 So, welcome to this session of lecture, welcome I think most of you are just interacting with some of you just before this class and some of you have already told me that do not go to in depth thing just give us the overall light so that we just want to know what is the area instead of going to some particular model. So, anyway I will just try to throw some light to an comparatively newer area I am not saying that other areas are obsolete, but this is slightly that we have to do if you see a means this is a very hot topic in the sense that if you see that today's Times of India at the bottom news is that means they are just means claiming that stop debating now whether the globe is warming or the climate is changing on stop the debate on that issue now start thinking what we can do to mitigate this thing. So, the climate is changing and they have given the criteria it is very nearing that 2012 we have to do something concrete on ground by 2012 until then otherwise it will be too late to do something and they have given some numbers also that 3 lakhs people will die just for the reason of this global warming per year all over the globe. So, that is very and there are other things I think you can hear it from the media and all these things. So, I think my second slide that I am going to show you it is just a just an overview of this climate change and most of you are knowing just to give you the idea before I start what I am going to talk. So, just to give you some of you I think most of you are aware of this is it flexible if it is not is that means in a system of the in an in a hydrological system if you see starting from the snow pack it is saying that due to the global warming the snow is going up and the glacier is retreating that you know that recently they have found out that the glacier just out of the Ganga 3 is depleting. So, we may get the very low rainfall in the Ganges that is very dangerous for us. Similarly, for the forest fire and the extreme weather extreme weather I think from the particularly from the Mumbai cars you know that some recent experience of this extreme flood and all these things that you know there are some agricultural pattern change ecosystem there are some total the flow that is required at the downstream of a reverse air system and due to the low rainfall or the spatial change in the spatial distribution of the rainfall you can see that the groundwater is depleting continuously and as a result of that you know and that generation of the hydroelectric power. So, it means what the picture is giving is that the impact is everywhere impact is from the starting from the early morning to that the end when you go to the impact is there everywhere. So, what we can do what we have to do that is the one thing that people are also thinking and another thing is that at this stage as being from the hydrological point of view we will see something some that the one most important thing that if we can say something about our future if you can see that yes that is going to happen. I am not saying I am not claiming that I can stop that if the tomorrow there is a heavy rainfall in Mumbai there will be lot of plug and lot of thing will have happened and people will die. I am not saying that I will stop the drain what I am saying that if I know in beforehand then some disasters can be mitigated. So, that thing I will say. So, just I was talking with had before this session he was telling that some are you talking about some GCM or not yes there are GCM is one of the most important potential tool in this thing how to see that the GCM stands for the global climate model. So, that is also a good thing to do, but thing is that you know GCM is having his own shortcomings also means the grid size of one GCM model is much larger than the requirement of the hydrologic scale. When we are talking about from the civil engineering point of view from our hydrological point of view we are talking very small area. So, that in that small area that is called as a subgrid phenomenon in that GCM that terminology that subgrid phenomenon cannot be properly just or I should say in other words that till now the state of the art of the GCM is that it is not capable of giving all the answers that one hydrologist need at his own scale. So, here what I am going to tell you that is another way that is recently identified that is called the hydroclimatic teleconnection. This hydroclimatic teleconnection means that that I will discuss in details that is there are some link between what is going on in the climate what is going on in the global circulation pattern and there is some link some statistical link between those circulation means some physical links are there with the hydrological variable at the smaller scale. So, if somehow we can establish those links and somehow we can establish what is the physical reasoning behind that then we can talk about these things in much more details and we can use those information in a statistical way and we can use the information for our own purpose that is the sole idea of what I will give. So, first I will give you some idea about this some large scale phenomenon those are actually my talk will be mostly deal with the Indian subcontinent which are important for our own country. So, this El Nino Southern Oscillation some of you know the name this is one large scale circulation pattern over the tropical specific ocean then the Indian ocean dipole mode is identified comparatively recently compared to Enso that is also another circulation mode over the tropical Indian ocean and the part of that there is a atmospheric part of that is known as equatorial Indian ocean oscillation and then I will show you mainly that what is the Indian summer monsoon rainfall the since this particular thing is dealing with the all over the India there is in the large scale. So, is there any link or if there is any link how to use that information to predict this rare rainfall and after that we will go to that basin scale in the small scale what we actually require and finally, I