replaced by GCNs. GCNs are more complex we have the ocean model which has all its components and the ocean models are on usually spin up from initial state of rest and then and then we have the data assimilation some of the models don't have data assimilation like the syntax frontier model which will be presented by those and later is not doesn't have a data assimilation only takes SSD nosing and some models also do the atmospheric initialization for even seasonal forecasts but we don't do that in syntax frontier model it's not so essential for seasonal climate forecasts to initialize the atmosphere model because once you provide the SSD or the subsurface conditions to the ocean the ocean will feed back to the atmosphere and then atmosphere will adjust to that SSD to provide the seasonal climate predictions but some models are also some centers are also doing atmospheric initializations and once we initialize both then we coupled them and ocean model runs and then exchange the flux fluxes with the atmosphere and atmosphere returns back wind strength and heat flux heat fluxes and in that sense we provide the forecasts. There are several WMO centers which are giving now the forecasts these are the designated WMO centers Environment Canada we have I think several representative here from some of the centers Environment Canada and then the KMA from Sea of Beijing ECMWF and the Moscow Washington CP Tech is the Brazilian Center then the Pretoria South African Weather Service Poama Melbourne Bureau of Meteorology and JMA from Tokyo and Meteor France Toulouse. So all these centers are designated center for providing climate forecasts global producing centers for providing seasonal climate forecasts but to get that information we must be one of the member of this. Either it should be member of one of those GPCs or the regional centers or RCOPs or NMHS. At the moment the data is restricted to those users. These are some of the samples from those prediction centers. This is a 2014 Lino predictions by ECMWF and the Met Office. You can see some differences in the predictions but usually the predictions are close to each other. 2015 Lino predictions when it was predicted as you might be knowing 2015 was followed by 2014 Lino which is very rare. Two consecutive Lino's are very rare and most models were not getting the 2015 Lino until February of this year. So the triggering also is an important factor in the models and it's important to observe when the model captures the signal. Most model gets around February March the following years the coming Lino of that year. Now if you are not able to get the forecasts from those global producing centers you need not worry because there are several other centers like IRI, AFRA climate center even in Jamstack we provide the forecast online for users. Like here we are comparing the 2015-16 NSO outlooks from IRI plumes. There are several models available, model predictions available from the IRI site. APCC also has several models. Jamstack we have only one model and then we can compare that with ECMWF model. For the 2015-16 NSO outlooks we expect this NSO will terminate around January February this year and then return back to Lannina's days somewhere towards the December 2016 to January 2017. A little bit about our prediction system the Syntax Frontier model. You can see here the Syntax Frontier model actually originated from Italy. There is a center INGV not far away from here in Bologna and they started this model and then we borrowed that model and tried to improve it using the R-simulator. At that time this was the R-simulator number one computer in 2002 but now I think in top 500 and the atmosphere model is T106 with 19 level. The OSEN model equatorial enhanced to half a degree but usually it's around two degrees away from the equator and both models are coupled every two hours. We have a second version now. Doshan is going to present a little bit about that. Yes we have more number of labels now and uniform half-degree global. Before we go to the prediction the first thing we do we try to understand if the model is actually capturing the Elino, the Dipole and some of the tropical phenomena besides of course the extra tropical phenomenon. As you can see here the model was able to simulate the Elino, the Dipole together. This is one of the model years not necessarily related to any of the actual calendar years but nevertheless it was simulating the IOD and the Dipole very well. So we had some confidence but not all the Dipoles are related to Elinos. So that's why we wanted to see if the model is able to capture the Indian Ocean Dipoles together with Lanina. So this is one of the years when the model is capturing one of the Indian Ocean Dipole phenomenon together with Lanina. So we had some confidence that this model has basics to capture the Elino, Lanina and Dipoles on different phases. Then before we go into the predictions we try to do the verifications which is one of the very important process in seasonal climate forecasts. I think Andy Robertson is going to talk about that later next week. So this is the simple skill score of the Jamstex-Sintax Frontier model. At 3 months lead you can see that the model surface temperature, sea surface temperature predictability is quite high. The correlation is 0.9 between model predictions and the observations. At 6 month lead we still retain very high predictability in the tropical Pacific but started losing predictability in the Indian Ocean. At 9 month lead we have still good predictability in the tropical Pacific. So ends of predictability remains even at 9 months lead time but we are losing the predictability in the Indian Ocean. At 12 months lead we have almost no predictability in the Indian Ocean. We have quite high predictability in the tropical Pacific which means the model is pretty good to predict Elino one year ahead. Now the new model is able to predict