#95 Future of 2D hydraulic modelling

thank you for joining us to another great australian water school webinar i'm trevor pillar chairing today's webinar where we're going to explore the recent advancements in 2d hydraulic modeling well it's been a big build up to this webinar there's a lot of people on board so we'll get this cranking pretty fast you saw a map as we came in of all the people over 50 countries represented thank you for joining us i'm delighted to be here today but you know what really makes this click is the presenters who are with us today three water modeling professionals from two flow led by bill syme we're so glad and honored to have you on board gentlemen bill chris and greg all working together as a team on two flow we also got over 30 years experience as you can see there bill is a software business lead at bmt but together they've been working for a fair while on um a fair while is probably putting it mildly i'll leave you to explain to us i've got one question for each of you and that is if you could just tell us what motivates you to uh take on this life it seems a bit of a mad life at times here in modeling they're changing so much as it does over to you first bill oh what motivates me well i guess i just love water i always have loved water love water sports and unfortunately i also like computer programming so that the two ended up marrying together and um i just have always been very passionate about you know how to simulate water as it moves down a river basically yeah yeah and chris i'm much the same as bill actually a love of water i originally think i almost went into meteorology but then shifted across into understanding modeling yeah no that's really right yourself greg oh yeah i'm probably the geek amongst all of you i just love mathematicians and computers and yeah it all just fits together nicely for me yep now that's fantastic and you're all um from the northern part of australia queensland [Music] brisbane's capital city yep i'm down in the south southern part putting up we're still with the cool weather a lot more to come but um so great to be doing this together and you bring decades of experience it's wonderful to be working with you again after many years well i think we'll leave it right there for the intros we want to get right into this thank you right over to you bill to take us on a bit of a journey um so thanks trevor thanks joel um and welcome to everyone so it's actually very um just fantastic to have so many people dialing into this is it's because it's pretty exciting stuff that i'd like to talk about before i get into it though i will just mention that we're replacing the old website so a lot of the stuff i'm talking about today isn't actually up on our current website um we're transitioning to a new website we're frozen the old one and there will be a new one up hopefully in october 2020. but if you have any questions in the meantime feel free to contact info2fo.com okay so an overview of today's presentation i'm going to step back and actually just think about the history of how 2d models thing solvers have evolved i find it quite interesting something i've thought about quite a bit um but it helps us understand too why we are where we are and where we might be going like all things there's still always limitations to be resolved so just this couple of years we sat ago we sat down and we focused on what were the things that were holding us back and we've been focusing on those and that's essentially what i'm talking about today how we've been addressing these limitations in 2d solvers i'm not talking about just two flow this is across the board so the evolution of 2d hydraulic modelling well i've stolen this slide from rory nathan who some of you would know basically it is an evolution and i guess the early days of hydro robusters the big question is are we now up to hydro sapiens have we how much further can we go with our 2d modeling so just looking back in time i mean the first 2d models started to emerge around in the late 1960s this is the work of lendurtzy in 1967. he put out fortran code to simulate 2d flows and he had a really large 100 cell model to demonstrate his his his scheme actually works 100 cells anyone make a 100 cell model these days i doubt it that's sort of pioneering work as well as the work of mike abbott and system 21 which became mike 21 the last of the what i would call the really good adi schemes was by stelling in 1984 and this is what the one the twoflow was based on and that was you know that in sterling's work he had a 1 000 cell model so that was progress but it's also worth noting that these guys were driven by modelling of coastal bays and that's where 2d modelling started in the coastal tidal area then in 1990 this is where two faces stepped in and well i haven't seen anything different but it really was the first scheme to successfully link 1d schemes with a 2d model in this case of the gold coast seaway and we're starting to see some graphics this is a graphics system that i developed around two flow on the right but the ability to model the whole tidal system as a combination of one entity really open up the door and we did so many studies using this system back then we had a 10 10 000 cell model of the gold coast seaway which by today's standards is very tiny but we've got some really good results then we moved in the late 1990s we started to see 2d models starting to be used for flooding and this is one of the very early pioneering studies and it's pretty exciting stuff to be honest because finally in the flood world we could start to see how the water was moving around you can really visualize water in this case coming across the levee through the bridge expanding downstream there's very dense vegetation here at this site so that was that was really exciting stuff and we're starting to make models around the 50 000 cell mark which we thought wow that's that's amazing then we moved into the 1d2d concept was extended to the urban environment starting around 2003. this is a model of bristol city in the uk on the very first 1d pipe 2d surface water models and there's a some of the flood maps that came out and we even started making animations back then yeah exciting time and we're getting up to 100 000 cells really exciting but then we had this quantum leap in the early 2000s and thanks to people like greg and others we went from cpu to gpu hardware and that actually gave us an extra order of magnitude on the improvements in speed and ultimately that means run time so we're getting up to 10 million cells and so we can go to even started to consider