 Hydrologic modeling using FEM remotely sensed data and GIS. So, what is hydrologic modeling FEM remotely sensed data and GIS? Modeling you are aware particularly at least in any field that you are involved. Modeling is required right to simulate the original right. So, why modeling? What is modeling we know yes or no? Ok why model? But why not correct but why if I have the actual process why model it? To predict but why predict why I should not gauge it? Why I should not measure? Change alternatives ok agreed. If you measure you know the trend. I can understand if you change the sum of the data a futuristic land use pattern because every village is becoming urban. Urban is becoming you know what I should say metros super metros and all that. So, that I agree except that do we need modeling? Planning is based on the present and the future ok future we agree present we know then only for future we should model right. Sensitivity is if you try to change this parameter you need to know that is all otherwise we do not need modeling. So, that is why we need not only costly it may not be even sometimes possible. Yeah costly is the correct thing but sometimes not possible tell me an example ok please. No that is ok anybody else somebody else was telling yeah that is true more so because every watershed is not gauged is it not? Of course in the west it is extensively gauged may not behind it percent even rain gauges we do not have forget about the gauging of the rivers. So, it is for that reason otherwise probably you know modeling is definitely required because in spite of all the advancements that we have we have problems of water disputes between the states. Do you agree? Probably it is going to be you know the fragile you know the democracy itself is threatened our federal structure is threatened because of these disputes. Because everybody feels that even if I have water to spare but I will say oh our state is losing. These considerations are weighing more can the modeling help? Yes or no? Yes. But that means modeling is already there people have been doing why these disputes mostly our own creation right? If we agree for something then we should be able to say but anyway these are all narrow regional considerations political all that we cannot help it. But as a modeler can you say yes this is what it is water being the elixir of life it is the on the basis on which are the availability on which the civilizations have come up agree? So, the most important of course there is no control on the air so it is okay it is freely available so far although there are you know air oxygen cabins that are being built at some places. Water is another next important or of course leaving apart the food. So and the basic needs water is the most important whether you want to make roti or grow wheat, rice, makhan to make even the materials as well as well making and even roti kappadavar makhan in everywhere it is required. So it is the basic necessity and then what does this modeling will do? Can we increase the quantum of water that is available fresh water somebody must have given an estimate you know most of it to see water fresh water fresh water is locked up in the glaciers of Himalayas and other places are architect and architect whatever it is. So what is it that we are modeling optimum solution for what can you increase the quantum? No. So what is it that we are modeling? Use. Proper? Use. Use is in your hands how does the model help? Will it remain or not is a futuristic prediction will it remain or not is not you know that could be another type of model. Okay will it remain or not whether the glaciers will remain or will become water in tomorrow that is another thing you know global warming and all that fine but what is this modeling? Correct but you see modeling finally you say increase demand of water because of industrialization population growth what not okay fine. What is this modeling is going to give you? Can it give you more quantum of water? Efficient use of resources. Modeling helps is it? I am not quite sure but how does that help the results are implemented what type of results you expect? Friends one thing before you know I am before you doing something and you are want to listen. No modeling nothing in case of the scarcity of water scarcity of water is created by us. All the modern techniques okay this has given you to spread out otherwise you would have come closer. Similarly every technical invention has two sides of a coin it is a double-edged sword. So with the industrialization we cannot increase the quantum of water that we how much utility actually here for a given source or for a given river how much is the quantum so that that is what we try to determine. Management as such may take care but where is the management? We are not able to plug the leaking pipes we do not we are insensitive to a leaking tap we are insensitive to put off the light when not required. What type of modeling and what is that optimization you would like to do? Are we sensitive are we putting off unless we start from there no modeling is of any use. A leaking tap you know how much leaking overhead tanks we keep seeing we keep seeing the corridor lights in my own department daily I put off. Why we cannot be sensitive and that should start from there. We just do not bother and the E in engineering or engineer stands for economy are we aware that is it. So the moment the on tap water is given people do not know the worth of it those who get and those who do not get know its value. They run for few kilometers to get a drinking water part of drinking water. So modeling is secondary first conserve and utilize that is the basic step which although it is not part of my lecture I want to stress that conservation. Because the moment you know I do not know how much you know I have to struggle to get that part of water if at all I struggle to bring it from a river to my house. Yes or no can I do that so that is the first thing that modeling alone cannot do anything in the