even two years ahead. However the predictability is not so high in the Indian Ocean mainly because the Indian Ocean is having several scales of phenomena and their interaction is not so easily captured by the model. So that's why at longer lead time the predictability in the Indian Ocean is not so high. So as you heard from Adrian and Frederick yesterday we can improve the predictability. This is just a semi-multi-model ensemble. So I think he is going to talk about that. What we did in this like we had three different coupling physics. The same model but we employ three different coupling physics. So in essence they are quite different because the coupling is different. So that's why you call it semi-multi-model ensemble. Although the model is same we have different coupling physics for different ensemble members. I think he is going to talk more about that. So we also participated in multi-model ensembles to see if we have any improvements in predictability skills by combining our predictions with several other predictions. And this was done by Bin Wang in IPRC using the APCC, Apec Climate Center models. They had I think eight models. And from that again we find that they separated the June-July August predictability from December-January February predictability. And we can see that the predictability is quite high. I think this is three months lead quite high for both December-January-February and June-July-August. They had a marginal improvements at 12 months lead but basically the syntax of predictability remain quite high among all the model members. This is another example where we try to understand the predictability again, the skills of the predictability. The black line here is persistent line. So we think that this month's SST is same next month and the following months. These are the lead months 1 to 6 months ahead and these are the anomaly correlations. So you can see that the the syntax predictability remain quite high above 0.9 even at 6 months lead time much above the persistence level here. And most of the other models we are doing quite well as well. So most of the models are doing quite well to predict a lead six months ahead. And we find that if we can add the trends in the global warming, the global warming trend itself gives some predictability. The problem is to capture that trend, the problem is to capture that trend in the model is not so easy. So if you just take the observed trend and add that to the climate predictions, we find there is some improvement almost 0.2 correlation improved at one month's lead time. Compared to Elina predictability the dipole predictability is not so high. We also have two barriers. For Elina we have only one barrier spring barrier but for dipole we have both spring barrier as well as winter barrier. So that's why the predictability of dipole doesn't go very well beyond six months lead time. And once we have those predictability we try to apply some of those as Adrienne was mentioning about this. We had a project in South Africa, Babatunde is here I think. So he was part of that project. For the first two three years we tried to understand the climate variability in Southern Africa and then the predictability skills of syntax model and also the models available in South Africa. And in this phase starting in 2014 last year it will continue until 19. What we are trying to do is using those climate predictions we are trying to develop an early warning system for infectious diseases like malaria, diarrhea and pneumonia. Several centers are involved from both sides and most importantly we are going to use the vector model which Adrienne has developed. And this is some of the preliminary results as you can see here. The bottom figure here is showing you the malaria prevalences in South Africa. As you can see here the Limpopo province and the coastal belt gets lot of malaria incidences. Mostly those are important from Mozambique and the model was doing quite well actually to capture that. And we are very confident that this model can be used together with the climate model predictions to provide the early warning system. So we understand that models are doing well but models also have biases and models also are missing some of the the the transition phases of Elino and other climate phenomena. And to improve that we have to improve model resolutions we have to improve model physics. In fact some of the models are models are now trying to dissolve clouds so that we don't have to have cloud physics in the model. Going into 3 to 5 kilometer resolution so we can we can get rid of cloud physics. Besides all those things at the WCRP working group for seasonal to internal predictions we have taken up some of the projects to improve the model predictions and to reduce the model biases. The the first project led by Bill Merrifield from Environment Canada and also Mikhail Toltsky from Russian Atmospheric Sciences. So in that project what we are trying to do is like we know that most models are actually having problem for longer term predictions. When you go from multi-year to decayal predictions most models are problem to to to have the drift from from observed climate to model climate. And to avoid that drift they are trying to initialize the model at every one year or two years so that the model stays around the observed climatology. In addition to that for subsistional predictions we know that the model shock when we initialize the atmosphere and ocean model we have a lot of shock in the initial state. And Mikhail is trying to improve that initial shock by by improving the initialization schemes. And we hope that through that we can reduce some of the biases and shock so that we can improve predictability both for subsistional as well as decayal scale predictions. The second project we are taking now is the snow cover how to initialize snow in the models. In last 3-4 years we had a project