a whole of catchment modelling so this is a catchment in the uk that would use two for gpu and you can zoom right in and still get amazing resolution it was just yeah very very exciting stuff so the big question is what do the 2020s have in store for us what do we need to improve how do we make this martini modeling even better what i find interesting over that course of those you know it's been 50 plus years is what i call mathematical elegance so in the early days there were the computers had virtually no ram they're slower they're way slower than your smartphone there's no parallelization and the people doing their pioneering work really focused on attaining as bigger time step as they could so that their solution could progress very quickly and so they really did some nice stuff in the mathematical space and they're all implicit schemes solving matrices lots and lots of line of code it was they're really driven to get the most out of the hardware by using these large time steps and matrices that sort of start to change when multi-core cpus and then of course gpus which have you know thousands of calls embedded in them came in that opened the door to finer volume explicit schemes which don't use matrices they're very mathematically much more simple you may call them a bit of a brute force approach they do require much smaller time steps because they're bound by things like the current condition but just there are so many calls that they just were so much faster and much easier to code up and so forth also things like ram or memory but in the early days there's a bit of an issue but in recent times that's not an issue you're still limited by the runtimes so the mathematical elegance sort of decreased we didn't have to be so fancy with our mathematics but since that sort of big shift we are slowly improving and fine-tuning these final volume screens which is really what i'm talking about today the other interesting factor is the accuracy have the schemes got more accurate over all those years well in my view it's a diverging thing in the early days there were very few 2d schemes but they were mathematically elegant they tend to be implicit second order finite difference the finite element and they really did have and they did use all the terms in the 2d equations including turbulence so they are focusing on the coastal modeling area which needs to have all those terms including things like coriolis and so forth so they're pretty high quality stuff they're doing back in those early days um but today we have lots of 2d schemes out there lots to choose from from the explicit second order finite volume schemes such as 24 hpc that use all the terms and 2d equations right through to much more simple schemes some schemes are just a 1d equations or a 2d mesh some schemes don't have all the 2d terms and some are first order which can suffer from numerical diffusion so we have a lot of choice here and it's really hard to sometimes understand what you know which other ones to use and which ones are appropriate today we're probably we're talking about the top end of town because that's what we're into that's what we focus on um but if some of these other ones are fine in certain situations but just be very careful using those and why should we concerned about that well it's sort of quite simple the way i see it these are the key terms that 2d equations might try and solve um you know you're trying to determine how water changes is moving over time and that is affected by inertia coriolis gravity bed resistance atmospheric pressure turbulence and any external forces so if you're modeling something like that on the road so that water is fairly moving slowly you see a guy riding a bike through it it's a mixture of ponding and slow movement you do need gravity because the water's got to move downhill and you need some sort of bed resistance normally mounting is in equation which is resisting that gravity and that's effectively all you need to model that type of flooding and that's fine but not far away from where that person is riding the bike we have this going on and that's really really complex flows um and whilst you might not might not need coriolis you might not need atmospheric pressure or any external forces such as wind you certainly need gravity and bed resistance but you definitely also need inertia and turbulence because those terms all play a role in how this water is progressing down here and so you do have to understand whether your hydraulic solver is suitable to your problem are the mathematics appropriate and a really nice example of this which i've shown in the past is this is showing water level rising with time from a levee breach onto a flood plain and originally software a was used and that was a result the person doing that modeling was a bit concerned about the results just based on how they looked and you know their unsteadiness you see that in the results there so they ran another piece of software software b and they got it quite a different result and very different propagation times so they did the right thing they didn't explore this further because being a fairly critical study they also ran it in software c and got that result incredi in fact almost two is similar to software b if you have a closer look at the software it really does they're all presented as 2d software but i mean look have a closer look you'll find out that the software is just a simple 1d solver of a 2d grid whereas b and c are advanced second order solvers from that coastal era so you know you will get differences when the hydraulics get complex and it's really important that you understand that so make sure your hydroxyl is appropriate and if in doubt benchmark or seek advice so what are the limitations of our sort of modern day accurate 2d solvers um the things that have we've known about for some time is the issue of sub-cell turbulence and all basically the turbulence models that are used uh often cell size dependent or inaccurate particularly when the cell size gets a lot smaller than the depth that has become a major issue some first order schemes if using one numerically dispersive but uh not a substitute for turbulence they'll give a similar effect but they are not a substitute for modelling the turbulence so some cell turbines has been known for a long time even mike abbott in his work in the 1970s identified that um mesh resolution orientation design well this is a this is always a challenge for modelers particularly for fixed grid models such as two for hvc but you know your flexible mesh models can also suffer from this if they're not well