 Hello everybody, thank you to invite me here, it's very nice to visit Boulder and this new community, I didn't know until now. So we heard a lot about the ROMs and now I'm going to present the Telemac system. I started my work when I was a student and a postdoc in the States a long, long time ago so I was studying small scale sediment transport processes and now I go to large scale applications so I'm mainly in charge of the development of the sediment transport modules in the Telemac system. So the Telemac system is now open source, it's a finite element free-surface model and it's made of many components so for free-surface flow hydrodynamics we have a Telemac 2D which is a 2D flow component and the 3D component is a Telemac 3D. We have sediment transport for the 2D, it's a CZF and 3D is a CD3D, it's actually part of Telemac 3D. It can be also a couple to Delwac which is from Delwac 3D and for the waves we have Tomawak for large scale applications and Artemis for smaller scale models. So these, the other things we need is also a grid generator. We recommend to use Blue Cano which is a new recently developed model and the heart of this model of the different modules are the finite elements library, it's called BF and you need also some post-processors like FIDAR and Parview. So I would say that this whole set of this system requires most of the, I mean satisfies most of the requirements that James told us about on Friday. So it's in Fortune 90, it's written in Fortune 90 and all the modules are well documented and I hope they do what they are expected to do. And we had, so it's been open source for a year now. So since July 2010 now we have a users club every year. So last one was in October, so a week ago actually and we had 130 participants, a crowd about this size and we have 2,000 people registered in the world and in more than 100 countries. I'm not sure they are all using it but I actually downloaded the code. For the developments are shared and organized and also well the strategy of developments are decided by a steering committee and mainly a European committee where we have 3 institutes, 3 French institutes, I don't know if I can use this. So we have and set math and EDF and we have 2 English ones, HR Winingford and Derasbury and this is a German one BAW. Okay, so next one. So it's a multiple scale, of course we go from small scale to large scale applications and so multi physics. So we need, we generally we need to couple the models, the hydraulic models and the sediment transport or the water quality or the wave modules. So the way to do that we have different options, we can change the models, that's the older way it was doing, we used to do that a long time ago and the most efficient way is to internally couple the models. So it means for example if you couple telemac 2D or 3D with CISIF, actually CISIF is like a sub routine of the main hydraulic module. And now we have a new internal coupling also for the telemac 2D and tomahawk and this allows us for example to simulate the longshore transport, the longshore wave induced current and for the longshore transport we still need the coupling with CISIF and we will have this probably at the end of for the next version. So the other way to do it is also use an external platform and it's been done by HR Wilingford, they couple the different modules using OpenMI, I think you can do something similar at CSDMS. Okay, so what is important also is parallelism. So all the options are parallel and we use MPI and we also include all the options and we have different options for the advection terms, for the advection terms and this also includes the method of characteristics, I think it's a very difficult point. It's been tested with up to a grid with one million of points and you can use up to 1024, I think they do more now, but it's been tested at this time up to 1024 processors on IBM Blue Gene, well yes it's tested up to 8096 processors, sorry. And okay so this is the results of that. So if you show along the red curve it's ideal speed up but somehow we start to diverge and have a lower loser, it's less efficient if you're using above 4000 processors. So as I said it's applicable to very different situations from rivers, estuaries and also littoral applications. Here's an example of sediment transport in estuaries in England and this is an interesting problem very important for EDF. It's salinity intrusion, it's in the bare lagoon, you can see in the red it's a medutal anion, high salinity coming into this freshwater lagoon and you need a very refined grid at the level of the stratification. A nice example is also the propagation of tsunami and we try to reproduce 26th December, it was in 2008, it was in Sri Lanka and can also have application in the ocean and this is a wave and tidal combined coupling for the in the Atlantic Ocean, the ocean model. So the BF library is a heart of the whole model so it's been basically written by Jean Chelaouet and he wrote a book so you can find all the information on this. I'm not going to give you the course on finite elements but you have an equivalence between finite elements and finite volumes depending how you define the cells of integration. So while you don't need to know everything about finite elements to use the models and even to program new processes because all the fundamental operations on matrix vectors and scalars are already built in this BF library. An important thing is all implicit