will give you the conclusion. This is a slide that gives you the and overall idea of what is we are going to do we are going to use the information from atmospheric science and we are using it using that information in the hydrology and water resource through hydroclimatic teleconnection. Now, this what is this hydroclimatic teleconnection that I am going to tell you now. So, these are some two classical example of time series is given here we these are very common thing in the sense that some cases in some time series you can say that the sudden trend that is a depleting trend is going up and again you can see the inter-annual variance that is that standard day deviation over some period of time that is changing over time. So, the statistical properties of one particular hydrologic time series is changing and it is found that this kind of change having a association having a connection with this kind of change in this circulation pattern and of course that change in the circulation pattern as a result of our climate change. So, this hydroclimatic teleconnection I am just reading it out the significant association between the hydrologic events and large scale atmospheric circulation patterns which are widely separated in terms of the planetary scale. Actors the globe is referred as hydroclimatic teleconnection and most importantly it is recently being established that this temporal structure of hydrologic time series is significantly forced by this large scale atmospheric circulation pattern through hydroclimatic teleconnection. Now, I will give you some studies related studies if not for this particular Indian sub-bornness some other states and also I will give you some brief idea about this Eldino Southern Ocean before I go further just to fasten the smooth go of this talk. Just look at this in animation this is the tropical specific ocean tropical specific ocean you can see right hand side this is America this is the equator and left hand this is Asian countries are there this is the equator this red color means the anomalous warming up of the sea surface water sea surface temperature is warming up and cooling down is the also it is anomalous cooling down in a cycle. So, this warming up phenomenon is known as Eldino and the cooling down is known as La Nina and this is also related to the sea surface height change when the when it is coming to warmer the sea surface height gets up and when it is colder it is coming down, but these are all in the anomalous anomalous this anomaly means it is the difference with respect to the long term mean. So, you can see this is the actual SSD SSD stands for sea surface temperature and this is the equator that you can see this one Eldino condition typical Eldino condition this is the La Nina condition and this below structure this the these figures are showing that this is the anomalous temperature. So, and this temperature this here there is a warm pool of the water that causes significant change in the circulation pattern over this specific ocean and also this is associated with the pressure this is the anomalous pressure over this topical specific ocean which is known as southern oscillation. So, these two are coupled to each other and known jointly known as Eldino southern oscillation abbreviated as ENSO. Now, how it is it affects see the middle phase this is called the Walker circulation. Walker circulation there is a rising limb on the western part of this and when there is an Eldino condition the right hand side side this rising limb shifted towards center or towards the eastern part. So, this because you can see this is the very small but if you see from the global scale the specific ocean constitute of the significant part of our globe. So, slight change of this circulation pattern is linked to it throughout the globe and that is causing some significant change in this and the opposite case in case of the Lanina. And these are some regions that are identified by the atmospheric scientists these are known as that Nino 1 2 3 Nino 3.4 is dot shown by this dotted black line and Nino 4 and the SST anomaly of this is average SST anomaly of this region is used as an index of the ENSO thing. So, and it is found that for Indian subcontinent the most important or the most correlated SST information that we get from the Nino 3.4 that is why we will use the Nino 3.4 SST for the for our Indian subcontinent. And also the three months running average is also known as the Osanic Nino index that is also from the Nino 3.4 area. So, these two index is used as a measure of that particular circulation. So, another circulation that I told there is the Indian ocean dipole mode it is rather recently identified and published in the journal Nature and it is also published identified by the scientist in Indian Institute of Science. What they find that a similar pattern between the eastern and western part of the tropical Indian ocean is also identified. So, this is known as the Indian ocean dipole. Dipole means that there are two point action one on the eastern side I think the next slide it is shown here. So, this is W E I O is the western part of the equatorial Indian ocean and this is the eastern part. So, when these two convection over these two region are it has been found the negatively correlated. That means, when the convection over this region increased it decrease over here and vice versa. And this convection is related to the wind pattern over this equatorial Indian ocean region shown in the red color. So, the anomalous Jonah wind over this equatorial Indian ocean region is known as the equatorial Indian ocean sorry equatorial Indian ocean index. So, that index is used as an index of equatorial Indian ocean oscillation. So, these two indices we will see. Just a quick review I am I am not going the details of that. So, I will just quickly give you what the other states or the others they are also find out. But before that I should say that even if I have given some idea about this NSO and IOD and equatorial Indian ocean oscillation that I told that these