optimum utilization of resources all that you plan but then implementation. And then finally what is important is the conservation whatever possible you conserve and remove all the guzzlers industries which are meant for guzzling water like beauty aids preparation you know how much of water it requires. All that is required for our survival fine but those should be restricted and of course the water guzzling crops if you know in a region where there is less of water. I think it is both politically with the governmental and individual that has to solve the problems otherwise no modeling will help because I cannot increase the quantum of water do you agree. So modeling is only for those for our things and say ok how much can come out of it that is it but we cannot solve the problem unless it is tackled at the root cause ok. So modeling is only for you know to know what is the quantum of that that we can get otherwise we cannot improve we cannot increase. So FEM and remotely sense data has been used in one of these case studies and the mathematical models can be broadly classified there are different researchers have classified in different ways. So one could be a black box in Central Water Commission produced for every region and sub region in India a unit hydrograph theory based on that and everybody can have access to it you can request them and they send it. This is Krishna basin in that again some sub basins they have divided and then given that is based on unit hydrograph theory I think those all of you must be knowing unit hydrograph. So let us not get into the details that is what is called black box why called black box. Why not known something garbage in garbage out like a first time when the computers came even now we do not know unless we check some values simply giving computer is giving these values have no meaning. So this is that you only know the rainfall you know the so you get an output black box any particular equation it is adopting. So black box models lumped models could be of course these mathematical models broadly classified into black box lumped and distributed ok. Lumped Stanford watershed model if you have heard about it it takes some you know of these parameters but not an equations of you know unsteady flow equations and all that. So they are lumped and Stanford watershed model is one of the examples I am not saying this one of the examples I have given distributed model. Distributed model is based on complex physical theory of solving the unsteady flow equations. Unsteady flow equations are initiated it is nothing new 1885 or so Saint Venant has given these equations and lot of approximations that have been assumed to derive those equations which we will see subsequently. So this is a general the type of models that that can of course then you can talk about you know what is the statistical stochastic oh there can be many types of models. I have just broadly deterministic ok. So as I am dealing with this distributed modeling so let me confine myself. There are some of the equations you must be familiar because professor Eldo must have talked to you about the FEM while doing FEM lectures on numerical methods. So Saint Venant equations of continuity and momentum are equations where A is the area of the Q is the overland flow momentum equation in terms. These momentum equation in its full form particularly earlier was difficult to solve because it takes lot of time and the numerical methods when it came into being and when the computers were you know but a rare commodity or bigger bigger computers right. Have you seen or have you heard? So earlier even the present PC on this is much much bigger than the whole of the IIT building that used to occupy those days that was the thing. Because it is the same growth that has happened like our what should I say? Radio sets with the wall type to the transistors to the integrated circuits and you know even further that is the type of advancement. So earlier it was almost impossible to deal with fully dynamic waveform and then finally even if we deal with approximately or taking first and second is kinematic wave that is the slope friction slope is equal to the slope of the terrain and then adding term by term it gives a diffusion wave and then of course pass is steady and static. So that is full form and then that can be again there are many people who have done by kinematic. Some people with diffusion and of course these things these various things. What are the merits and demerits of distributed models? Distribute models are based on complex physical theory, mathematically rigorous prediction effect of watershed change possible we said somebody said futuristic so that is possible easily because we can incorporate. Of course what are the demerits? It is data intensity and no closed form solution numerical techniques are used. Those unsteady flow equations are difficult to solve in a closed form. So we have to use some numerical techniques of method of characteristics, finite difference, finite element, finite difference could be explicit and implicit. I think you must be already aware of them. So some of these methods have to be used to solve them because either closed form solutions for a simple case may be possible but it is difficult and sometimes impossible. So that is the reason why numerical techniques are required. Do the numerical techniques give equivalent a closed form solution? Yes, no. It can be very close not the exact because after all that you know it is not a closed form solution. What are the requirements of a good model? It should be based on physical laws. Yes or no? Black box model there is no. Incorporate spatial and temporal parameters of watershed on the storm. What is spatial with respect to space coordinates? Rainfall will change. Yes or no? Land use land cover will change. Can we incorporate them? Similarly temporal with respect to time there is variation of rainfall. So if these can be incorporated that can be a better model. Yes or no? Instead of an