designed you can get sawtooth effect along the edge or drive due to dry cells protruding into flow pass if your mesh is too coarse it'll block or restrict flow paths um typically schemes just have one elevation per cell or in two phase case spine elevation for cell and for cell face so that can be a poor representation of the underlying terrain and of course 2d fixed grids historically don't you can't bury the 2d cell size which is a major limitation so we'll start with our turbulence and then we'll move on to mesh resolution orientation and design so why is turbulence a problem well cell sizes are getting smaller and smaller and smaller and that is a problem as we've discovered through over the last few years there's some pretty rigorous benchmarking testing outcomes not and not all done by us the one that really opened our eyes was the prison river model which was a regional model extremely well calibrated there's just very little doubt about the uncertainty in that model it was developed at 30 meters and then when that model was released to for public use um other consultants were making cut down models that reduced the cell sizes down to five or ten they were looking at infrastructure works within the river wanted faster runtimes so to make a little cut down model just to cover the area their study area and it's a good example where the cell sizes start to get well underneath the depth so the brisbane river flows at four to six meters per second 20 to 30 meters deep it's a big fast flowing system as a consequence they needed to recalibrate those cut down models and to do that they had to use non-industry standard meanings and values so that immediately rings alarm bells you have to start thinking about what's going on there and we suspected evidence was the cause but we couldn't confidently say that other things we've been doing or have done in the past is models of fluent tests often very very poor reproduction using industry standard parameters so that's you know we had to change the turbulence coefficients or things like that um and of course the other thing we've noticed is poor mesh size conversions can occur particularly once your cell size is much less than the depth so that's essentially what is happening here in the brisbane river reduces cell size down you get quite a big change in results and that's a big issue for quatering and flexible measures because they have a multitude of different cell sizes so our conclusion was we really needed to find a turbine scheme that is insensitive to cell size so why turbulence well da vinci and others were just fascinated by turbulence this is a sketch by turbulence and i think the thing that's to appreciate is the turbulence happens at an infinite scale from the very tiny molecules up to very large eddies that form so it's got a huge scale there's always turbulence happening a sub-cell and it's important because your surface elevation is dependent on your velocity field your velocity fields depend on your turbulence and turbulence affects how your momentum is diffused as water comes out of a bridge for example and then that affects your velocity field which affects your surface elevation so it's a vicious cycle so if you don't get the turbulence right you won't get the rest right when your flows are complex so on the right we have a image of some eddies forming and basically what we're talking about is the effect or the energy loss that's happening sub 2d cell due to that water churning around like that that's what we're trying to model and the traditional approach to uh is use large eddy simulations and this is the approach used by those pioneering people back in the 60s 70s 80s and the favorite approach was this megarensky formulation and that's fine for big large cells when you know that much relative to the depth but it is not designed when the cell size is much less than the deaths megarensky is for portions to sell area and as you sell area gets smaller and smaller it's magalinsky tends to zero turbulence state so what do 2d cells are becoming smaller cell sizes less than the depth measures that very sales size really need a cell size independent turbulence parameter and that's what we're about can we come up with a 2d cell size independent turbulence model so our approach was to test a range of models that were published they're the ones we selected the k amiga and k epsilon were knocked off early in the game um we benchmarked to think to known results so flume tests as in measured data from flim test real world data where they had low uncertainty such as the brisbane river where there's very very little uncertainty about the flows and the water levels and try and come up with an optimum parameter for each of these benchmark models for each of these different models can we develop a one-size-fits-all turbulence model the three models we started with was a right angle flume bend at kansas uni 15 centimeters across here very small uh measuring the head loss around the bend this is the band brake test of water flowing against the building used for the uk benchmarking test six is about three meters wide and the risen river so on 200 200 meters wide river but with very little uncertainty over the flows and the water levels measured along there so what we did for each of those models was to run them at multiple resolutions so in this case the right angle bend we've got two cells across there all the way to down to 6.5 millimeters which is a whole bunch of cells across here so we ran all the different cell sizes for each of those three models and now for each of the different turbulence models we're trying to establish what was the optimum parameter and so on the chart on the right we have the energy loss across that right angle bend the dashed lines are what were the most measured you know they had a range of measurements which fall inside here and this is the number of cells across that little flume model so from 2 up to 32 the black squares is the no turbulence zero turbulence model so as you would expect it gives the lowest um energy loss around the bend and you can see you start to get a convergence around eight to sixteen around this h cell mark is converging but at two cells is very coarse and it's too coarse to pick up the flow patterns and doesn't give a good result so we tried then increasing the constant value this is 0.001 and 0.005 the right answer is probably around 0.004 and what you can see convergence is attained around 8 cells across the channel and if you're going to use a constant viscosity model you need to have a value of about .04 to calibrate that model so we're going