schemes and it's interesting because it allows us to have current numbers up to 10 so it's nice to speed up the calculations and to have to go to large scale and large scale and long term morphodynamics. So we use the power scripts and it's MPI for the parallelism. Okay we had the talk before I was who presented briefly the difference between 2D and 3D models so I try to put this on the same slide I know it's not too confusing so if you have 2D models your your main variable are the depths integrated velocity and water depths and you want to use a 3D model you're going to have insight in the in the flow on the vertical structure and also calculate the vertical velocity so it's much more precise much more expensive too. So for in order to use a 2D model it's generally valid if you have a water depth less than the horizontal scales and you can assume hydrostatic pressure and supplicable to non reciprocating and well mixed flow so no stratification for the 3D model you need you have two versions one for hydrostatic pressure and one the other one non hydrostatic pressure so if you have a stronger reciprocating vertical flows you have to use a non hydrostatic pressure and you need a turbulent flow model so that's important so the equations for the 3D model are the shallow water summoner equations, depths integrated equations and so you have the continuity first equation and the second one is a momentum equation I hope I can use that one yeah so the and in this momentum equation the important term is the bottom friction I I already presented it because we it's very important for a sediment transport application to have a good representation of this bottom friction so in a 2D model you have a quasi a quadratic friction law it means that the bed stress is related to the depth average flow velocity and generally the physical you have different friction laws and different friction law coefficients but the most physical one I mean is the one which relates the product with friction coefficient to this bed roughness so bed roughness is classical relation when generally the velocity in most situations can be in boundary layer problem it can be it can be represented by a log profile and in the 2D approach we assume the log profile to be valid up to the free surface so in the 3D model we solve the full Reynolds average Navier-Stokes equation where here you have the turbulence coming in turbulence term and you need a turbulence model for that deterministic term and the difference between the 2D and the 3D is also in the representation of the bed friction so it enters the boundary conditions of the 3D model and you don't need to assume any more log profile up to the surface but you have to do this assumption at the first elevation plan so the bed friction is now related to the bottom velocity so this is a people who are working on who are responsible of this 2D model and this is sheet one who is responsible of the 3D model so you can contact them if you decide to use the models so just other things to say so the 2D model is basically the most robust and efficient model and you can apply it to large scale and long-term applications the 3D model is more precise more accurate very interesting and you can get some very detailed results on the structure vertical structure but it's highly sensitive to the choice of turbulence model so you have a you can go to the k epsilon 2 equation model or use a mixing length turbulence model it's also highly sensitive to the vertical grid resolution and the way you're going to set up your plan vertical plans so you need to do a lot of tests and to be very careful when you use this model one important feature common to both models is a treatment of dry zones on tidal flats for example so there is no elements removal all points are treated even if they are dry and generally just take the well just if you have a dry element you just get rid of the pre-surface gradient you set it to zero but there's a new recently developed algorithm which is very which is considered to be the best because it actually it's mass conservative and it's also positive it ensures the positivity of water depths so you can find details in this paper so an example of application is a classical for us it's a terrible event it was in december 1959 we had the break of this large dam malpassé drum dam break and it's located in the south part of france well it's about 10 kilometers from the middle middle and nc so it was very sudden and unexpected the break of this dam and the was i think 220 meter long and 60 meters high and we had the release of 50 millions of 40 and 40 50 millions of water flew into the valley and was very quick so we use a grid mesh of about 26 000 elements and time step of four seconds and the way the length of the simulation is 1000 time steps so it's less well it's about one hour and you can see that right at the start after 60 seconds the velocity is just at the very narrow part i mean at the level of the dam are very large like 10 meter per second so an animation i don't know if i can well this is an animation you can see the you cannot see it but i don't know why anyways it's you you can see the you should see you should be able to see the wave going into down very quickly and it's all dry zones so it's very important to have this nice treatment of the dry zones and you have a comparison valuation with comparison with the maximum elevation water elevation and at the distance from the dam