are the most important thing for Indian sub subcontinent. But the list is not existed there are some other similar kind of pattern that are also exist in across the globe and that may be useful for some other part of the globe. For example, that Pacific North American pattern. So, here one study that Redmond and Koch has identified there are some statistical analysis of the nature and magnitude relations of precipitation over this and they have found some link is there with the SY is the Southern Oscillation index. So, Southern Oscillation and the Pacific North America pattern and also that Khaya and Drakap they have seen that there are some link between the stream flow pattern over the core four core regions over the United States with the Eldino and also similar influence is there from the La Nina. Massattel has shown that the link inflow in the Kulotha river basin has some influence from the Southern Oscillation index. The Elthair has shown that the river Neel the stream flow in the river Neel is the 25 percent of the natural variability can be explained just from the information of this NSO. And of course, there are a lot of studies in this way for other parts, but for the Indian part in this size research that we have it is still in the nascent stage even at the larger scale and that is our main goal now and but what we have seen what the established link that we have we have seen that there is some link between the NSO that is that Elino Southern Oscillation and Indian summer monsoon rainfall. So, they call that it is the it is very correlated. So, when that is higher. So, then we get a less rainfall over India and vice versa and so earlier means when the earlier works they generally use the information of the NSO and use for our prediction here and but it has been recently found that this relationship in Elino and this Indian summer monsoon rainfall is not properly established. There are some recent experience that even if there is a lot of land we get normal rainfall or it is already even if there is a mild and you know we get very drought in 2002 for example. So, Gargil et al commented that the relation between the Elino and ISMR is not properly understood yet. So, now the recent investigation of this IOD is showing that there may be a possible link between the the the joint influence from the NSO as well as the Indian ocean circulation pattern is influencing the variability of the rainfall over India. So, if small study is preliminary study has been done what you can see here is that there are some for different lags. Lags means that what is my target variable what is my causal variable what is the time lag between them. So, there are 4 lags we have used lag 0 that means that on the same time lag 1, lag 2, lag 3. The red dots these circles you can see the different size of the circles. So, these circles that you can see if the circle larger means the greater deviation from the main. So, those are and the red circle means the positive anomaly and blue circle means the negative anomaly. So, now just from your eye inspection you can see that there is for particularly for example, the lag 1 there is a somehow some difference is there over this 45 degree line this line indicates that when this NSO index and the equino index are equal. So, this plot you can see that there is the x axis is the NSO and y axis is the equino. So, when the it depends on this which is dominating. So, if the NSO is dominating that means we are getting lower rainfall here. So, it gives some one indication that it may not be only linked with the what is going on in the tropical of specific erosion that may also linked with the IOD. And these are some conditional probability has been calculated and it has been established that when the NSO is greater than the equino when we get the much larger conditional probability compared to the probability of the higher rainfall and adjust opposite when the condition changes. Let us see what the in for particularly for the Indian sub going to what can be done. So, we know that here the Monson pattern is so that the more than 80 percent of the annual rainfall occurs in the summer Monson pay period. That means that June, July, August and September and the studies of this hydroclimatic telekinetic in the Indian sub continents at the larger scale itself still in the nascent stage. So, we have to first establish if there is some link where is the link how to use the link at the large scale and later we can go for this smaller scale because that is more important for the hydrologic scale. So, we will see how the suitably because you know that whatever I think you have now got some idea about the subcomputing all these things. So, subcomputing is a technique where you get one input and it gives one output and it just trained the network in some way either back progression or whatever it is. But when you when we are talking about the larger scale when we are talking about the larger scale there is a source of uncertainty. Uncertainty means that you know that on the laptop on my computer I can do ok one input one output I can get, but in the when we talk about the nature it may not be that straight forward for a similar input it is not always guaranteed that I will get the same output. So, there are lot means the source of uncertainty is lot. So, in some in the at least at the larger scale I am not say I am not promising that in all scale it can be done at least at the larger scale if we can give some idea about the answer certainty along with the predicted values then that will be more useful to the end users. So, as I just discussed that is that uncertainty I have stated that we have to consider from the large scale establish link to the basin scale I have to come. So, what we can do and so these are the specific objectives that I am going to tell you. So, in this work first of all I will give you that idea of the larger scale what way we can take our research further and later I will give you one example of course that using the commodity tool at that case at the basin