average over the thing and approach for engaged watershed. The best thing here we are modeling because most of the time there is no data. If there is data I don't need to model except for futuristic change. Otherwise to know even the quantum of water because one state says no inflow the other state says yes there is sufficient inflow. Who can decide? So, incorporate spatial and temporal parameters and approach for engaged watersheds. Mainly these approaches for engaged watersheds. You check and then say can you incorporate land use land cover alert parameters and they can be from the ground truth or a quicker method could be remotely sensitive. So these can be incorporated into the requirements of a good model. Can the model be almost nearly like a where do we stop this modeling? Is it possible to model it almost 100%? No. So we have to have a compromise. It may be possible to have it 100% but then what happens? It becomes so complex. It may not be worth using. Similarly how much simplification that you can bring it to? Then it may not serve the purpose. So our objective should be to have a tradeoff between the two. That's it. You can make it as complex as possible. But finally no modeling will be able to take the variations in the natural that it comes at every stage. That is the truth. But we can be closer to it depends upon how much of time, how much of energy, how much of money, how much of data that you have. It all depends. So it is for you to decide what type of model to go depending upon the data, depending upon time, depending upon sometimes. There can be a particular watershed for a particular storm even the black box models or conceptual models may behave better than a distributed model. It does not mean distributed models are. That is it because you have to go by the physical theory. So you may not be able to get sometimes a good results. So but that is that we have to accept. But you put in a lot of effort in the whole process. So what are the numerical techniques? As I have told you I think method of characteristics, direct methods, finite difference methods, explicit, implicit and finite element method. What is an explicit and implicit method? Must have talked about FDM was one lecture if I have seen the notebook or the notes. Yes, explicit, implicit? Yes please. Explicit, implicit. So a time step becomes a problem in explicit so we go for implicit. And so these are the methods of. The solution methodology that has been adopted is one is discretize the watershed. Comparison this computation of excess rainfall because excess rainfall is one thing which contributes to the runoff and evaluation of overland flow and evaluation of a channel field. What is that overland flow based on drainage, based on rainfall variation? You can take the some of the parameters are most of the parameters in take on rainfall variation topography, soil characteristics. And of course channel flow, straight reach, unique channel slope, no abrupt change in parameter all that are not possible. So whatever that is possible you can do and you can define your own criteria for choosing the discretization. So here mainly the case study that I am going to talk about is mainly based on drainage for overland flow. And these things straight reach is almost impossible any river system but to the extent possible and channel slope also may be changing. If you can take care of that no abrupt change in any parameter. So here although I think I do not know whether first of all they talked about these infiltration equations of this. Here what has been adopted as a first step is a loss rate or a pi, this pi index method which I think all of us are familiar. Yes or no? Yes sir. So how is this determined? One is of course based on this. Another thing is for calibration based on inflow and outflow you can calculate pi index. Yes or no inflow is the rain outflow is the runoff. If you have the data you can calculate pi index. Yes. Alternatively if you want to use it as an engaged. So what has been adopted is CWC has given these catchment and these catchments all over India into various zones they have Krishna, Godavari, Kaveri like that they have divided. And then again sub zones they have given they have given some an average value of loss. This is all for you know those the models that you know the modelers that they do not want to get into the equations for the loss rate. So they can use as an engaged. If you have a gauge data you yourself can calculate the loss rate. So if it is calibration it is based on pi index you calculate. If it is you know as a you do not have any data then still you want to do take it from the CWC reports. They are available if you write them later they will send you the reports. Last rate is calculated so average rainfall excess have been calculated. Check this whether the total outflow is equal to this including the base flow all that you have to do that exercise and check. This rainfall excess for each overland flow region has been calculated that becomes the input for the model. Is it okay? Here my friend will say why all this ANN can do. Yes ANN I know you have already experienced I mean must have undergone those lectures and the outputs. I agree there is no but problem is again like a model which you do not know really what is happening. You try to change some parameters and get an output based on the training. So the equation what equation it is solving? Is it unsteady flow equation? So reality of that I am not saying 100% real here also because to what extent you can go into details that is missing. So this modeling is based on there are various types of modeling. So what I am talking about is based on Saint-Venet equations. So that becomes the inflow to the models that you know this overland flow is based on although subsequently we have done with respect to the Saint-Venet equations. But here what has been taken as a mass balance rate inflow minus outflow is equal to increment in storage. Pt plus delta t is the next time step. Pt is the