called GLES Global Land Atmosphere Couple Experiments. In that the soil initialization was very important and many many centers participated in that to improve soil moisture initialization and that was quite successful. Now now what we are trying to do is to initialize snow both land snow as well as the sea ice and polar snow in the models and it is led by John from Seoul and even from Norway. Then of course we have the problem in teleconnection although we are getting the Elino dipole and other climate tropical climate phenomena very well the teleconnection from that those climate to extra tropics are not well captured by the models. So we have the third project to to understand that the the projection of tropical climate to extra tropics and the the atmospheric gross we are propagating around the globe taking those signal from one region to another region. To understand those variability in the model and the biases in the model this project is going to work on that. Besides those three projects as Adrienne was mentioning yesterday we also have the database climate system historical forecast project in this we are going to archive all the hind cast results at the moment we have already several models 1 2 3 4 5 or 15 models here and all these model results will be archived not only for from for the present state but also all the future states. So we have one of the archives where you can understand why one model went from bad state to a good state what kind of model physics they change over the time so that we can we can replicate those things in other models that's one of the ideas where why we are archiving all the data. At the moment most of the monthly data are available and now they are trying to archive the daily data. This will be a good point if you want to understand or try to to to work with the seasonal forecasts before going to forecast you can try to understand the the the model physics using those models. I'll stop there and maybe Doison will take over from here. Any questions to me? You can ask up. Okay good okay I will talk about the syntax of seasonal prediction system. So first three slide I'd like to run over the basic information about the seasonal prediction so you know that there's a difference between seasonal prediction and weather prediction so seasonal prediction is a kind of climate prediction so it's a target is to predict the climate so climate is a statistic of weather for example the monthly average as a number of any day in a month and the number of typhoon in season so so weather prediction is try to predict the weather itself a few days later and so this is an inter-season time scale this is important so S2S project but my talk is a target is a seasonal so target is a monthly average or three months average and read time is three to sometimes up to nine months read time so as the beheaders are already explained the potential source of seasonal predictability is mainly due to answer predictions so because contribute for success of seasonal predictions the contribution from ocean is important because ocean variability is very slow relativity atmosphere and the ocean have a large heat content relativity atmosphere those character of oceans can work as a seasonal potential source of seasonal predictability in particular tropical ocean is crucial because topical ocean is very warm and it can drive the global atmospheric circulation so here I would like to introduce the two pioneering pioneering work by Dr. B. Hackness so he found the AC couple phenomena in the tropical Pacific and also in 1964 and three years later he found the teleconnection from tropics to meet Russia this is a fundamental theory of the seasonal predictions so so we need to the numerical numerical seasonal prediction system should be based on ocean atmosphere couple model so the numerical weather prediction mostly employed the standalone atmospheric model because on the assumption that the oceans do not change in the relatively short prediction period for example one week however for prediction of the answer and it's induced seasonal anomaly we need to application on ocean atmosphere couple model so this is a schematic diagram for the numerical prediction system as a first step we know the current we have to know the current state of the climate in particular ocean state is important for seasonal prediction so we need to the observational system for example morning buoy or satellite observation and those information should be assimilated in the model this is in other word it is called the initialization and also we will conduct the numerical integration by a couple GCM using the supercomputer so by the way this is our new supercomputer called the action rate so we have a basically three step for the numerical seasonal prediction the question is which step is the most critical for prediction scale actually this here is the shows the relative reduction in SST error SST forecast error over the NINIO 3 actually NINIO 3 is a L NINIO index by the ECMWF seasonal prediction system so ECMWF successful improve the seasonal prediction scale for L NINIO actually total gain is shows by yellow column so about the 35% relative to the probably 1996 operational original one and this red column is a contribution from the ocean initialization and this blue column shows the contribution from model development so this figure shows the model development and ocean initialization are equally important for improving the seasonal prediction scale so I would like to focus on our seasonal prediction system index F so Bayhara already shows this slide so this model is developed at JAMstack under the use of collaboration so here I would like to introduce how we developing this model and its initialization scheme so first about the model development so actually the in classical method modeler try to tune to the individual uncoupled GCM separately so tune the atmospheric