scale what how these tools can be used to the basin scale and what the further thing what we can do just to how to take this research forward for your if it is useful to you. So, what is our further objective also. So, first I am starting with the larger scale in this large scale first one thing we just now I have shown you that this large scale circulation pattern is always having some signature in terms of the sea surface temperature or sea level pressure that I have shown you earlier in case of the ANSO. So, these signatures are there. Now, I am talking about that these two this ANSO and the IOD are most important for the Indian the rainfall of the hydroclimatology of the Indian subcontinent is the most important, but thing is how we know that yes these two are important. So, a global analysis is done taking the SST anomaly of all over the globe and a contour plot the global contour plot is prepared and we are finding out the most effective regions over the globe. And this is these are the data that we have even the 5 degree by 5 degree entire globe that sea surface temperature is used land surface temperature and the month 3 rainfall over the 13, 13 sub subdivisions is used should be selected from the all over the India. These are subdivisions identified by IMD Indian Meteorological Department. So, this is one typical example this plot is plotted for all 13 subdivisions, but this is only one example that you can see the star point is showing that most important most influence there is the in terms of the Euclidean distance measure is used. So, in terms of the Euclidean distance which is the minimum Euclidean distance from which point you are getting that point is identified. This is only for the OLTC and also that sea surface temperature is also identified anyway. And these are the that for the particular season what are the locations this this gray scale where is the field up in the gray that means these regions are located in the Indian ocean regions and when these locations are not highlighted those are in the tropical specific ocean. So, you can see that when we are considering only SST most of the regions are in the tropical specific ocean. And when you are considering that some local influence also that is the local temperature, land surface temperature then most of the regions are located in the Indian ocean regions. So, the measure identification that we have found it here. So, we have seen that there may be a possible link with the IOD and the Indian subdivision rainfall. And when we are considering the only SST then we can see that it is mostly located in the tropical specific ocean. So, the circulation pattern from these two regions should be used in this way. The thing is that once we have identified we need some suitable tool how to extract those information and give some meaningful prediction along with the information of uncertainty. So, in this purpose we have used one basin dynamic linear model this basin dynamic linear model in hydrology this is very new this has some good important features one is that it is the it is the dynamic nature. So, the climate is changing and the once we in any more modeling there is a subcontinent modeling if you go if you go to some other modeling also. You see that the first part is the parameter estimate the parameter with some past record. Once you identify the parameters in the subcontinent also that connection weight and the bias all of the parameters. So, once the parameters are identified you freeze it and then you do the use. But thing is that over the time the relationship the nature of the pattern of the relationship may get change. So, what will happen in that case? So, here but this particular model is a dynamic. So, each and every time step depending on its performance it is changing its own parameter ok. So, the dynamic nature of the model is inherent inside the model itself. So, and it allows the external input of course means some input we need from the climate information relaxation of the stationarity assumption. This is a particular known as the stationarity assumption. Most of you might be knowing what is the stationarity thing that is it is the statistical properties of the time series over different times are fixed. But that may not be the case under this climate change scenario. So, thing is that. So, if you go for some relaxation. So, this model give you the relaxation of the stationarity property. So, that is most useful in this case and of course, we get the online external intervention. So, online external intervention means that I am saying that there is something going to happen in the next time. So, for example, classical example you take I am doing one stream flow prediction ok. And my model does not know anything about it ok. It is that getting one input and giving me some output. Now, somehow I came to know that next year there is a construction of the dam in the upstream ok. So, my model does not know anything, but I know from my as a human being I know that if there is something there is getting constructed there the flow may be affected in some way ok. So, I can give the model my own justification to it to give that you can change your parameters in this way that is the way. So, this is known as the external intervention. Anyway, for this particular example that I am going to see that online externalization is not done, but just to tell you that this can be done. And ultimately it is the probabilistic forecast. So, I think after showing this slide I should directly go to the results as you told that no need to go further details ok. These are the data that is the target time series is the all India rainfall and the exogenous inputs are the end so, and you know that I have discussed and these are the time period. And after discussing this I should go directly to the results what is going on, but just to give you some idea ok how this model work mathematical formulation of this work is that means this y stands for this what is my target time series as the rainfall. And this f is my regression times