 I'm I'm I'm hoping that you all have noticed like how careful we select our speaker sets. So yesterday we had like all weird English accents. Today we have all like Germanic accents. I'm including myself too. Janis is in the Netherlands right now I think. He's a collaborative player in Robert Schwartz. Like he couples models together and tries to like enhance them and he uses BMI. And so that gives you a student talk here. But we're excited about his work. Glowfrim is a globally applicable computation that works for a hydrological model. Thank you for that. Glowfrim that is basically my PhD work I did in the two in the Netherlands University of Detroit and just kind of give you a really brief summary of the work I did in the past three years. Glowfrim the name stands for a globally globally applicable framework for hydrological modeling. Sometimes you may even get a better name. So if anyone has a nice idea and I'm pretty happy to hear it to improve that. So it all comes down to this equation. 1 plus 1 is 3 or saying hydrology plus hydrodynamics we find any synergy here. Can we by coupling hydrologic and hydrodynamic models can we improve our simulation results? And this is not just any weird assumption of course it's based on hard facts. And if you compare hydrology and hydrodynamics we see that they have distinct properties. And for instance the hydrology if you look at that particular global scale hydrology we see that it's forced by global meteorology so it's basically distributed we have values for almost any any place on the earth whereas if we run hydrodynamic models we are depending on observation stations and many models are just forced by upstream boundaries. So really lack a lot of locations where we don't have that inflow or cannot properly simulate inundation. However if you look at other aspects as the routing of the models you see that the large-scale hydrologic models are not really performing well because they for instance only solve the kinematic wave approximation which is definitely inferior to the full shade of water. For instance it comes to backwater effects so that can really have some big impact. And also the phase resolution differs remarkably we see that the global hydrologic models they run a pretty coarse resolution so I think the finest we can get at the moment is around five arc minutes that's about 10 square kilometers of equator whereas as Paul already showed that two days ago that the hydrodynamic model can really go to finer space resolution for instance like up to 30 meters which of course improves the the representation of the flat extent that also shown in the bottom right picture. So I thought okay coupling that is nice and I would develop some code and one point I had to give it a name and I called it glow for them. And in glow for them that is a framework it allows for spatially explicit coupling and this means that we do a grid-to-grid assignment and the coupling between the two models really happens from one grid to the corresponding grid in another model. It's online coupling so it happens at a time step level so it's not just running one model and then using the output as input for the next one would really yeah happen simultaneously which gives us a lot of flexibility in establishing for instance feedback loops between the model. And it's free and open so everyone can use it can download it if they're interested it's not behind walls and it's smaller so people can add other models if they want if they think okay I have a model I want to put in there or yeah just out of curiosity or got it performs better maybe it's it's quite straightforward to add models. So the entire framework then contains a lot of Python functions to couple the grids and exchange the data. It also contains like an interface script that actually performs the coupling of the of the model so it calls the the BMI it happens when I come to that later for the different models and then just thinks the different variables input and output variables and also performs unit conversions for instances necessary. And it contains a set of BMI models so models that are or were implemented some some BMI adapters so there can be coupled within flow frame. Those models are shown here so we have hydrologic models we have PCA Global WB that's from the University of Weed Direct that's a global scale hydrologic model and also the W flow suite so this is not global scale but the nice thing about W flows it contains a lot of other hydrologic models which is the HPV or SPM model. We now recently edited the global routing model camouflage and it contains two one the two the hydrodynamic models the one is called this what FB I think you heard about that right now and some maybe even followed the clinic by Paul and the other one is self 3D flexible mesh that's developed by the TARS in the Netherlands as well. So I've already said it's models are coupled by using the BMI concept the basic model interface and I don't know I guess many people already know about it heard about that here in a CSDMS but just maybe provide some course outline how it works so it's every model has an adapter and with that adapter or wrapper we can communicate with models and we can actually perform the steps the model would otherwise do automatically but now we can do that by using commands and I think the five major commands I think there's a big discussion I had two days ago what are the main commands within the BMI but I think that are really really important so we can initialize the model then you can get variables for instance runoff from the hydrologic model you can set variables so we can convert the runoff into discharge for instance and then get this discharge into the hydrodynamic models and then for the online coupling we update both models until the end of the modeling round and once we're done with the simulation we can finalize the model what we then only need is really this interface script in between to execute the different functions for different models to give you a nice visualization how it works we see here a lot of lines and the blue blue lines I hope you can see that it's the 1D network for DELF 3D flexible mesh for for the ELVE I think it's pretty obvious and that's when we force it with output from the K-Moflag model and really see okay depending on the if the models are at the edge of the 3D model we use the discharge that was routed before for if it's already within the base and we use the runoff that's coming from PCAC. And that happens really on a cell to cell basis so it's a nice we really get all the spatial variations in simulation runs and also in input data and you can use it for many many things so I just show a selected number of applications and for instance we can replace the routing as scheme from PCAClub as I already said it's not really good it has the kinematic wave and gives some issues if you want to simulate discharge and in addition to that. So what we did is we used the DELF 3D flexible mesh model you can see it here for the Amazon river basin or just a part of it but still big enough and we we force or we use the output from PCAClub and put it into DELF 3D and then we compare the simulated discharge and that's a lot of lines so I can keep it short we see here that on the left side the left side yes the DIN route so that's routing scheme for PCAClub that has a in good efficiency of 0.64 whereas if we