to play a game now which is which is the odd one out i can't get you to put your hands up on this virtual material but um i guess you have a good crack at that one so which is the odd one out so we have constant but this is the chart we're just looking at this is megarinsky 2d form of woo the 3d form of woo and prandtl now for those who are astute enough you'll notice that the old one out is smegorensky and that's because there's no matter what value you specify as coefficient you never ever get the right answer it converges down to the no viscosity case and that's because smankarinsky's proportion of the cell area and therefore proportional it tends to zero as the cell area tends to zero um constantly double o four woo value is for two d's in a round close to point five and so forth but all the other four schemes um successfully converge to an answer typically starting around the hcl mark bristol river similar story and no surprises smegarenski is the old one out the dashed line is the head drop measured across this part of the river and uh smegan is he you just using smackdown ski on his own will never get there and that's why twofold actually uses by default uh from the latest release smegarinsky plus a little bit of constant because you need a bit of constant when as this magnitude tends to zero um so constant i mean what's interesting here is you've got a value of 10 keep that number in mind the previous one had a value of 0.004 move to the values and so forth so just to summarize that testing um this 90 degree bend case dam brake flim test prism river and the optimum coefficients for each of those cases so basically to summarize cemensky is not an option um constant impractical because huge range in values you have to have a different value for every cell size and it's probably also depth dependent as well um richard e is okay some cell size dependency um prandtl is pretty good but it was very computationally memory intensive slowed down the simulations enormously um with 3d is a real sweet spot and we've done subsequent modeling as well which is just confirmed that a really good default value is around that six or seven mark so we've locked into blue3d as our default for the 2020 hvc release and it's really been a really good move so if we feel we're in a much better space now sub cell tubing so we're giving that one a tick we can really be confident where we're going with our cell sizes can now go really really small and and the depths can certainly exceed the cell size um mesh resolution orientation design so we let's have a look at those issues so the first of our new features is quad tree and that's really dealing with the last one which is 2d cell two fixed grid models couldn't vary their cell size well quad tree's changed all that it allows splitting one grid cell into four and it really allows you to fine tune your mesh resolution where you need it the really nice thing about it it's very very fast to set up literally minutes just feed in a road slayer gi layer of your roads and the rest is done for you so you just automatically generate your mesh something like that on the screen and away you go it just hooks into the pipe network system if you have one etc and what does that mean in terms of your simulations well in terms of cell count it really obviously has a nice effect we go from six million cells down to one million cells in this case of this study here um going through three meters to a three six twelve arrangement it's all about making parts of your model corsa where you don't have much terrain variation such as your floodplain or parks and gardens and so forth or up in the hills or areas that aren't expected to flood just make yourselves coarser a nice effect on memory from two gigs of gpu ram down to 0.4 so you can that that's a big saving if you've memory constrained and you also get a nice speed up in runtime as well so the last one is twice as fast more than twice as fast as the first one so it really is a sweet feature to use um yeah and of course you need to cross-check the results change so this is a change in results from the first level so no quad tree down to three levels of quad tree and 80 percent of the site um has a flood level difference of less than 0.1 and if you're comfortable with that you can definitely move to that sort of quad tree 3612 arrangement it's always worth doing that comparison do my results change much so yeah very comfortable with that sort of difference for that type of model sub grid sampling so this is the one that surprised us all so subgrid sampling is all about using the terrain information that sits with inside a 2d cell making full use of that train information conventional schemes use a single elevation per cell center or cell triangle or quadrilateral if you're flexible mesh and then some schemes like tifa also use a elevation on the cell sides so this so just sampling at those five points with sgs on we start to sample at a resolution that you can control typically there's no point going finer than your underlying dm or underlying terrain but in this case here we're sampling all the elevations on a 5x5 grid and so you're picking up the shape of the terrain underneath and what that means from a computational point of view is that for example if the water level in this cell is down here this cell would actually be dry at this point if you only had a single elevation model but if you've got sgs on this cell actually operates as a partially wet cell so only this surface area the blue surface area there is is wet and this cell face is flowing but not fully and this cell face is flowing it's almost like a cross section but not fully these two cell faces are of course both dry as water level rises these cell faces are now fully flowing and cell's around 85 percent wet and the other two faces are partially flowing so you get much better movement of water through your 2d system so let's look at some of the benefits of sgs and this is the one that i guess surprised us all the the problem of having a deep sided channel unaligned to your grid so this is this is the nightmare of course or the the pain for 2d fixed grid modelers and for flexible mesh models they have to cut a really nice mesh down through there if they're going to model it in 2d and the problem is that if you had a fixed grid model there angle to the to the channel to a rectangular channel you get this disturbance developing whenever you're around these sharp edges here so you can see the streamlines that are not smooth and that generates a bit of energy lost of course that world has been forced to take a bit of a bend and