regression parameter regression vector what I am using and this theta is the regression parameter. So, these are all in terms of the vector. So, this f stands for the first this f this f is the relative weight is factor to En stands for the Enso index that is taking from the ith year of J minus kth month and similarly for the Equation index. So, I J are the target time and I am taking some lag kappa and lambda ok. And these are some initial information we have to give some belief we have to give that belief is we gradually evolve in the general time step. One small thing that I want to give you that this is one factor called delta that is taking care of the dynamic nature of the of the model ok. So, this delta can lies between 0 to 1 if delta equals to 1 means that is a static model it is not changing anything. And delta cannot be equal to 0 of course, but if it is very close to 0 that means it is very volatile means it cannot it something giving is the memory of the past experience of my performance ok. So, these delta has to be these are all the model parameters that we have calculate from the experience how it should change ok. So, the dynamic property is controlled by this parameter delta. And one important thing is that this is the prediction in most of the supplementing tool also you will see that one prediction you will get here you are getting the prediction in terms of some probability ok. So, this one t distribution with some scale and with some mode and scale and from that you can calculate what is your expectation of the A and what is your associated uncertainty. So, these two are important that I have already told. Now, let us see how it is. So, model is calibrated from the 1958 to 85 and these are the result is shown from the 86 to 2003. So, this you can see that the these are for all 12 months there are two bars one is the grey is the observed rainfall and the black one is the predicted. And as I told that observed observation prediction as well as it can give some information of uncertainty. So, that 90 percent of the confidence intervals are also given one is the upper another is the lower band ok. So, I just want to tell you one thing that in 1987 it was an Elino air ok. So, Elino means that we are supposed to get some lower rainfall and you can see that after considering that IOD also it has it has given that slightly lower rainfall prediction is also lower ok. Immediately in the next year 1988 it was a La Nina. So, India got higher than normal rainfall and the model has also successfully predicted the higher than normal rainfall also. Most importantly in 1997 case 1997 is the century record Elino that we have seen in that 20th yes century century. So, if the strength of the Elino is so high that means we are supposed to get very drought in that year, but fortunately we got the normal rainfall in that year and if you see in that model also when you are considering both the influence we are also predict the model is also predicting the normal rainfall in this case and just in the case of the 2002 you are most of you know that it is got the severe drought in that year and only except that July 2002 other months it is predicted the lower rainfall. Now the question comes is it means some other factor that is creating the role or is there any need for taking the both the influence together or not that is the one question. So, what this model is done with three different cases the case one that I have shown that is I am considering that both the cases of both the Enso and Equino has been considered. So, another two cases are the considered only Enso and considered only Equino ok. So, here you can see that for a particular month there are four bars ok. The first black one is the observed what we have observed second grey one is the when I am using both the things third one white one is the predicted using only Enso and fourth one is the only Equino. So, in most of the cases that you can see when we are considering both the things the prediction is good that you can see just from the visual thing and here if for here say this is the scatter plot this the A is the only Enso index if this is only Equino index and the last one is the considering both the cases. So, when we are considering both the information it is giving the better results ok. So, thing is that some of you may be working in some other area, but thing is that this particular model that is BDLM can be used in that models also where you are you know that there may be some uncertainty associated with it to capture the uncertainty this BDLM can be used not necessarily only in the field of hydroclimatology. So, efficacy of the BDLM model to capture the dynamic relationship is demonstrated in this. So, prediction of the uncertain future value in a distributional form so that information of the uncertainty also you will get it and here means these are obvious that I have discussed the concurrent influence of these two unusual experience of 97 and 2002. So, both Enso and Equino have a significant influence on their info over India. Before I go to the next slide I should tell that there is another tool that I have used that is known as that is very recently coming in the water research and hydrology is known as Copula. But thing is that here in this particular model BDLM one requirement of the data set is that your data set should follow or approximately it should follow the normal distribution. The next thing that I am I am just talking about is that it need not be any particular distribution that is known at the theory of Copula. So, here we have what we have done we have developed one composite index that is called the monthly composite index of Enso and Equino and we have seen how it is related with the different different homogeneous monsoon region over India. So, this is one plot that is the that is showing that correlation coefficient of the different regions that is the different months. So, June this June means that it is I am considering the rainfall anomaly from the month of June and considering the Enso