 Today I am going to continue the topic monsoon prediction and what we are going to focus on today is actually the prediction with dynamical models. So called dynamical models, these are models based on the laws of physics. So I am going to talk on predicting the Indian summer monsoon rainfall. So again we focus on the seasonal time scale June to September and the spatial scale is all India. So average Indian rainfall for June to September for that specific season that is what prediction of that is what I am going to talk about today. Now the logical approach to prediction of the monsoon rainfall is by integration of complex models of the atmosphere or the coupled ocean atmosphere system based on equations governing fluids in a rotating system. It is important to note that the breakthroughs in seasonal forecasting over the tropics have come from the phenomenal progress since the 80s in understanding the physics of the El Nino Southern Oscillation phenomena, ENSO phenomena which is the dominant signal of inter-annual variation of the coupled atmosphere ocean system over the Pacific. So there has been phenomenal progress in understanding of ENSO since the 80s, elucidation of the nature of ENSO, unraveling of the underlying mechanisms led to the development of models to a level at which they could realistically simulate the phenomena and its impacts on the climate of different regions. As we have seen already it is important to understand ENSO not only if you are interested about the climate on the Pacific but ENSO does have impact on regional climates over large parts of the tropics other than the Pacific region and so it is very important to understand it. And with the kind of insights that were gained in the advances made in understanding of the nature of ENSO, the models could be developed to a level at which not only was it possible to simulate the phenomena realistically but also its impacts on the climate of different regions. Now we already have seen that the year to year variation of the Indian monsoon ISMR is related to ENSO. So given the links between the Indian monsoon and ENSO, it was expected that this revolution that occurred in atmospheric oceanic sciences of it being possible to predict ENSO would lead to models giving better predictions of the inter-annual variation of ISMR. So it was expected that it would be possible to simulate the inter-annual variation of ISMR with atmospheric general circulation models when the observed SSDs are specified as boundary condition. However, the results of several such studies suggest that the problem remains a challenging one. It is not as if solving the ENSO problem has also led to the solution of the monsoon problem, there is more to monsoon than ENSO. So consider first the inter-annual variation of ISMR during the satellite era and the link with ENSO. So let me just remind you of the nature of the inter-annual variation of the monsoon. This is only from 79 onwards. Remember that droughts mean deficit more than 10% or one standard deviation is the same thing. Excess is above normal rainfall which is of which the anomaly is larger than 10% or one standard deviation. So during this era then that I have shown from 79 to 2009 we have seen several droughts. This is 79 itself was a drought then 80 to 85, 86, 87 and then we had a reasonably good period here but it ended with frequent droughts 2002, 2004 and 2009. The excess rainfall years are in fact relatively few there is 83, 88 and also 94. Now we will see that 94 and 97 as well as 83 are very special years in this talk. I just want to remind you that 97 was the largest El Nino of the century. So now what we see here is plotted again the same ISMR. This is the same color combination used so red means droughts, green means excess rainfall and in between is normal rainfall, black is normal rainfall. Now next to the stick with ISMR is a stick showing whether ENSO is favorable or not. So it is when it is negative ENSO is unfavorable, when it is positive ENSO is very favorable and actually if we have a very highly favorable condition then that is called a lanina and you see an example here this blue stick means it is a lanina and green means it is an excess. Excess associated with a lanina and El Nino would be orange this is again departure of the ENSO index beyond one standard deviation deficit. So this is highly unfavorable situation these are the El Ninos. Now the link let us look at the known link between ISMR and ENSO. Now actually 79 ENSO was unfavorable but not all that much 82 you had an El Nino event highly unfavorable ENSO and a drought. So this is as expected but then 87 also we had an El Nino but and a strong drought 88 is a case where you had excess rain associated with lanina. So these three years demonstrate the known link between ENSO and monsoon that there is a high propensity for droughts during El Nino years high propensity for excess rainfall during lanina years. So as expected El Ninos of 82 87 are droughts lanina is excess however why was 94 an excess you look at 94 ENSO is unfavorable it is negative and yet we got excess rain and you look at 97 where you have the strongest El Nino of the century you can see how large the ENSO signal is and yet the monsoon rainfall was normal. However then later on what happens is in 2002 with a much weaker El Nino you are getting large deficit here big drought here and here also it is associated with an El Nino. So why was 94 an excess despite of it being a weak El Nino it was not an El Nino the anomaly was not minus 1 but it