so you get a higher water level and you stop obeying manning's equation um the solutions to this of course have been the good old 1d channel cut through the 2d which has been done for a long time now it's very time consuming but at least you get a much more accurate representation of the conveyance down these channels but you do lose the benefits of a full 2d solution once that water starts spreading out flexible mesh so you need to spend a fair bit of time with sure quadrilaterals down here and they should be aligned with the flow and you need to have a sufficient of them to pick up that shape that's a very time consuming exercise plus you're getting very small cell sizes sometimes another option of course is just put more fixed grid cells across there but that's very quick to set up but your runtimes can become unworkable so we thought yeah that's what it is um it doesn't work very well we've always known that known that for a long time but we thought let's just try sgs see what happens and when you turn this gs on these cells here are now all partially wet so this cell is only tiny bit water flowing through it whereas this cell here a fair bit of water is flowing through it and you can see the stream lines are nice and smooth the velocities are nice and smooth we've gone we're not getting this disturbance and not getting his head dropped so it really did seem to do a really nice job of handling this situation so they started to benchmark it let's take a really basic case a rectangular channel sloped according to the mannix equation so that the depth that comes out of the model should be exactly one meter um in these charts you see the solid line is water surface dashed line is energy but the theoretical answer is exactly one meters depth so if we have the grid perfectly aligned to that rectangular channel the really nice thing is we perfectly obey available equations that's that was very comforting just to get started we actually do a bay mannix equation but as we rotate that grid and this is without sgs you'll see how the water disturbances are creating you know a bit of noise and you can see that in the shape and the undulations along here and we start to pump up the water surface because the water is finding it hard to work down there and you're starting to deviate from the theoretical answer and if you rotate more in this case we're really getting to probably 15 error to the theoretical solution 45 degrees is a bit better again because everything's sort of straightened up with the stream lines but we're still getting a sort of uh some effects along the edge there and it's not perfect so we've always known this and that's why we've always said you need to have lots of 2d cells across a deep sided channel so you don't get these edge effects so let's turn sgs on see what happens yeah we know that one works that's all good 15 degrees spot on and what you should realize these cells outside the rectangular channels they're actually partially wet cells they're not fully wet and these arrows here just artifact or the displays of the display issue so those cells out there are partially wet 30 degrees on the money and 45 degrees we this is we're just so excited about this because it really does open the door for using fixed grids in deep solid channels at any orientation let's see what happens with different resolutions so this is without sgs so 50 meters pretty coarse water's finding it really hard to get down there you know choking down to one cell here um and of course a terrible comparison to the theoretical result if you go to 25 meter resolution is better 10 getting close and at 5 meters it's actually starting to get something that you might feel comfortable with for reproducing the theoretical solution let's turn sgs on and we're going to start with five meters you know that works yeah that's perfect 10 meters perfect 25 meters still perfect and 50 meters it's also pretty much on the money so it's also saying that we can pretty well model a channel like this with only one or two cells across it if you're using sgs absolutely fantastic news very very exciting let's uh let's take a more rigorous test so this is a uben test so measured water levels on the outside and down the middle um and on the inside so the bit of surf elevation on the outside it's pretty well straight down the middle and a little bit of under elevation on the inside so this is from a typical fe flex for mesh model that we made some years ago is the quadrilaterals nicely shaped and it gives a pretty nice reproduction of those measured results and spot on with the upstream water level a little bit under on the super elevation side so we've always known fixed grid models just can't compete with that this is the result we will get so you know as the water goes around that bend you see the distortion you're starting to see streamlines divert and you get quite a messy result and you're overpredicting the head drop quite substantially due to all the noise that's created around the outside of the bend so yeah we've always known that but let's turn this gs on so now we have a whole bunch of partially wet cells around the edge and the velocities are now nice and smooth streamlines are smooth and we get a really good result and and this has been no this is just using the default parameters in the new version of two flow with the new tokens this absolutely spot on um and this was actually if you like a real like a validation of a lot of our earlier work it's a bit a little bit wobbly not quite as smooth as a flexible mesh one but it still is producing the right hair drive um absolutely brilliantly so let's look at cell size effects so this is 34 centimeters quite chunky cell size across there it's a bit noisy in results but it still produces a head drop correctly upstream which may be the only thing you're interested in as you get finer it smooths out 10 centimeters and then for 5 centimeters it's giving as good a result as the flexible mesh model really really sweet outcome so what that's telling us if you're just interested in getting the right hair drop you can even model something like this at a very coarse cell size as long as you're using sgs to pick up that make create those partially wet cells around the outside very nice so what are some of the benefits of quad tree and sgs i'm just going to go through a couple one is the improvement of in urban environments so here we have a intersection model with a single domain five meter grid we've got four pits feeding into a pipe network on the right we've got a quad tree model so we've reduced those cells down to a much smaller cell