model and change the ocean model and try to couple between between two that's in contrast to the classical method our model development we tune the it directly by improving the AC couple physics so we focus the potential impact of the strong ocean current on wind stress actually this is important to simulation of the climatology and the acid viability so this figure shows the linearly regressed SST and the surface wind anomalies map on the linear theory SST index and this figure is from observation so we can find to the El Nino structure right but and this is a control run by the syntax F but this is the effect of an ocean surface current on wind stress totally neglected in the control run and this F C P L run in this one the surface wind stress is calculated by like this so this is a density of air and CD is a drug coefficient and we are is a surface wind speed and we were the surface ocean current so this key is considering the impact of the ocean surface current on wind stress so you know this model and focusing on the warm water pool region and this is actually the warmest ocean in the world and it's very important for their connection so if focusing this area F C P L run it's much similar to the real ocean right so this is the one example of our model development and also for about the initializations so we use the SST matching scheme and this scheme is a syntax F model is a couple model and OGCM SST is strongly matched towards the observed SST in a couple mode so in a couple mode the AGCM forced by such generated OGCM SST and then the OGCM forced by the AGCM simulated front but with matching to observed SST this is the SST matching scheme and this is one of the simplest approach for the initialization but it can provide a compatible initial condition between the atmosphere and ocean this compatible initial condition is very important because of this the balance between the ocean and atmosphere can reduce the initial shock you are in forecast and also this simple schemes can capture the elliptical subsurface ocean structure in the tropical Pacific this figure shows the Hockmiller diagram you know the time devolutions and longitude and you know the shows the 20 degree iso-sum steps and normally along the equatorial Pacific and left column is from the soda a soda is the ocean simulation data is including the subsurface information but left column is the SST matching run with syntax F so you know that we are just using the SST information so you know this is very similar between two so we can say to the SST matching our SST matching run can successfully capture the subsurface structure in the tropical Pacific actually this success depend on the performance of the AGCM and OGCM and also how do we generate ensemble member in our system so yesterday Tomkinson already explains our necessary of the ensemble member because the atmosphere ocean couple system involves a strong non-linearity so variation in initial condition and the physical schemes needed to diverse solution so need ensemble prediction to reduce the prediction uncertainty so we generate the ensemble member associated with the different initial condition and different physical schemes so I have three strengths of SST matching initialization and we have a three coupling physics scheme for considering the effect of the ocean surface current on wind stress so three times three times we have a totally nine ensemble member employed for seasonal prediction and the initiated every month in 1982 to present okay this is a summary of our system and this shows a schematic here if we start the prediction from May 1st 2050 so we have a continuous SST matching run with a syntax F and if we start the prediction from May 1st we can use this to restart fire from the SST matching run and we can run the freedom actually this is a focus run and running to the up to the target season okay so how skillful is our system for El Nino prediction so this is a time series of NINI 3.4 NINI 3.4 is an index of El Nino it defines the sea surface temperature normally average the in this region and this black line shows the time series of NINI 3.4 and generally speaking when the NINI 3.4 beyond the 0.5 degree we can say the El Nino is occurring so we have a several El Nino event but this event is the strongest event in El Nino in 1997 and 1998 actually WMO estimates the 34 billion US dollar loss in the world due to this the strongest El Nino so as a prediction it's very important from not from scientific viewpoint but also the economical and societal viewpoint so first I'll try to focus at this event and show the prediction scale and this is just focusing on the NINI 3.4 in 1997 to 1999 to 1998 so black observations and this blue line shows our prediction from the 1997 April 1st initialization here and this red line is an ensemble mean so when the model initialize at the point the other point so you know the tropical Pacific was almost neutral state when the prediction started at that point our model success or predict of El Nino occurrence actually the all nine member is beyond the 0.5 degree so we can say the our model success or predict El Nino occurrence from the neutral states however unfortunately the amplitude is rather underestimated this one oh yeah it's a prediction that is the April 1st and prediction end by match so it's about 12 months predictions okay and and the next so when the model sorry initialized okay sorry at that point actually at that point we El tropical Pacific was already El Nino like state but actually this is a moderate El Nino state when the model initialize at that point you know our nine ensemble member predict those El Nino like condition enhanced in the coming six months so actually I already said the 1997 event was the strongest event so Nino 3.4 beyond the about the 2 degree so now is some research of course this extremely strong El Nino as a super El Nino so I can say the