was still an excess rainfall season and why was 97 a normal monsoon. If we look at normalized the SMR anomaly versus ENSO index again this is something we have seen before and what you find is that ENSO index is sufficiently favorable. So this is about 0.8 I think then you have no droughts and when it is sufficiently unfavorable there are no excess rainfall seasons. So one can get a one sided prediction provided the ENSO index is beyond this or is in this range but in between these two there are lots of years where you have droughts as well as excess rainfall season and with ENSO index alone you cannot say anything about these extremes here. Now as expected El Ninos of 82, 87 and droughts of La Nina see you see 88 is La Nina and excess rainfall season 80 to 87 are El Ninos and this but these are the years we want to understand 94 which is an excess despite El Nino being unfavorable and 97 being a normal year despite El Nino being so strong. Now I must remind you that we are going to focus on extremes for reasons I mentioned in the last lecture it is most important to predict the extremes of monsoon rainfall much more so than the fluctuations within the normal range. Now what you see here is nature of the impact of ENSO of Equino and the relationship of ISMR we already have seen that 94 and 97 what happened was that they were straw they were years with strong Equino Equatorial Indian Ocean Oscillation and that is the second mode which is important in determining ISMR this is something we have seen and so how what is the nature of impact. So if you have an ENSO then suppose this is a La Nina case where the OLR correlation with OLR of ENSO index is shown and what you see is during La Nina the entire region will have lot of rain except for this headway region whereas the impact of equine is different equine implies excess rainfall here this is the positive phase of equine corresponding to La Nina and suppression of convection or rainfall here and this is associated with high rainfall over the Indian region. This is seen here if you look at correlation of ISMR with OLR it is highly correlated with rainfall over the western Equatorial Indian Ocean and negatively correlated with rainfall over eastern Equatorial Indian Ocean and this is of course the ENSO link where you have a negative correlation between rainfall over the central Pacific and rainfall over the Indian region. So if we look at both the indices equine and ENSO we have seen already that there is a clean separation and the reason that 94 was an excess is because equine was so large and positive and the reason 97 in fact is a normal year again because equine is large and positive although Lino ESO highly unfavorable we have seen that in the phase space of both these indices the droughts and excess monsoon years are well separated. So if the prediction is for a point for the forthcoming season implies that it is a point somewhere below this line here then we can say that there is the possibility of excess rainfall years is 0 the probability of excess rainfall years is 0 if the point is below the line above the line the probability of droughts is 0 this is from historical records. Now let us this is inter-annual variation as we understand it now let us look at simulation of inter-annual variation with atmospheric models. Analysis of the simulation for the years 79 to 95 by 20 state of the art atmospheric general circulation model which we call AGCMs was organized under the atmospheric model inter-comparison project AMIP which showed that almost all the models simulated the correct sign of the ISMR anomaly in 1988. Now remember the models have still to progress considerably so the first question we ask like we asked of the empirical models I discussed last time is at least the sign of the ISMR anomaly captured when observed ISMR is an extreme in other words when observed ISMR is an excess are the models at least predicting a positive ISMR anomaly when the observed ISMR is a drought are the models at least predicting a negative ISMR anomaly. So this is the question we asked of the AMIP and what we find is that for 1988 which I will remind you was a year associated with La Nina excess rainfall year associated with La Nina black is the observed ISMR anomaly and this is these are all the different models and you can see but for one model all the models have got the anomaly sign right the amplitude may not be right at all and may be exaggerated in some models but at least most of the models vast majority of the models can get the sign of the ISMR anomaly right when the AGCMs are forced by observed SST. However a vast majority of the models fail to capture the anomaly for the excess monsoon rainfall season of 94 during which ENSO was unfavorable as we have seen. So this is the excess monsoon season of 94 and you see almost all the models are getting in fact negative anomalies here negative ISMR anomalies. So vast majority of the models have failed to capture the correct sign in this case also same thing happened for 97 none of the models participating in another experiment which was a Clivar monsoon GCM in the comparison project could simulate realistically the observed response of the 97 El Nino event. Now occurrence of large errors only for a few years so what are we seeing for some years like 82, 87, 88 there are not that many large errors most of the models are at least able to get the sign of the anomaly right but there are huge errors only for some years like 94 and 97. Now what does this suggest occurrence of large errors only