just along the road gullies and and you know course sell out in the parks and gardens and around and then down to 2.5 and 1.25 for the roads let's see what happens at the say the pit up here so these pits each have a depth discharge curve associated with them for the single domain five minute grid the depth of that pit at their 2d depth is taken onto this chart interplay into that curve and you're getting a peak flow of around that much but with quad tree or sgs or both you're going to get a much better definition of the depth at that's at that pit so that's picks up a bigger depth which then in turn picks up a bigger flow and pushes a much more accurate level of flow into that pit this has huge benefits for modelling in urban environments it really does allow you to get a much better handle of that pit pit flow capture let's look at some catchment modelling where sgs and quattri are having really really big positive effects is in direct rainfall cash flow modelling as i'll show so we're just going to look at the river tamar in tasmania this is a quad tree model you know 10 meters for urban areas 20 minutes for rural areas and 80 meters up in the higher ground we have a have a a low a high res model the 10 20 80 and a low res model which is double that which is simply changing one number and two flow let's look at what happens so we're going to zoom into a small part of that catchment and look at the flows across these two major rivers so without sgs what we're really interested in is are we getting convergence due to changing our cell size and this is the hydrograph at this south ask and the bottom one is the hydrograph at macquarie the gray line is the 80 40 2010 um flow and the yellow line is the 160 80 42 so doubling the cell size and they are very different so that's saying we have not got cell size convergence our cells are too coarse we need to keep making them finer and finer same response at macquarie so no good we would have to keep making ourselves smaller and smaller and then you have a model which you wouldn't be able to run let's turn sgs on this is a result with sgs the orange is with the 80 40 2010 and the blue is 160 80 14. not identical but very very similar and what that's telling you is you've got convergence you can be very confident that um you can change that cell size even smaller and you'll get a very similar result it's also very different to this uh to the no sgs case which means that you really would have to make your cell size so small to get up to somewhere like here so this is having huge benefits for models which have cell sizes which are much coarser than your underlying terrain resolution as is the case here yeah so big tick for that okay to wrap up in conclusion just to reiterate basically 2d hydraulic modelling it's um it's been on a pretty amazing journey for the last 50 years and as i said it's sort of transitioned from a few 2d schemes that were mathematically elegant and you know very accurate for what they were and now today we have a wide choice so be very careful about your choice of scheme but the even the nicer schemes still have well we identified they have deficiencies which we've largely known about for a long time however as we've seen cell size independent turbulence model has really solved on the issue of making your soil sizes smaller and smaller and smaller particularly when they become less than the depth this magnesium is no longer applicable for that case it's only suited to the large eddy simulation modeling and it now means in two floor hvc you can model from a flume scale to something like a large river scale without having to even think about changing your turbulence parameters well if you did change them it would be a very minor change very exciting quad tree well that's the easy one to understand and it's just so sweet to be able to just drill down and put a finer resolution where you want it yeah it's just so easy to do it it's a matter of minutes to do that type of thing very sweet and soft green sampling this is the um it's a surprise for us i guess it really has allowed us to confidently say that fixed grid models can be rotated in your orientation you can model deep sided channels such as a concrete urban concrete drain at any angle and you can also model it quite a bit coarser than what you were previously that's got i can foresee and we're doing a benchmarking of that for the throsby cash in newcastle where all the 1d channels can be replaced by the uh as 2d very very interesting space we're moving into very exciting and it has big benefits for direct rainfall cache and modeling you know mesh size convergence is absolutely brilliant so look above combination you can have all those three things together or you can just pick and choose but the valve combination is absolute game changer and look from my point of view the future is looking really bright for accurate 2d modeling so thank you that was fantastic the depth of experience coming across here and watching a q a line go at the same time it's just a huge area of study which you've thrown yourselves into bill chris greg the questions are are coming in click thick and fast so we'll get straight to them but you know what comes to mind straight away if you can't measure the behavior of water you certainly aren't going to be able to manage it and this is this is serious in-depth measurement going on here as you'd all agree i'm sure thanks everyone for your questions it's been absolutely delight to watch them and thanks chris and for greg uh both uh hammering away at the answers there look there's a couple coming through here right now should we should we get onto those straight away thank you everyone for upvoting them to the um to the highest to the last prioritizing for us so let's go over this first one i think subgrid computations are already implemented in the hec-ras 2d am i right if yes how different is the two-flow approach from the hec-ras 2-d um i'd i think yeah hec-ras 2-d um and all credit and they were the sort of pioneer in that i guess what i would i haven't seen is any the sort of benchmarking in some aspects but yeah certainly they they do have that feature and it's a credit to them it's the one thing they've forced us to um take up um but ultimately it's about benchmarking and whether your scheme is you know all schemes will behave differently we'll treat this differently so that's what i would strongly stress yeah no that sounds good thanks a lot uh boro dwelch that's great uh turther asks this model uh using the step gradients uh can be used for high bank cutting and high erosion mountainous areas yes yes