and an example of the cpu time so if you use one processor it's only 52 seconds and if you go to eight processors it's only 10 seconds now you can do the same thing with a 3d model but you will have a even if you only have two planes it's equivalent to a 2d model actually but you multiply by three the cpu time okay just a word on the wave generation model called tomahawk so it's a third generation spectral wave model so the main variable is the variance density directional spectrum and it's solving the conservation of the wave action so it includes shielding reflection nonlinear interactions and for the source term it's wind generation and the sink is wave dissipation due to breaking white capping and bottom friction and we now have the wave current interaction for literal applications little currents taking into account radiation stresses this is put back and into the calculated by tomahawk and going into the telemac model and can be applied to oceanic or coastal zones and the person responsible is Giovanni Matarolo so this is an example of application it's we wanted to have a represent the wave propagation during a storm along the coast of France so we have two nesting models one is the oceanic model is a very large grid and a much more precise coastal model so the large-scale model give the boundary conditions to the coastal model and this is just a validation and with comparison at different harbour on the measurements between measurements and the model and model results for the wave height time period and direction so we can see it's very good and this is for the cpu time so it's a bit more expensive and takes about well the minimum is using 20 processors and three it's more than one hour for one year of simulation on the oceanic wave model okay now i go to sediment transport so that's more familiar with um so this is how we sketch the problem we have basically the flow acting on the mobile bed and the water depth is uh splitted in two zones the near bed very thin layer you have a bed load and in the upper water column you have a sediment transported as a passive scalar and suspended load so it's quite a classical way to uh split the transport rate into bed load and suspended so for the bed load we have a choice of a formula to calculate where they are semi-empirical and they come from the literature one of the most we have a choice of 10 and among them some classical ones like van wein like mega peter and the transport bed load transport rate is related to the excess bed stress minus critical bed stress stress so once you once you derive the bed load transport rate you solve the x-nail equation finite elements so here is a bed porosity this is a elevation due to bed load and now we have two ways to so this is done in cd bed load but if you want to solve the suspended load you can of course use a total load transport formula but it's not it's not uh probably recommended if you have a variation in the flow it's only if you have a steady state formula so you you need to have to solve this advection transport diffusion equation for the depth averaged sediment concentration and if you go to the 3d model you solve the full 3d equation and you have this additional so you have advection and the settling term take into account and turbulent diffusion here this is only a horizontal diffusion once you work out well this is important but i'll i'll go back to that it's erosion and source and sink terms coming from the bed erosion minus deposition rate and once you work out this concentration you know what the erosion and deposition is and you can solve the uh evolution due to suspended load well the difficulty here is that uh i'll go back to that but the deposition and erosion and deposition rate need to be defined at reference elevation which is at the limit between bed load and suspended load so we work on that we are with Pablo who is now responsible of CZF and myself okay we have different way different erosion and deposition law depending if you have non-creasive course of grain and creases very fine sediments so if you are in the coarse grain generally you have also to take into account sand grading effects so you represent your grain by a number a finite number of grain sizes each represented by a diameter and by also its percent is available in the bed for each each of the fraction you're going to solve the x-nail equation calculate the bed load suspended load and so they are more or less individually you solve each each of these fractions individually uh so the equilibrium for the erosion rate if you have non-creasive sediments you're going to use equilibrium concentration formula so basically the erosion rate is proportional to the equilibrium concentration and this is a settling velocity you have a number of again a choice of formula so this is a generally used van Rijn formula but you can also use biker or cisman and fresso and they are all defined and the each formula gives you also the depth of this reference evaluation so each formula is associated to this reference elevation and so this term is explicit in the model equations and this term is treated as an implicit term so the deposition rate is proportional to the concentration times settling velocity but all of this is taken at the reference elevation okay if you have a cohesive sediments you're going to have a crone and partenades erosion and deposition laws so here the difficulty is that the critical erosion rate critical erosion