our model is successful predicted the occurrence of the extreme strong El Nino from a moderate El Nino state okay this is for the 1997 and 1998 event so next question the how skillful is our model for other ends event to answer this question I'd like to explain the first is about the three months read prediction line and the next few years so this is also Nino 3.4 index and this green line is the ensemble mean prediction from the 1997 April 1st initialization right and so April match April May June and just a three month read prediction I brought it this red cycle and also this purple line the prediction from the May 1st 1997 so May June July I brought it the three month predictions here and also blue green blue color shows the June first initialization so I also brought it to the three month read prediction and this connected this wine and this wet lines cause a three month read prediction and same as this six month read prediction is shown by blue line so this is a fear of the time series of Nino 3.4 black is observation and the red is our three month read prediction and blue line is a six month read prediction so after first grant so it looks similar right and our model is very skillful to predict El Nino actually correlation between observation and three month read prediction is close to the point nine and also observation correlation between observation and the blue six month this time also beyond the point eight five no it's very high however not good for the timing of initiation for example here we have some failure for initiation of El Nino right and also we have failure of the termination phase of El Nino so our system have a problem to predict the timing of initiation and termination okay anyway our system is very skillful for answer prediction relative to other model so this figure is also already Bayhadasan shows but I will explain the detail so this is a correlation between observation and prediction so wine is a perfect point five is a relatively low scale and X axis shows the forecast read month from one month to six months and this line is a SSD persistence line so this shows a lagout correlation and this is a each model prediction system and actually this is just for the Nino 3.4 index in 98 to 2004 so that line is our model so our model is shows a very good scale relative to other model and question is this is up to six months how about predictions beyond six months this time so actually our system is a skillful for the two-year answer prediction this figure also similar to previous one but up to 24 months read time so this is also SSD persistence and this is a nine ensemble members prediction by syntax F so up to 24 months still it's close to correlation close between the observation and prediction close to point six so this is a skillful from a statistical analysis significance okay this is an ensemble member yeah each member yeah each member is below the mean okay so this is a each ensemble members prediction and this black line is a bit average in the nine ensemble and calculate the correlation so sometimes there's a ensemble means can reduce the uncertainty of predictions okay so by the way how about this year's answer predictions this is also Nino 3.4 and blue is our observation of the DCS so now is a strong El Nino is occurring actually this year's El Nino is almost similar to the strongest El Nino in 1997 1998 and actually this wet line is a prediction from the May 1st initialization this year so our mother nicely predicted the super El Nino go in this winter from the May 1st and now is a way we are here so we have a strong El Nino now and also what will happen next two years so I we predicted the this strong El Nino may turn to a La Nina in next year so this year may be similar to the strongest El Nino in 1997 1998 other time also strong El Nino as quickly phase transition and to turn to the La Nina I sorry model predict to the probably this year's El Nino also phase transition to La Nina in next year and also this is an example for our prediction about the coming winter coming December January February average by our model and this is a sea surface temperature anomaly and we can find to the very strong El Nino we would pass it really persist in the coming winter and this is a surface air temperature anomaly and most of the world we experience the warmer the normal condition in the coming winter and this is the rainfall predictions and green colors of the wetter than normal and gray three the brown color shows a drier than normal so probably the is a western euro so Western US will be drier than normal but for Brazil is a will be drier than normal due to the mainly due to the El Nino okay so this is my summary so syntax F1 is now is a very skillful to predict El Nino however I show the there is a room of improvement for the timing of initiation and termination prediction so how should we improve our system so I already explained the model development and also initialization I call important for that so strategy one I'm another developing the model the previous one version of the syntax F1 and I'm not preparing the second version so updating the model updating version of the model and enhance the vertical resolution in atmosphere model and enhance the horizontal resolution in ocean model and include the sea ice model so I'm no writing paper about that and but I want to show the preliminary result about the syntax F2 so this is also prediction scale actually based on the correlation for the December January February averaged between observation and prediction from November first initialization so you know prediction from November first initialization and target is a December January February so it's about read month is about the two or three months read time and left column shows prediction scale for two meter air temperature normally and right column for the prediction scale for rainfall anomaly and upper panel