for a few years suggests that the low skill in the simulation of interannual variation of the monsoon arises from a poor simulation of an important facet of phenomena and or the teleconnections rather than omission of an important process such as coupling because if the models were not good because they had an important process wrong could be the parametrization of clouds or boundary layer or whatever then you do not expect this kind of a bias in error where the errors tend to occur only in some years. Note that 94 and 97 seasons are characterized by positive phase of equino associated with strong positive IOD events. The anomalies over the equatorial Indian ocean associated with the positive phase of equino is simulated by the IGCMs forced with observed SSD. So actually what happens is since we are forcing with observed SSD the local response which is the equino equatorial Indian ocean oscillation is simulated accurately by the IGCMs however for some reason they are not able to get the link with equino. So actually it was suggested by Gargill et al after looking at the experiment in India called SPIM seasonal prediction of the Indian monsoon in which 5 models in the country were run for a period 20 year period beginning with 85. What this project tested was a hypothesis put forth earlier by Gargill et al which said that the poor skill in simulation of monsoon equino link leads to the poor skill in of IGCMs in simulation of interannual variation. So this was one thing that somehow the link is not being captured although the local response is okay. Then the question was why are IGCMs not able to simulate the link between the Indian summer monsoon and equino. Now there are two possibilities one is that models are inherently incapable of simulating the link with equino and I must tell you that initially when people began to look at how good was the monsoon prediction in models with IGCMs atmospheric general circulation model they were not able to simulate the link with ENSO et al. And what happened was under an international project called MONEG people worked on the models to get the 87 El Nino drought and 88 La Nina excess monsoon right in the models. So development was done on the models to get the link with ENSO right. Now so is it that the models were incapable inherently incapable of simulating the link with equino or that models are not inherently incapable of simulating the link. But are unrealistically sensitive to the anomalies over the pacific that is they have become in the model monsoon is much more of a slave of ENSO. But in reality we have seen that the monsoon anomalies can even be of opposite side than that suggested by the ENSO phase. So it is possible that they are not sufficiently sensitive to equino and are unrealistically sensitive to ENSO. So these are the two possibilities that arise and the latter hypothesis is supported by results of SPIM this is the project that I mentioned this was an Indian project seasonal prediction of the Indian monsoon a national inter comparison experiment with 5 AGCMs used in the country for seasonal prediction for 84 to 2004. Now when the models were forced by weaker SST anomalies than observed for 94. So in SPIM two experiments were done firstly for all the 20 years observed SST was specified as a forcing for the AGCMs. But there was another experiment done in which for few years the models were forced by April SST anomalies rather than the real anomalies. Now for 94 when this was done what it implied is that it was forced by weaker SST anomalies and actually weaker El Nino over the Pacific and when that was done the two best models could simulate the link with equino and a positive ISMR anomaly. So this leads to the conclusion that it is not that models are inherently incapable of actually simulating the link with equino but rather they are too sensitive to ENSO. You know when the effect of ENSO was artificially weakened by replacing the SST anomalies with weaker ones then the models could actually simulate the link with equino. So far we have been talking of predictions with atmospheric general circulation models but as I mentioned before for prediction with atmospheric general circulation models you have to specify the boundary conditions in particular the sea surface temperature which means that predictions with atmospheric general circulation models involve two steps these are called two tire prediction. One is that you have to predict the SST by some way and specify that SST predicted SST to run the AGCMs to predict whatever you want to predict in this case the monsoon. Now of course there may be errors in the prediction of SST itself and it is to see what would happen to the atmospheric models if there were no errors in prediction of SST that experiments like AMIP were designed where there was if you wish a perfect prediction of SST the observed SST was specified to run AGCMs. But in real life if we are going to use AGCMs to predict then we have to predict SST as well of course a more natural way of predicting the monsoon with models would be to use coupled ocean atmosphere models where from an initial condition the ocean evolves and the atmosphere also evolves and the boundary conditions for the atmosphere come from the ocean which is also evolving in the model. So to assess the skill of prediction by atmospheric or coupled models what is done is retrospective forecast which used to be called hindcasts are generated. Let me explain what these are see for each year or season models are you run using conditions which