present time step. This is the rainfall. Average of these two is taken during the time step. Ac is the area of the watershed. Qt plus t, Qt are the you know time step, advanced time step and the present time step overland flow parameter. And that should be equal to increment in storage. Based on this you know a simpler mass balance equation of overland flow has been given. All the equations have been given in your report. The channel flow is based on the Saint-Venet equation where W is the width of the channel, H is the water surface elevation. Q is the overland flow and FEM formulation. I did not give you many details here because there are lectures dealing with these details of FEM. So, this equation, this LH is the one that is represented by this equation and that has been solved. FEM formulation has been developed and solved. So, that solution details I think he must have talked about under numerical methods. So, let us get into the details of what has been done further. The solution is a numerical method. So, which he must have talked and where is the remote sensing part here? Remote sensing data has been used to get the land use land cover. What can remote sensing give you? It can be used in various applications no doubt. But what essentially can it give you based on the SDS lecture? Please tell me. What it can give? Land use classification. Land use or land cover? Land cover, yeah. That is more appropriate. Is it different? Land use is different from land cover. What is the use? What is the cover? No, no. What gives? Is it land use or land cover? What is land use? After classification you may get. Land use you will get is it? But classification gives only, classification gives you land use or land cover? Classification gives? So, classification gives land cover. Land use? You have to you know based on other evidences or field you cannot say it is a water body. But whether it is a swimming pool, whether it is a irrigation tank depends on the shape and surrounding things. You know I told you interpretation based on so many things. So, directly classification gives you land cover. Based on classification we get land cover. Land use can be decided based on circumstantial evidences as well as the ground truth. So, either supervised or unsupervised method can be used to classify. So, that is what. And here it is a Pimpelgang Jogye water shed which is in the western Ghats of Maharashtra has been taken. And the discretization of the water shed as well as that we can have a look at this. This is the drainage, the boundary and then the location. Fortunately, can you imagine such a 102.1 or 2 square kilometers catchment has about 8 to 10 gaging stations, rainfall gaging stations, rain gage stations is very very rarest of the rare. I do not think now they have and the World Bank aided project that they were asked to install and report that aid came from there. This I happen to get some data from there. So, it has some 8 to 10, some events it was not there. So, that is what. So, you can see that dots, black dots are the rain gage stations. Along with that the thies and polygons have been drawn the area of influence of the rain gage station. And of course, these are the things that you can see the drain discretization of the rainfall. See these lines if you see that is considered as one this to this. Can you see this line dash line? You can see even from your reports this one. This is one of the overland flow from that side of the river and this side of the river this that is what is overland flow region. And channel flow at every region wherever it meets the channel flow regions have been marked with a circle that discretization of the channel. All that is there in that. So, this is the way that it has been discretized and this overland flow coming to the channel segment and the channel flow being routed to the outlet has been simulated in this work. These dotted line I will show you because I would have allowed to have a pointer. This dotted line for the model yeah good question that is what I want. Actually it should whether it catches your attention that is the reason why I kept I need not keep that. Good I am happy you are the first person to ask this question after all these lectures. Okay good why that this modeling when I had the problem of when I was doing I had the problem of last region is not being accounted. I was wondering what is it so for the modeling purpose I created an equivalent region at the end which is just before that. Are you getting my point and then as a pseudo but I did not include in the final thing thereby I got the outflow up to the outlet. Are you getting my point it is only for modeling purpose it has nothing to do with that. That was you know for me but I kept it so that at least one day one person will ask and I found you to be the first person. Okay so far nobody has asked so that is it for me for modeling because I was not able to get the last region outflow into this. Okay any other question channel flow you see what happens the it was you know last outflow was this one. See this region of these two were not coming I was getting up to this and here is the site where this river was gauged. Are you getting my point river was gauged at this you know black triangle and in my you know the outputs I was getting only up to this it was not adding. So I thought that time it was best to put another region hypothetical which is equal to that and then I got that. Okay the logic it was not no there is no logic is that I was not able to get this. So once I have distributed that I am getting one node earlier one node earlier in the channel. So now I get because this is not getting added this is no I have taken it almost equal to the same area. It does not matter as long as it does not get added to me because I keep balances and checks in that. So it does not matter actually any area not zero some areas. Okay 102.2 square kilometers depth of overland flow based on the previous equation. You calculate you can take yeah you know