information from a particular month June, July, August, September, October and what is the correlation in that ok. So, this is giving all 12 all 12 correlations it is not necessarily always we are considering the lag we are also considering because the atmospheric people they say that it may not be always that the circulation is giving me some input to my rainfall it is other way around also my monsoon in this India the monsoon also giving some feedback for the for how the circulation pattern will be. So, we have considered all these lags and what is this black and white that I will tell you and similar graph is prepared for the Equinoa also. Now, my intention is to pick up one particular month which is best correlated for the present month ok. Now, you see for this first you see the Equino case. So, this is for the June I see that March is the best correlated and similarly if you see for all the months the best correlated month is always the preceding month. So, some month from the previous K cases, but in case of the ENSO there are some K cases where we see that the best correlation is always coming to the after the month ok. So, we can use that particular month we can use that particular information of that month, but the difficulties is that so first of all we have to predict the ENSO that predicted value of the ENSO we have to take in care. So, instead of that what we have done is that we have taking the best correlated month from the preceding months ok. So, these are the months ultimately for June it is March and so March Equinoa July, August, September and this coefficient that you can see 0.18, 0.29 these are calculated in terms of the least square method ok. Now, thing is that once we have identified this coefficient is this coefficient are stable or not. So, minimum what should be the data length that we have to consider to get the one more or less stable coefficient that we have done in this way. So, this is the x axis is showing this alpha June, alpha June means that for the month of June what is this 0.18 ok. This is the this is the beta June this is alpha July like this there are 8 coefficients and there are 8 curves 8 graphs here. So, you can see and this x axis is the moving window size that data length is gradually being increased ok. So, we can see that after 35 years most of the parameters are becoming more or less stable. So, what we have done we have taken a moving window of 35 years and we have checking how these are stable ok. These are fluctuating like this, but we can consider ok these are more or less stable. So, using this again we have seen that the 25 percent of the total variability of the all India rainfall can be explained by this MCI in this way that monthly composition index whereas, only 11 percent can be explained by N so information and 4 percent can be explained by the equinox. So, we always go for the combined information ok. And this is also again we feed it the MCI information in the BDLM and how we are getting this information that prediction as well as the associated uncertainty. And this graph is showing you how this association is specially related ok. You see there are three correlations can be shown here. The first one is the correlation with the MCI monthly composition index, second one is with only N so and third one is with only equinox. So, you can see that for all the homogeneous monsoon regions in India the when you are considering both the information this is giving the better result. And one more observation is that for the northeast part of the India this correlation is not as good as the as for the other areas ok. So, for the basin scale study we have picked up one basin here that is the Mahanadi basin which lies in this area where this association is significant. Now, this is one basin that is the Mahanadi basin as I told that at the basin scale how we can do this thing. So, this is one basin and initially we are seeing at the seasonal scale. Seasonal scales means that what is the total monsoon monsoon and infant is considered. And if we see is there anything any link that we can see is that before taking up any work we just do one pilot study and if we see yes there is some scope then we go for the basic thing. So, here also I am just showing you one thing. So, and this is the location of the Hirakod dam and this is the Hirakod river and the Hirakod reservoir inflow that is the casement of Hirakod you can see from this dotted line. So, this is the casement. So, inflow in the Hirakod reservoir is taken and that analysis is done for the possible link with this large scale information. This is the graph showing that what is the average monthly inflow. So, you can see that most of the inflow occurs between June to October. So, total inflow from June to October is used as the monsoon and inflow and the in this total season what is the link that is being done. So, now if we see if we total this monsoon and inflow and if we see that this this graph is showing the annual, but thing is that as you can see that mostly it is the June to October. So, that graph will not change much if we consider only the monsoon. But what the most important observation is that as I showed at the beginning of this lecture also that is that inter-annual variation of this graph is getting changed over the time step. So, is there any link between this inflow variation and the large scale is our present we have done we have now we are using because here we are using the seasonal scale. So, they are also the informational also we are using as a three months running mean which is for the NSWIT is ONI, Osanic Nino index and also the three months running mean of the equi-oen. So, we have taken and we are doing some some previous monsoon period correlation and we are getting that the for the ONI the December, January, February and for the equino the January, February, March is giving me the best information. But I should tell you that there is one statistical significant test is there these things are just marginally failed from the significance test but thing is