would have been available for prediction for that year or season. So even if you are doing the experiment in 2005 as the experiment I will talk about was done what you do is if you want to make a 30 year kind of run then 30 years run then for any specific year say 1994 you take the initial conditions as were available at the time of forecast for 94 season it could be first May for example and you take the conditions for the state of the ocean as well as state of the atmosphere and then integrate the model to generate the what will happen in the forthcoming monsoon season. So in a way these are forecast which are done retrospectively so they are called retrospective forecast somewhat misleading name for them is hindcast. Hindcast seems to suggest that time is going in the reverse direction that is not true at all. In fact it is that forecasts are done long after the season is over but using conditions which were available would have been available at the time of forecast to assess the model and these are called retrospective forecast. Now an ensemble of such runs is generally generated by perturbations of the initial condition and that is because we know that atmospheric models are sensitive to initial conditions and therefore one would like to run them for a whole host of initial conditions and average over in those initial conditions to remove the chaotic element that comes in because of the sensitivity to initial condition. So generally an ensemble of such runs is generated by perturbation of initial condition. Now the skill of the model is derived from the ensemble average forecast. Such retrospective predictions were generated for 1960 to 2005 for 5 state of the art models from Europe and I am talking of coupled models, coupled ocean atmosphere models under a project called ensembles and for 82 to 2009 for two versions of their model by NSEP, National Centre for Environmental Prediction USA. I discuss now assessment of the skill of predictions of ISMR by the models again focusing on the extremes to get an idea of how good are the predictions by these models. What is the skill of these models present state of the art model in predicting the extremes of the monsoon. But we have seen that extremes of the monsoon in real life are very much related to two phenomena Enso and Equino. So I will also discuss the prediction of these two modes Enso and Equino and their teleconnection with ISMR in the models. The mean JJAS rainfall patterns from the 5 models from the ensembles project and the observations are shown in the next slide and from CFS 1 and CFS 2 in the following slide. So these are the mean patterns and what you see here is CMAP corresponds to observations. These are the observations and these are 5 models UKMET office, Meteor France, CMCWF, this is the European Centre model and IFMGOMR. And so you can see this is reality and in reality this is the kind of rainfall you get the scale is given here. And this is the kind of rainfall that is simulated by the model by enlarge the patterns over the Pacific look ok. But you also see that for some models like UKMET office and Meteor France there is hardly any rain over India. You also see that UKMET office gets much more rain over western equatorial Indian ocean than eastern one whereas reality is the opposite. ECMWF model is not bad most of the other models do get some rain over India. But you can see that in Meteor France you get much more rain here this is north of the Himalayas. So there are differences from model to model but on the whole they are basic features of the mean rainfall pattern over the Indian region and equatorial Indian ocean are reasonably well captured by the models. So those were the ensemble models these are two versions of the NCEP model CFS1 and CFS2 and these are the observations. And again you will see that the second version has a dry bias there is very little rain over India in this and it also seems to have in comparison with this some differences over the Pacific. But the major difference is over the Indian region where you have a dry bias here which is relative to the observations here. As I mentioned for some models such as Meteor France and UKMOT there is hardly any rainfall over India. The major rain belt in the mean rainfall pattern of Meteor France over 60 to 90 and UKMOT is over 50 to 80 is over the equatorial ocean rather than the Indian ocean region. So what you will see here is that the major rain belt is here and not over Indian region very much so in Meteor France also. You see the rain belt is extending all the way and it is just over the equatorial Indian ocean. It really does not have any rain here there is some rain over the Himalayas here. So there is a major shift in the rain belt. Now for the UKMOT model the rainfall as I mentioned pattern over the equatorial Indian ocean has maximum rain over the western part and hardly any over the eastern part which is opposite of what is observed. Now let us again look at observations. This is from 60 onwards because you remember ensemble runs were made from 1960 onwards and these are the excess monsoon years and these are the droughts here. Now for this we have taken mean rainfall for the same period 60 to 2009 so that what we take with the model is comparable because for model runs we have from 60 to 2005 so we have taken mean rainfall over a similar period here and what you get is these are the excess monsoon years. In fact 61 is the year with highest rainfall recorded and then we have of course 70 which is an excess and 75 which was a lanina