uh yeah it certainly could because well hpc uh will push into any supercritical flow automatically um and sds will give you that definition in terms of you know if your core cell sizes are a bit coarse up there or you can't push a quad tree cell size in sgs will allow you to really pick up those uh cuts we're seeing that over and over again with these steep catchment models i'm hammering through these because i can see from our clock where we're into the last 10 minutes but the questions are coming through thick and fast and it's great i can see the questions also coming people um you're putting on the chat line um can you can you um everyone uh go to the q a so we've got a bit of a record of some of the questions and comments but thanks everybody for joining in it's fantastic oh martin jacobs has got a question here in your benchmarking did you use constant depth manning's uh in or depth fairing mattings in how would the different approaches affect the outcomes yeah so no they were all um constant depth management those all that benchmarking there's no death varying meanings in uh look i think there's certainly a place for depth-bearing manning's in which tufo supports um but i would say that this needs to be improved guidelines and it can be a mechanism for calibration but none of that work that i showed you there today use depth screen meetings then yep that's great look um what we're going to do right now i want to include greg and chris with you bill uh i can see that um greg's uh answering martin jackson's uh question at the same time so i'll go chris greg come on screen now and and let's let's uh chip into this discussion going on here it's just fantastic uh where do you want to go with this first up who wants to go first as you saw greg you wanted to answer that one i'll just answer did you want to comment on that further yeah oh that was the yeah um i was about to answer that one but then i heard you talking to us i thought maybe yeah look i think uh all of the calibration work and testing work was done with fixed manning's end but we do have uh we routinely run a very large suite of models testing our latest uh development versions against previous versions and many of those models have have got depth varying mannings and values in them and certainly they all perform equally well to uh to previous releases so um um don't have any uh there's no red flags to me about the uh the validity of their approach between fixed mannings and or death varying management all right let's keep going and um uh there's a lot to get through here we may not get through them all but we'll hit them as you've um prioritize them yeah chris chris you're wanting to answer tim craig oh sorry i'll go up to the top i was scrolling down through the other questions you've marked yourself on that one yeah oh hi bill great talk how would you deal with the typical 3d problems i piped it's a great question tim um look um if basically if it's really got strong 3d flows you may need to resort to some sort of cfd modeling to establish the losses through that space and then you can you can model that in 2d or one day by applying additional energy losses and there's something that's quite commonly done in two flow world generally speaking um something like bridge losses uh yeah you do need that additional energy loss and that's sort of how we model it in 2d but if it's a really complex problem you may need some other higher level modeling to handle that that's great okay we'll look just so we've got uh keeping track of what's going on here those two questions um from martin and tim craig sorry greg and chris if you could just click them off as answered or i'll send that one from tim across there's one here from i'm sorry i can't pronounce your name rita hutter so memory and cpu comparison between uh change in cell size plus sgs versus um just the change in in cell size uh i got that so i see riyadh memory and cpu cpu comparison between dx and sgs yep did you want to address that chris um yeah look there's there's put probably about it's a ten to twenty percent increase in memory requirement for the sgs bill greg is that is that fair to say yeah that's fair the big gain is when you move to that quad tree mesh from a single resolution model there's a video i posted on our linkedin and youtube channel about a week ago where we implemented three levels of quad tree nesting and we were able to reduce the ram requirement down by about a factor of 10 um by introducing quadtree there's just a much more efficient way to get the same level of accuracy result great that's good should we go back to the top uh gentlemen do you think uh back up to the um and i'll just add trevor that we plan to take all these questions and try and come up with an answer for all of them in a written response as well so don't if people don't get answered though we'll try and get an answer yep no that sounds good that's great actually the top one's a really good one bill sorry i'm sorry to cut in there trevor that's all right on the recording the reason i'm probably making a point of reading them out is on the recording when people come back to these which is where most of these webinars end up on a web on a youtube website or aws australian multiple website and people watch them they don't actually get to see the if i'll get this right they don't actually get to see the question so here's the reason for sorry about making yeah so the tough one um if your cell size is 10 meters but you have a small ridge line within that ten uh will uh not define with z shape or two flow past water through the ridge line uh hds are turned on yeah look potentially uh it will because if your cell faces either for either side of that ridge line they'll be sampling along the actual face line the storage will be well represented because it does the whole cell but the actual cell faces is where the sampling happens for the conveyance so it's still really important to push through your brake lines or road crests um ridges or levees um because you know they will control the flow so you basically just feed your z-shaped layer in of your road crest or um levy in after you've done your secret sampling of your underlying terrain that's great any comments on that one chris greg one sentence uh look i just say that when it comes to the subgroup sam sampled models it's probably more important than it was previously to introduce those ridge brake lines yeah all right down to federico could you please over elaborate on reasons for not turning sgs on every single model yeah there is actually um there is a penalty in run time i think it's 10 