finally the overland flow is calculated the depth is based on this equation. You want me to go to the slide 2 by 3 yeah best than what now tell me what is this equation represents can you tell me anyone. I think the equation because the depth of flow is so small that is taken as that's these are the assumptions that's what I said no model can be 100%. In fact most of the equation that we get are based on approximations. So how to what extent you can model it is like drawing the whole earth one to one scale feasible. So that is if we want to have 100% modeling all these problems are not there in his modeling right. Okay so tell me anything else can I proceed further that we'll come back to the discretization if you have questions. It should be always questionnaire session no lecturing because that is what gets to the student to the participant anybody. And as long as we are comfortable with the receiving questions. Sometimes it may be little thing where you may not have accepted what is how does anybody know answers for all the questions in this world. Then what is the point accept it as you said what is the logic there is no logic because I was not getting results up to one node less. So I thought okay that was the best thing that I can do no logic that's okay. I may not be able to satisfy your thing but then there's a question yes at least it can keep me thinking. To find out a logic which if I if not now maybe next time okay yes please. Shall I go to the boundary of the watershed can be fixed by either that's why I said questions because rather than otherwise spoon feeding everything that is right. One is toposheet you can take those ridgelines and somewhat you know it may not be easier maybe between the two where from these starts if ridgeline is there fine you can get to that. Otherwise halfway through these you know drainages where they end that is the way from toposheet. Now if you are talking about the modern day remote sensing methods these DTMS again can give you if you can generate. DTM can be generated one from the contours that are digitized from toposheet. It has its own problems but that is what is normally being done or the GPS values that are being taken and interpolated or the SAR interferometry. Side looking airborne radar interferometry can give you the heights and of course these spot IRS 1C, 1D, Cartosat they have you know a tandem to use to generate a 3D model. Any of these things can be used to generate and then once you have a good you know digital elevation then you know. You can mark the boundary easily in fact DTM can help you even generating the drainage because you know the heights that is it. Of course you may have to correct it for some of the flaws that is what you know sinks have to be filled up flat areas have to be decided because she might have talked because my job is on water resources and this so that sinks have to be deleted flat areas. In the process the original sink sometimes may be deleted the original flat area we may be forcing it to go otherwise the program does not work is it not. See mentally you can think suddenly what happens if this happens you can say but the program does not know unless you give a way out for the program so that is it. So, by any of these methods it can is it okay. You see finally it is like this. Every present day parents want the after the birth of a child what IIT or even before birth possible okay then IIT is not getting the next best you know anomaly will do okay fine anomaly. That is not possible Satyabhama will do that is then something so finally it all depends upon what level of data because everybody may not have the same intelligence coefficient yes or no. So, we should be again there is no point in getting frustrated both parents and child because I have not got into IIT or my children has not get there is no meaning because there are ample opportunities in this world for everybody. Everybody is born with a purpose we should accept it so if that is not there what best can be done. I am talking about in general and in particular with this so that is not there I have at least one brain gage if it is there yes or no. If that is not there no we will come to the worst how does it matter because that is the engine meteorological department will have some brain gage stations all over India. The nearest meteorological station we can ask them to give data what I am telling you the worst that can happen. So, with that that may be a single value only thing is your model becomes course your values may not be that reliable that but still you can model. Model with the available thing that is what I am telling the same model on war land flow we have switched over to two dimensional same 20 equation flow later. So, all the eight brain gage stations hourly rainfall and hourly runoff I was fortunate enough to get otherwise it is difficult to model a distributed model I better go to the other models. Hourly rainfall and runoff for this catchment and all the eight to ten some events I had ten some events I had eight rainfall hourly. I actually what happened you know I waited quite a long for this so I said at least ten for two three storms I got for two storms that is it I did not get. See finally there is a problem with this data everybody thinks that is a secret one oh no it becomes what happens tomorrow there is a state dispute between the two and then what happens. I think we should have a national data center although NIH NIH was also asking for the data when I was looking for the data. So, I think there should be it should be mandatory for any developmental project or anything that is being done in any state any nook of the country should reach a particular data center. I am sorry to say in Central Water Commission who took my program at that time promise many things have not delivered. I am there is nothing there the director himself flew down and then took but then he said ok I will you I said I want to you know calibrate more catchments but it did not come through. Somehow