and an excess. We have 83 which was an excess, 88 which was an excess and 94 which was an excess and then we have a whole lot of droughts as well. So prediction of the Indian monsoon rainfall on a seasonal scale is important but challenging problem in dynamical models. On the whole the skill of the coupled model studied here in predicting the extremes appears to be reasonable with the models being able to predict at least the sign of the ISMR anomaly for a majority of ISMR extremes. Now I must mention this was not true about 5 years before this experiment when models could not even get the sign of most of the extremes right and the statistics is like this. If we consider all the extremes observed over the period in which in the period over which the retrospective predictions are available the skill of prediction of ISMR extremes seems to be reasonable for several models. Consider first the predictions by the 5 models from ensembles. This is a set run by the European centre project ensembles of the 9 droughts during 61 to 2005 negative ISMR anomaly was predicted by for 8 seasons by 2 models ECMWF and UKMD office and for 7 seasons by the other 3 models. So this is a pretty good record and for the 7 excess monsoon season positive ISMR anomaly was predicted in 6 seasons by 3 models and only in 4 seasons by the other 2. So the success rate is not bad for 82 to 2009 for which ensembles model predictions are available CFS 1 and CFS 2 predicted negative ISMR anomaly for 5 out of 6 droughts whereas CFS 1 predicted positive anomaly for 4 out of 5 excess rainfall seasons CFS 2 predicted positive anomaly only for 3 out of 5. So in this sense CFS 2 seems slightly worse than CFS 1 it is also worse in terms of the dry bias. Now for a few extremes of monsoon rainfall and the special case of 97 the observed ISMR anomaly and that predicted by the 5 models of ensembles we will see here and see this is 61 this was the year with maximum rain and this is organized you know with ISMR decreasing as you go. So next was 88 you see in 61 all 5 models got the sign right in 88 also they all got right but look at 83 in 83 all the models simulated negative ISMR anomaly but actually it was an excess. Now here most models got it ok there are two models which got the sign wrong but the amplitude is very very small. So 94 has certainly improved from the AGCM experience 97 on the other hand you see huge deficits simulated predicted by the models whereas actually it was normal. Again for the other droughts 82 87 and 2002 most models seem to be getting the sign right. Again huge discrepancies between model predictions and observations in the year 83 and 97 as far as ensembles is concerned. Interestingly the same years have very large errors also for CFS 1 and CFS 2 what you see is CFS 1 are blue but with initial conditions in April and May this is CFS 2 and remember the simulation was from 1980 onwards so we do not get the earlier excess and so on. But again you see for years such as 94 CFS 1 is doing fine but CFS 2 is not but again you see this is the year 88 in which things are doing things are actually pretty good as far as CFS 1 and 2 are concerned in 88 83 and 97 are again two culprits which will completely spoil the skill of the model models are somehow not able to get the skill right. So the most remarkable feature is the coherence in the signs of ISMR anomalies predicted by the different models for several years. So there is a large coherence and even here when they get all get it wrong there is a coherence between the models only thing is that the sign is wrong when we compare it with reality. So there is a coherence between models which is an interesting thing to see thus all models predict negative anomalies we have already seen this for drought of 87 positive for excess of 88 all but one predict a negative ISMR anomaly for the droughts of 82 and 2002. However all the models predict deficit ISMR or droughts for the excess monsoon season of 83 and normal monsoon of 97. Now since all the models predict deficit ISMR or droughts for the excess monsoon season of 83 and normal monsoon of 97 there is a marked improvement in the correlation coefficients of the predicted with observed ISMR if these two years are dropped. So what you see here are two things one is the pattern correlation coefficient between the simulated and observed mean rainfall for all the available years and the reason we have taken for the comparison is shown here this is the reason over which the actual pattern is compared for the years for which the runs were made with the model simulated pattern and the correlation is derived. What you see here is the correlation between simulated and CMAP or GPCP mean rainfall for all the available years and what you see is that in fact the correlation the mean patterns are very good the mean pattern turns out to be pretty good for CFS2 and it is pretty bad for UK Met Office where you remember UK Met Office had very much more rain here and almost no rain here whereas the opposite is true of the observations where you get rain over the east and not west and UK Met Office also had any hardly rain over India. So UK Met Office has relatively poor pattern correlation pattern correlation of ECMWF model is pretty good 79 and CFS2 is also good but you have to remember that CFS2 the sample size is much smaller because the runs are only for 79 to 2009. Now what is the correlation coefficient between predicted and observed ISMR this is why we are trying to test what is the skill of the model and the simplest measure