for the syntax F1 and lower panel syntax F2 so we can find some improvement in the southern hemisphere by syntax F2 and also some improvements of the rainfall prediction in syntax F2 okay so about and also strategy to about the ocean initialization I said the success of the SST matching initialization scheme is that over a region where ocean variability is strongly constrained by the couple ash interaction so actually this figure is from the Arun Kumar papers in 2013 shows the local simultaneous SST and rainfall anomaly correlations and point by point correlation and it so correlations for each season and for example if the correlation between the SST and rainfall anomaly is highly positive that means a warm SST is correlated with a more rainfall so that means a warm or SST drives a conviction as a more rainfall that situation there's a ocean drives atmosphere and this is good for SST matching so for example in tropical Pacific we can hide we can show the high positive correlation between SST and rainfall so that means that tropical Pacific AC couple is strong however for Indian ocean we can find the negative correlation right so negative correlation means a cold SST is correlated with a more rainfall because a more rainfall is less sunshine it's cool sea surface temperature so that means the atmosphere drives the ocean temperature so in that case is a atmosphere is driver so and the AC couple feedback is weak so SST matching is good for the capturing the subsurface structure in tropical Pacific but not enough for the tropical Indian ocean and also tropical Atlantic this is the one problem of SST matching system and another one that this scheme cannot resolve high-frequency variability related to the mud injury and oscillation or a westerly wind burst etc so this event can have strong implications for the timing of initiation and termination of answer event so we have to include those information of a system so I know that developing the new scheme based on SST matching that's using the three lever collections and collaborate it with Dr. Andrea Strollt, OCMCC and Bologna. Okay so this is a two strategy and actually we have another way for the improving the seasonal prediction that's a strategy 3 to discovery of new potential source of seasonal predictability I already explained the traditional approach for seasonal prediction to predict the tropical climate variation and then to predict their pair connection to the mid-ratch that's another possible way discovery of other potential source of predictability for example regional air-sea couple phenomena in mid-ratch due because mid-ratch due roughly speaking mid-ratch due is an atmosphere-derived ocean but sometimes in mid-ratch due we have a regional air-sea couple field market and also soil moisture overland and stratosphere, snow cover, sea ice and ozone okay many other we have other possible potential source of seasonal predictability. Here I would like to introduce the one example for the regional air-sea couple phenomena because I recently published a two paper about that so we found to the new potential source of seasonal prediction for West Australia this is named the Ningal Minyo so actually after late 1990s global warming and the negative phase of the inter-decadal Pacific Ocean can let the warming oceans of the west coast of Australia. So absolute SST values is warm-ups and here is actually the mid-ratch due but it's kind of the topics region and new climate phenomena was born that's the Ningal Minyo. This is very similar to the Pacific El Minyo but this is a close off the west coast of Australia and it directly influenced on the rainfall over West Australia. So actually for Ningal Minyos so Kataokaetowal is a pioneering park for about that he calculated the EOF first mode of the year-to-year SST variability and defines the index for Ningal Minyos as an SST anomaly averaged in this region and this is a time series of the Ningal Minyos index and this is a one degree line so from 1984 to the late 1990s so no event beyond the one degree for the Ningal Minyos index but the recent 50 years we have a four event beyond the one degree line right so that means in recent 50 years we have experienced a frequent occurrence of Ningal Minyos. Actually our model is nicely predicted this one because the duties are linear prediction skills and this is published by my paper and also this is important for the potential source of seasonal predictability of western Australia. I also calculated the simultaneous correlation between sea surface temperature and the rainfall. This is a previous period and this is a recent 50 years period. In the period one no correlation between the local sea surface temperature and the local rainfall. However in recent 50 years we have a high correlation between sea surface temperature and rainfall so roughly speaking we can say to the local SST can drive local rainfall in the recent 50 years and also I checked through the prediction scale for rainfall. Actually this is a correlation coefficient between observation and prediction from September 1st. Sorry target is the austral winter, austral summer. This is the rainy season in Australia and our model have no skill to predict the rainfall, seasonal rainfall over western Australia in the period one. Recently our model shows enhancement of seasonal prediction skills of rainfall over west australia so that's why the Ningal Minyos can work as a new potential source of seasonal prediction for rainfall over western Australia. So detail is published in these papers. Okay so this is a take home message. Our system is very skillful but still it's not good predicting the timing of initiation and termination of the Ningal. So how should we improve this system? The model development and ocean initialization are equally important and also another way to the discovery of new potential source of seasonal predictability. Thank you.