use the same input or same forcing meteorological forcing and put it into a hydrodynamic model we get an increased 1.979 so we can conclude coupling models is beneficial in particular if we talk about discharge simulation let's say we can also use it to benchmark hydrodynamic models so we use the same output from PCAClub and put it into different hydrodynamic models and then we compare and see why are maybe results different in this case we compared the DELF 3D flexible mesh model with a this flood model again for the same same area in the Amazon river basin and we also see that models are different the flexible mesh as it says it's flexible and contains different grid sizes so they vary between two and ten kilometers whereas the this flood model has regular grid two kilometers resolution and that of course impacts the number of cells 1d and 2d cells but the main thing is about is that we use the same boundary conditions so we use a 0 meter water level boundary also the hydrologic forcing is identical between the models there is no uncertainty at least as far we can say at least there's no difference and what you see is that the results are actually quite similar in terms of discharge but if you look at this inundation extent we see that the differences are quite remarkable and there are two locations I want to point out one is at the at the river mouth I don't know what happens there so I don't have any explanation because it should be the same boundary conditions but still the models perform differently and then also in the upstream part we see that by using a flexible mesh and using coarser grid sizes in those areas compared to the regular grid see that of course inundation extent differs and with the flexible mesh we get larger inundation extent whereas maybe the number of inundated cells is actually the same and recently we we added the camoflat model so that allows us to do some nested modeling across scales so nested modeling is just to use one bigger model and then for a certain location we use a finer model for instance yeah 1d 2d hydrodynamics and that works like this so we have the runoff from the hydrologic model we put it into camoflat so that does the intermediate routing throughout the entire basin and then we couple that routed discharge as I showed before to the delta 3d model so we really see that there are really different scales and there are different levels of detail in the model so we have a relatively coarse hydrologic model we have then a pretty detailed dark-scale routing but such 1d and it can really simulate inundation extent in a a high level of detail and then in the where it becomes important so at the river mouse the river delta we use 1d 2d simulations to explicitly simulate inundation extent that we call nested modeling and I made some really cool animations if it works a dozen bummer animations today yeah that's a good idea is there anything happening no that's really sad put so much time in that bummer well what you can see is that that's inundation extent from the three different models you can really see that if you use a coarse hydrologic model and simulate inundation extent we see of course it works but the resolution is not really what we want and making some kind of risk assessment whatsoever on based on that information is really hard so that's the top left corner and the top right corner is then the same with camoflat we already see okay it's a nice much nicer resolution and but still it's of course only applicable maybe for large-scale and coarse risk assessments but if we were going to go to high detail oh there is some but anyway um so and then if you look someone took over the um yeah if you look at the bottom picture you see that this is the delta 3d 1d 2d model and that is really really detailed so I think there the space resolution is around one kilometer um two kilometers at the river mouth and we see that the inundation dynamics in the inundation depth and extent is much better resolved than in the other model and I would go on to the next slide so maybe this is this work so again we see the same nice discharge hydrographs and we see a lot of differences and if you look at the bottom picture we see this three different coupled models we see indeed that the dynamics and also the dynamic the variations over time differ depending what kind of model we use expressing that in some easily understandable numbers we see that we just use the pc o globe with the din route for soon we get okay results it doesn't really take a long time but then if we add a proper 1d 2d 1d routing with the camouflage model we see that the runtime is just a third of what it was before where is the accuracy doesn't even change I think that's a really big improvement in particular for large-scale simulations where runtime can be a bottleneck and then if you add the um 1d 2d models model we see that the runtime explodes which is not really handy maybe but the accuracy increases even more and I think that's just looking at discharge of course but if you compare in a dating extent I think the difference between the models get even more pronounced so what am I planning in the remainder of my phd which is unfortunately not really long anymore um I want to add the mod flow model in there so we can apply groundwater um simulation groundwater processes to the inundated extent in the high inundation model so it would look like this that we have output from pc o globe put it in the inundation model then couple that back to pc o globe and mudflow and then again mudflown pc o globe are updated to update the inundation model again if that makes sense yes it does um also it could be used to replace the on-scaling approaches for hydro launching models so it's just using large-scale models but if we really want to look at detailed inundation extent I think that tool can be can be used or at least you want to try it and also by maybe implementing like a crop growth model at one point we can also look into agricultural flutterers so now just to wrap it up globe frame provides a black and black framework it can easily be extended with other models I think that's a really strong point so it's not just a fixed code and it's flexible it's open you can work with that you can perform nested modeling simulations and it is able to provide simulation from the mountains to the coast so we can in the hydrologic model we can have some snow snow module solved then we have the 1d routing over the larger domain and in the river mouth we can have actually really really detailed inundations and yeah we can get it stuff so there is a first version is available as you know though but that's already a bit outdated there are a lot of things happen in the meantime and if you want to have the most date version you can just go to github or you can come before end this presentation I also want to thank my collaborators from different research institutes and yeah what I also want to say the model coupling is really nice to bring together models but it's also really good to bring together different disciplines and different research groups I think that's pretty really uh thank you Yanis um I think we can work with you to like make available these movies for people um on the repositories or on the website afterwards like give people a peek into them yeah later on still are there any questions out of the audience a short one is what we