that my idea is not to establish the in this way I do not want to establish what is the link what I just want to pick up on particular period yes I should use instead of using some other period I want to use this one as this is giving me the best correlation that is all. So, I am using the ONI from December, January, February December of course means the previous year December and for the equi-oen I am using the January, February, March and this is the list this is the this is showing you the correlation coefficient because ok fine here the correlation coefficient I told these are not statistically significant and this is one that this is called the partial correlation coefficient the partial correlation coefficient means this say this R is the correlation coefficient between inflow and ONI when I am partialing out the information of the equi-oen that is giving me the 0.36 so you see now from 0.27 to it is going up to the 0.36 and the you know most of you know the hypothesis testing and the p-value comes to a very low that is 0.03 and similarly for the partial correlation coefficient between the inflow and equi-oen I am taking out the effect when I is minus 0.38 so this partial correlation indicates that yes both the information is coupled together so if we use both the information at the basin scale for the stream also we may get some more information in it ok so these are the coefficient again by the least square technique we just develop this thing and the conditional probability if you calculate this number of low inflow condition on that when the compulsory index is in the lower one third range is much higher conditional for Cp is the conditional probability is much higher compared to the high in inflow and similarly in case of the when the CA is upper one third range then we are getting the probability is the probability of the higher inflow is much higher than the other these are just the conditional probability and number of flows now we have just fit one very simple regression model linear regression model and the model parameters is developed on the 58 to 80 and the model performance tested from 81 to 92 and we see apart from this highlighted yes we see that most of the yes the prediction is sufficiently good except for this couple of years so thing is that there may be two things one is that you see that instead of going for this very simple linear regression model you can go for some improved model that may that may give me some better information and one more thing is that where we have due to admins the data was not with us at that time so we just consider up to the 1950 to 80 so better the data if you get the more information you can extract with the view of that we went to the monthly analysis and we are going for some model that means maybe some of you are looking for what is the sub computing in it so now what we have done is that at the monthly scale monthly scale why we have used the sub computing is that because the knowledge at the monthly scale is not because you know that when we are go for some sub computing technique is that the knowledge the background physical base is not clear to me I don't know what is actually going on inside it so that at that place we are going to use one ANN so what we are here we are going to do is that we are taking the stream flow at the monthly scale now not in the scale monthly scale we will see and we will use the information of the large scale information also and you also know that in some model what we do is that we use the information of the stream flow from the some previous time steps that is the classical time series modeling so that use of the previous time steps is that because the time series itself might have some memory in it that is also showing some what is the prevailing condition of the water set for the earlier time step so when we are coming to the small scale in the in the temporal direction it is monthly that time apart from using the large scale information I am also using the information of the stream flow from the previous month if they are related okay so this is the location of the stream flow basin puts slightly upstream it is taken just to it is some 100 kilometer upstream of the Cirogondri reservoir and as I told that both the information large scale information that I have seen from our earlier studies for those particular months we are using those large scale information as well as significantly correlated previous month stream flow is also used you can see for July it is June stream flow SF stands for the stream flow June stream flow is used but you know that June is the first month and if you do the correlation between the June stream flow and the May stream flow that previous month absolutely will not get any correlation so that is why we have not used that May stream flow in this case on the other end for the September we have seen that July stream flow and August stream flow both are important so we have used both the things now the genetic algorithm based evolutionary optimizer what is it step by step we go first you know that some parameter in initiation that I think you by this time you come to know but I think you should ask me the one question why PC is so low because you know that PC generally varies from the 0.8 or so so here it is 0.2 I will come to that who it is so and this is the probability of the mutation number of generation is 50 here and you know all the things the generation of the initial powers randomly we are creating some architecture some 1, 2, 3, something like that random architectures are created at the maximum number of neurons is 50 now what I am doing I am training each of them with the back regression algorithm with my goal that that with the goals are the maximum number of neurons used in the network that is some criteria is that so that is called the model parsimony the mean squared deviation and the maximum squared deviation in the training data set so those are the goals and once these goals are achieved at least so and ok fine so how these are performing based