is correlation between predicted and observed ISMR and what you see is if you take all the available years then UK Met Office has the highest correlation here and CFS2 is coming rather close to it. Now it is important to see see there are many studies which suggest that models that get the mean pattern better are likely to get the year to year variation better but actually this study shows that that is not the case at all actually the mean correlation is very high relatively speaking for UK Met Office but pattern correlation is the lowest now mean correlation is high for CFS2 and so is the pattern correlation. So this is a case and which I think further studies are required to understand completely that fidelity in simulation of the mean pattern is not related to fidelity in simulation of the year to year variation. Now we have seen all the models almost go wrong in fact why almost every model gets the sign wrong for 2 years 83 and 97. Now suppose we were able to improve the predictions only for those 2 years leave the others rest as they are then what would happen we will see that correlation will actually increase substantially if 83 and 97 are omitted from the calculation. So the coherence in the successful predictions for 87, 88 and the failures for the special seasons of 97 and 83 which were false alarms despite the difference in parameterization can arise from success or failure to predict a critical phenomena across the board for all the models. Now we have seen that 2 modes Enso and Equino play an important role in determining ISMR it is therefore, pertinent to consider the observed teleconnections of the monsoon to the rainfall over and the SST of the Indian and Pacific Oceans and compare them with the teleconnections of the predicted ISMR with the predicted rainfall over this region and predicted SST of these oceans. Now in this lecture I will focus only on the predicted rainfall and not look at the predicted SST the correlation of the observed and predicted ISMR with the observed and predicted rainfall over the Indo-Pacific region for all the models and for observations is shown here this is the correlation for observations. So what is this? This is the correlation between observed and predicted rainfall and so this is the correlation between observed and predicted rainfall over the India for the set of models from ensembles and for CFS 1 and CFS 2 in the following slides. So this is just the observed means it is the correlation of the observed ISMR with rainfall everywhere else similar to the slide we have seen earlier. So this is saying observed ISMR is highly correlated with West Equatorial Indian Ocean Rain negatively correlated with this and this is the Equino link and negatively correlated with rainfall over Central Pacific which is the Enso link. Now let us see how the models are doing by and large the Enso link is captured by all the models because remember the models were tuned were developed to capture the Enso link properly. So the Enso link by and large appears to be captured by the models but if you look at Equino link just see ECMWF seems to get it right that you have positive correlation with this and negative with this but all the other models are getting it wrong. So you have positive correlation with EEO rain over EEO and negative with WEIO same thing for UK Met Office it is getting it wrong same thing for CMC and same thing again in fact for Meteor France it is opposite sign it is negative correlation large negative with this very similar to this model here IFM model. So right sign of the link with Enso for all models and right sign of the link with Equino only for ECMWF model. Now same story again with CFS 1 and 2 see CFS 2 has very very strong response to Enso relative to observations as you can see CFS 1 looks a little bit better but both the models have the wrong sign of response to Equino and CFS 2 is much worse than CFS 1 both have the wrong sign but the amplitude of the correlations is larger for CFS 2. So correlation of ISMR with Enso index and rainfall over WEIO this is the correlation coefficient between predicted and observed and this is what you have predicted and observed Enso indices and what you see is the correlations are ranging from 0.65 to 0.82 pretty high correlation. So Enso index is reasonably well captured correlation coefficient between ISMR and Enso index which is the link now. Now here is the observed link which is correlation is only 0.54 you remember that is because ISMR is correlated not only with ISMR but also with Equino therefore the observed correlation is 0.54 but actually many of the models are overestimating the link with Enso except for ECMWF and CFS 1 except for those all the models overestimate the link with Enso and now we look at correlation coefficient between ISMR and Enfoil over WEIO which in the observations this is the western equatorial Indian Ocean in observations it is 0.51 whereas almost all the models are getting it negative except for ECMWF which is getting it positive. So ECMWF captures the correct sign of the link with Equino all the other models simulate the wrong sign of Equino link with Equino. So the predicted Enso index is generally highly correlated with the observed Enso index with the correlation coefficient ranging from 0.65 to 0.83 the correlation of the predicted ISMR with the predicted Enso index is higher than the observed correlation coefficient for all the models except the ECMWF model for which the correlation coefficient is 0.29. The correlation of the predicted ISMR with