rate here is going to depend on the structure of the bed due to consolidation so we again if we have if we are dealing with cohesive sediments we have to have a consolidation model so we are building that it's a new it's a new it's in the new release in 6.1 and we have now a solving the gypson equation which gives you the evolution time evolution of the concentration due to consolidation i'm not going to have time to go into details in that so if you have a non-cohesive sediments you need also to have a vertical structure of the bed due if you want to take into account this sand grading effects so all of this my plot here is not correct this is all here this is the active layer so the composition of the transport is related to the composition of this top layer which is called the active layer and you need to prescribe the active layer sickness in the model so it's one of the limitation of this kind of approach so basically i i i could say many things about the differences and detail which are important when you want to simulate the transport so here's an example of the difficulty you have if you have a 2d model you have to have in this transport term the advection term for the concentration it's actually the depth it's a flux it's a depth integrated concentration times velocity but you only know what the mean concentration and mean flow velocity is and you make a mistake if you just say it's mean concentration times the if you don't time the depth average velocity just because you have a vertical structure and most of the sediments due to this very large gradients of the concentration most of the sediments is transported near the bed where the velocity is smaller so you have to take this vertical structure into account and you can work out making some simple assumption on the structure this ratio between the convection velocity and the depth average velocity so you you can work out this term and it's actually built in you can find an analytical solution for this term this alpha and it's less than one so it's important for this so if you do a 3d model of the suspended load then the problem the problem is that you have the bed for the flow velocity is at the bed but the suspension stops above the bed at the reference elevation so how are you going to do that and the suspension concepts break down if you if you extrapolate things down to the bed mainly because the concentration becomes infinite and so where the settling velocity the settling term is infinite and the erosion rate product of viscosity zero times concentration gradient is undefined so we are trying different ways to it's a tricky problem we're trying different ways to to deal with that and one one of them is to add a fictitious vertical plan where you define the boundary conditions and you stop the simulation the other important thing that's highly sensitive and the results are highly sensitive more more eventful than for the hydrothermics highly sensitive to the choice of turbulence model and the vertical grid resolution near the bed one other trick i mean you can do also if you do the 3d model is to add a laminar diffusivity so this is a rouse profile but it's modified if you have this finite laminar well defined constant laminar diffusivity you add this into this well i know i'm not sure you're all familiar with rouse profile but it's basically an equilibrium between settling and diffusion vertical diffusion but if you add the laminar diffusivity into this problem you get actually finite values of this bottom near bed concentration so the the value of the laminar diffusivity it shouldn't be important in the turbulent flow it's important here in this turbulence way we set up the model so this is a just a perfect way also to validate a 3d model log profile here and concentration profile comparison even concentration profile and the theme actually are the crosses and this is the analytical solution so another typical validation test case for the 3d model is this van Rijn experiment it's been also used by other people it's very classical and he made some detailed measurements of the propagation of a trench and concentration and velocity profile at different vertical plans so we use a 3d model with 16 vertical plans and non-iocetic pressure and we get a really good agreement i mean depending on the while we tested different models but basically we have a best agreement with the k while depending on vegetation but we get sometimes better agreement with the k epsilon order using the mixing length model but overall it works well this is a bad evolution and now so you compare the 2d so the 2d is really fast i don't have the cpu time but compared to the 3d model it's way way faster to use a 2d model and okay so here you see that the 2d model fails i mean it goes too fast rather you get a good agreement with the 3d model so in this case where you have this well kind of a strong recirculation in the back of this trench you need a 3d model well you can see also the effect of this correction on the convection velocity and it's slightly better but not so not quite good another example you can use a 3d model to represent the flow pattern in a meandering or river complicated the river difference here and this is a kind of thing related to what rohe told us about so you have a secondary recirculation and this is a 3d model and they are in the process of validating and comparison with adcp measurements here is a paranya river so it's a collaboration