on this criteria I am evaluating some of their what is their fitness ok and based on their fitness in this in this 50 network in this population I am selecting the best population using the principle of GA ok so GA principles are that the two parent network are selected based on their fitness ok once we select those things now to those two parameters are taken care care for this crossover and but I think most of you know this basic things of this GA the crossover this thing all of you knows then I will go in details in that ok so and the termination criteria is that once in a set of the population that we have generated if at least one network has developed all these things all the goals are achieved then we are stopping the evolution and using that particular network for my further thing ok so these are the best network that we have identified for different month ok and this is the correlation coefficient during the training period and this is for the testing period for different month so here you can see that month that the monthly rainfall for different years it is it is mostly means compensated with the year with the observed one and here for all the months you can see two lines the top line is that mean plus 50% of the standard deviation and bottom one is the mean minus 50% of the standard deviation and here if some inflow is above this line that is categorized as high inflow if it is in between then sorry if it is in between it is normal and if it is below it is called as low so now what we are looking at how many low inflow I have seen and my model how many low inflow it has actually predicted ok so we have this is called the contingency table you know so here that is that category of stress inflow low normal and high and here it is the predicted category of inflow so you know that if we get one heavy diagonal here that means the performance is good right and there is one skill score called the Hedgegate skill score this is this can vary from minus infinity to plus one ok so if that Hedgegate skill score is more than one that means there is some predictive skill of the model exist and it has been tested from the experience that if the HSS is more than 0.15 that means the prediction performance is really good and here we can get one very high HSS ok so 0.5 to 5 again again once study is done whether both the information we need or not ok so first the green column that you can see this is considering only the stream flow information from the previous month ok now the people can say whether should we really go for this information large scale information or not otherwise the model performance may only due to that you have considered some previous month stream flow that is why it is given so the first column that the full procedure is repeated for all these three cases we have identified the base network we have identified the what is the model performance in terms of some statistics so first column is showing that stream flow information for the previous month only considering only the large scale information and the third one is considering the both and we have seen that the performance is the better always in this the last case ok so there is a need for in incorporate those large scale information also for this for the at the basin scale also ok so the influence of the large scale atmospheric circulation phenomena at the basin scale is shown here here and so the base basic idea is that even not only that previous month stream flow the information of the large scale information is also will be most useful in this cases also before I go to the overall conclusion to this thing the I want to tell you one thing is that it is not that only that stream flow information from the previous month we should use at the basin scale there are there are some more local influence also that may create some for example the local temperature, local pressure and all these things also so we can take the research in this way that apart from using the large scale so what we can do the large scale information we can use we can come with at the scale and at the basin scale also we can we can see for some of the local influence also if we can incorporate these two then definitely we will we can hope that we will get some somewhat better results also so in this talk what I just wanted to convey apart from this sub combating what I wanted to convey is that apart from the GCM there is one parallel way of this hydro climatic teleconnection also that which establish the link between the large scale information the different parts of the world and this large scale circulation patterns are getting the most immediate effect of the global warming on the climate change is coming on that one so if we can use the information from those information directly to this one then my information will be more immediate we will be getting otherwise you know that so global is warming circulation pattern is changing that's why the convection pattern is changing the conversable is changing that's why cloud is changing so you can come out to roof but thing is that at each and every step you are accumulating some uncertainty and at the end when you are coming to the hydrologic scale the uncertainty is so high that there is no meaningful prediction will be exist okay so this way also we can come so anyway more this thing I have discussed so I just I don't want to read it I just want to read that for the assist we have seen that the most influence of zone is from the tropical in specific ocean as well as the Indian ocean so use we have used both the indices commonly and we have seen that yes there is a scope to improve the prediction performance considering this facts this information and also at the basin scale also we have seen that apart from considering only the local information that is the only the previous month stream flow we can if we use the large scale information the better prediction performances can be achieved and also further we can use some more local influencing factor for the further improvement of the predictions