the predicted rainfall over WEIO is of the correct sign and comparable with the observed correlation only for ECMWF. So this is an important point to note for the other models the ISMR is negatively instead of positively correlated with the rainfall over WEIO. Also the correlation between ISMR and rainfall over EIO is negative as observed only for ECMWF correlation coefficients of all the other models are positive ranging from 0.21 for CFS 1 to 0.71 for CFS 2. Thus only the ECMWF model has realistic links with Equino. It is found that generally the coupled models predict strong phases of Enso reasonably well. However they were less successful in prediction of the strong phases of positive Equino in 83 and 97. On the whole the ability of models to simulate Enso monsoon linkage is quite reasonable. Though most models over estimate the strength of this relationship the most surprising result of this study is that Equino ISMR link is opposite to the observations in most of the models. Only one model out of all these models ECMWF model is able to simulate both these linkages reasonably realistically. Now detailed analysis of the cases of 83 and 97 when all the models failed suggests that improvement of the prediction of the phases of Equino the simulation of the monsoon Equino link and of some aspects of Enso is a prerequisite for better predictions of the Indian monsoon. So, let me say that the actually there has been steady improvement in prediction of the monsoon. In fact we had seen here that the correlation coefficient between the observed and predicted you know for ensembles is ranging from about 0.3 to 0.43. And this is significantly higher than was found in an earlier study with earlier versions of these models. And so there has been a significant improvement in the correlation overall correlation of the predicted and observed ISMR value in this ensembles related to Demeter which was an earlier experiment similar experiment on retrospective predictions done with European models under the European center thing. So, there has been definitely an improvement in the models both in terms of better representation of sub grid scale processes like clouds, boundary layers and so on and also better resolution. In addition now as opposed to earlier the models are in fact assimilating ocean state initial conditions much much better. So, there has been a market improvement in data assimilation as well as in the models in terms of resolution and their physics as well and this has led to considerable improvement in the skill as measured by the correlation between predicted Indian summer monsoon rainfall and the observed summer monsoon rainfall. But we have to improve them further that is very clear because we saw that in some years they give huge falls alums such as 97 which was a normal year and all the product models predicted a huge drought and 83 which was an excess monsoon year for which also several models predicted drought and all of them predicted deficit rainfall. So, once if we can improve these facets of the model if we can get them to do you know simulate equino better and the link with equino better and almost all the models failed to simulate the link with equino then we should be able to improve the skill beyond what it is today and avoid these big falls alums. So, that would require a systematic study to see why are all the models getting the wrong sign of link with equino. Now this is a very important thing to do and it is since equino has been discovered relatively recently whereas, Enso the big advances have come already in the 80s and so on. It is not surprising that the models are not able to get the equino monsoon link right because as I mentioned earlier models were actually tuned to or developed to get the Enso monsoon link right. So, this is not a surprise, but what was a surprising part of this study was that even some aspects of Enso need better prediction if you have to get the monsoon right. For example, the evolution of Enso in 1997 was probably not realistically simulated all the anomalies were there and in 1983 the El Nino retreated halfway through our monsoon season. So, June, July were very much deficit, but August, September we had the El Nino had retreated. So, Enso conditions were no longer unfavorable and we had the development of a very strong positive equino. This combination led to heavy rainfall in August, September and excess monsoon rainfall for the season as a whole in 83. So, it appears that definitely as we had expected the simulation of equino and link to equino of the monsoon have to be improved in the models and these years may be good years to work on to get the links right, but it appears that more work is also required in simulating certain facets of Enso more realistically. So, on the whole then we have reason to be optimistic about developing the coupled models to generate reliable forecast of ISMR in not too distant a future. Because as I mentioned before from dimeter to ensembles has been only 5 years and considerable progress has been done. If there are concerted efforts focused on trying to understand why all the models fail in some years and why are they getting the wrong equino link, I think we would be able to improve the models to give to a satisfactory level in the next 5 years. So, I am very optimistic that very soon our coupled atmosphere ocean models the state of the odd coupled atmosphere model in the world today would be able to do a good job of generating reliable forecast of the Indian summer monsoon rainfall. Thank you.