we are very secretive about data because they may be worried that we are publishing or something like that but I think you know publishing is on your own data you cannot publish it is only something you do on that then you can publish yes please. That has come now not all this right information I do not know this data also maybe. So, I have never used that but maybe my students can use now. So, we do not know anyway information that commissioner will decide if at all it is coming into dispute. Yeah did I answer your question but yes. So, it all depends upon to what extent we have the data but modeling can go on only thing that they may not be distributed you go to some other model yeah please. On the basis of. What station where the stations are placed. No actually here one is I have gone mostly by drainage as if this overland flow if you look at this first one can you see this here. This is the first to that you know on the left hand this is the overland flow region. I I said if this I divide here or here somewhere this should not go on club or I cannot divide like suppose if I draw a line like this I am dividing the drainage into two separate regions which I thought may not be correct. So, I went though I have given you criteria of various things I went mostly by drainage. Okay. That is the criteria I have used yeah. Specific. No this discretization is very. I think the drainage network can be divided. We are seeing the drainage network it is possible to do that is not. Yeah drainage network if you see this has been done like that. Okay. Anything else? Shall I proceed further. So, the maximum likelihood classification which we have discussed so that these density lines are along the spread it depends upon how they are spread the points. So, it becomes it is not you know parallel pipe it is based on the probability density function they may become ellipse somewhere circle somewhere if you have seen you remember yes or no. Parallel pipe it is something like this or a rectangle like this or a stepped parallel pipe these things we have seen and the maximum likelihood what happens if the spread is like this this becomes a circle if the spread is like this this becomes an ellipse. If you have if you can recall my SDS slide if you want I can go back but that is what is maximum likelihood probability of each pixel belong to a particular class because the supervised classification. Based on supervised classification you calculate the mean variance covariance and every pixel you compare where does it fall. Probable density of belonging to each class is calculated from this formula and whichever is highest this pixel is allotted to that particular class is it okay. So, that is how a group of pixels will come to a particular class it is quite time consuming but that is the what the maximum likelihood is supposed to be maximum supposed to be is a better classification amongst the supervised methods okay. Based on this classification you can see these are the types of the land use land cover classes that I have got thick forest, forest, forest, agriculture, pasture, fallow, open land, settlements, river is it okay. This is based on that and this has been even a color code depending upon the class that it is. So, these are the classes that have been obtained from the maximum likelihood classification of Landsat data this satellite is the Landsat okay. One of the first few satellites that were launched for a civilian purposes since 1972 and what has been used based on these land use land cover classes shadow covered foothills ridges. The roughness factor has been from the literature has been taken for these particular classes and they have been used for this model and those values are given as shadow covered foothills of ridges is 0.35 vegetation 0.35 Season agriculture 0.53 barren land mixed all these values have been used for the model. So, overland flow roughness values for overland for regions have been used and the simulation studies with respect to the parameters obtained from Landsat data have been used and simulated. You can see the observed is given by these circles empty circles and the triangles gives you overland flow from remote sensing overland flow N that has been used and then if you look at that there is a variation look at the triangle versus the empty circle there is a variation yes or no. So, I thought then by reducing you see the difference is something like this. This is observed right and this is what has been simulated I think it is moving with the actual overland flow roughness values. So, I made it half to see something like a calibration and it has improved to this and both the results are shown there this is for storm 1 and another storm also is there another storm. But the what is important is not only this in addition to model that you have seen discretization and if I can go back of course, those results I do not have here, but suppose say if you look at this you look at this this whole thing and this you can consider this as a tributary instead of overland flow. Do you agree? Because it has an equal amount of area. So, if you consider as a tributary, tributary means this is another channel instead of only one channel I consider to be two channels merging at this point. So, that has improved definitely the results, but because I am stressing more on remote sensing I did not get into those you know I did not present those results for you. So, this is how it is. So, remotely sense data is useful yes or no. So, this is you know the similar hydrographs using these roughness values are in fair agreement with the observed hydrographs and of course, as one of you suggested futuristic you change the land use land cover put them you get the new one. This is real modeling usage or the quantum that is all, but optimal utilization and all that is for people to do. What is in our hands is only this much that I would like to say what are the limitations of our modeling. Because how best to use goes on that. So, this is one aspect and detail.