with carlos uné another typical test case we are doing is this program of meandering and you have a deposit here and erosion on the the outer edge so you have some detailed measurements and you can do the model in 2d but you need to parameterize the secondary currents or their effect on the sediment transport in 3d you just need a good turbulence model so if you use this parameterization you get a fairly good agreement between the model and the measurements now a nice application is ongoing work it's a phd which is done at ba w by an alien goal so you you can use a 3d model now you need a 3d model very detailed to get this bed bed forms formation and how the 3d model is going to reproduce these bed formations so a very detailed model it's a very small grid and you need this they also perform some experiments on the on this formation so I go very fast it's very neat so here you have a result of the model animation and you can see starting from a flat bed you start to have a initial perturbation and you have the built up of these dunes kind of feature and it's kind of realistic and very nice but the problem is you get really different results depending on the number of the way you you set up your grid basically you need a very fine grid and number of layers good enough in order to represent to have realistic dimensions of the dunes so that's ongoing work so I'm not going to say much about it really another so this is if you have a 3d model very detailed you can go to represent the bed formation the dune formation in your model but if you use a 2d model you're going to average the these bed forms and their effect on the flow and on the sediment transport so you need to parameterize the bed forms so basically you don't know how to do that and in another another way that here generally for the 2d model the usual way is to use a bed roughness as a calibration parameter but the alternative to do that is to have a predictor so I'm going to say a word on that so here we use a vanwein predictor so it's based on vanwein 2007 and the bed roughness is decomposed into small scale we per roughness medium scale and dune roughness and they all are function of sediment parameters and flow velocity so you can see that very small flow rate you have the ripples and then the dune is forming and then all the all the bed forms are washed out yes we have here an example of validation of these bed forms I'll try to be short but so it's just it was the objective of this this estuarine application of the coupling between telemac 2d and cesif it's done by Alan Davis at Bangor so I'm just won't have time to do go into that but basically you can use this multi-beam battery very detailed to work out what the scale of the bed forms are in this estuary so we found we have some dunes at the deep entrance and the scar of it and then he has ripples everywhere and he could use the model and compare the model results here with the average dunes here and this is a way compared to the two of them and he found fairly good agreement so it was a way to to show that this approach is realistic bed roughness predictors is a realistic and it's also very important not really for the flow you can you cannot see very much difference here is a comparison between the method of bed roughness predictor and here is an imposed constant bed roughness and you cannot see much difference on the flow velocity itself but very large difference on the bed shear stress it's a very strong effect and even more so for the bed load transport rate so I just go to the conclusion we have a this nice tool you're welcome to use it I you can download it from the open telemark website or from the CSDMS website and can you apply to very different complex environments so we show a few applications from small scale due information to large scale method scale let's say application morpho dynamics application we have this problem of uncertainty in the results there are lots of parameters in the models one of them is very important the bed roughness so we show that in order to reduce some uncertainty in the model results we recommend to use this bed roughness predictor and it's also a way to avoid some possible inconsistency between the model hydronomics and sediment transport models so we are still building the some more physical processes into the model we have a improving the sand grading algorithm we want to have a more continuous approach and we want to go to mixed sediments we have the consolidation and we are going to put the fluctuation we want also to go to mixed sediments as a main issue conclusion that despite we have all our efforts we still have a lot of uncertainty and it's difficult to assess also what the uncertainty in the model in the sediment transport models it's larger than for the hydronomic models so we need a new validation test case we need your help and we also are working on this automatic differentiation of the code to get more qualitative information on this error we are doing I can't finish this talk but this is these are the main people and they should be here to give this talk actually so Emil is the head of the project and Jean Michel is a father he's been working on the development of the model since 1987 so thanks for your attention for the challenge to see to give you an overview of this big model I think it's number one for the number of lines in CSDMS it's rated number one I don't know if it's a good sign or not do you have any questions