to 30 percent it's not great it's not a big change for all the benefits you get and and other things are increased as well like memory but the reason we didn't turn it on by default is that for like a lot of models if you if they've got a small cell size relative to the underlying terrain and they're well designed models you won't see a big change in results but some of those catchment models you'll see a very big change in results so we really want people to turn it on particularly with legacy models i would say that it will be the default in a future release but at this point in time we're still all learning about the benefits of sgs and you know it's we're circulating that feedback and i would say at some point in the future it will be you can turn sgs on layer by layer so some layers you can have it on some days you can have it off so it's also flexible in that manner as well all right done with that one alexander karash does bmt or 2flow intend to release sgs verifications or commentary for dam break case studies such as duncan kits has done for bmt uk in the past yeah yeah that's actually fine i mean that that i think probably referring to the multip um damn calibration damn break calibration at duncan put ups uh last year i think or earlier this year um yeah that's all public data so we can easily provide that information or point you to it and duncan is actually part of our team so oh yeah he is he's representing two flow there yeah yeah so yeah duncan's our uk person so um yep yep all right where do you want to go keep going with this one that's prioritizing it's not too bad uh frank fernando how does how sgs and quad tree how does s effect map out asc dot flt yeah i didn't talk about that today that's almost another whole presentation in a sense you have to be very careful with um so what we're actually developing now is a high res output so um sgs has you've got that turned on you've got lots of partially wet cells the really nice thing is that your flood surface will now cut right into the high ground so if you subtract your flood surface from your dem you do not have to do any buffering of that flood surface you get a really really nice high resolution depth map and other things so look it's a tricky one we like to output how the model sees it because as models we need to understand what we're modeling but there's some really we've got some really nice post processing already available in in some of our utilities where you can turn those uh sgs and quad tree outputs into really nice high-risk output but you um yeah it's a very good question yeah um who is that frank frank if you're just looking for information on that you'll find a really nice write-up on the wiki if you do a search in the ask you to ascii utility page and it's at the bottom under the the remap section that'll show you how to get some really nice results yeah and do and do let's uh out of session after this webinar's over let's be in contact to um to wrap up some of these order take them further uh matthew christens uh good question hey team excuse my ignorance but for a narrow and deep meandering channel let's say 10 meters with utilizing sgs grid cell size of say 40 meters will it provide an inaccurate representation of average velocity in cells that straddle both other bank and in channel areas um yeah so a good question matthew look when the water is just confined to the 10 meters width so those cells are all flowing about a quarter wide um the velocity pro it would be probably rough but it'd probably be okay but as soon as it breaks out above the terminal bank of course the it's like a 1d model you'd have a depth and width average velocity for the whole um including that bit of flood plain that's either side of the 10 meters so um yeah you want to be very careful i think modeling at that resolution is that sort of critical channel you've been wanting to put a finer mesh uh quad tree mesh down there to get better handles on velocity if i have a dem with one meter grid size and two float model with a one meter cell size would there be significant difference in the results very little difference in results um you get some maybe some slight differences but if your 2d cell size is commensurate with your underlying resolution of your dm there's no point in using sgs basically so baron kumaresh has asked if we are building a pure td 2d model where the imagery is limited by the lidar water surface elevation right yeah how do we handle this do we need to condition uh burn the bathymetry the dcm for a pure 2d model oh yes short answer that one is yes i mean and that's it's i've seen so many models where people have just taken the lidar they've not gone in and burnt through the symmetry either from other data or have it surveyed and you just get nonsense results if you don't get your topography right in your main flow channel the rest of your model is going to be hopeless yeah all righty uh if sds is turned on we'll have the effect of presenting on the presenting flood mapping results considering that the larger grid sides will be used to present the flooding but the whole cell is not really yeah yeah good question um yes this comes back to my answer a little bit before is so basically you'll what comes out is that the whole cell is wet but it's only computation is only partially wet but the nice thing is it if you subtract that water surface from your underlying dm you'll get that high resolution flood mapping come out and as chris was saying there's some very quick ways of doing that as a post-processing exercise and it's we're going to be filling in to hopefully by the end of the year the option of producing that out directly from to flow as the simulation progresses yep so keep tuned uh carlos has asked are there any plans to revive revive the z shr lines with sgs that are useful to certain times of submergence and will be great to see them come back oh i guess this is a z-shaped evacuation routes um feature in two flow um good question evacuation risk have been now built into hvc they're going to be in greg can you i'm just trying to remember yeah they'll be in the um i'm pretty sure they'll be available in the uh our next update this year the uh the ac 20 ac release they should be available um and again with sgs uh yeah it all um that's all in their works because they're based on water elevations or depths over the minimum um the minimum cell depth or phase depth for the particular faces so yeah that's all been put in and it's uh it's been tried and tested already i thought so and good add to that one there's also a spatial result if you turn on the times output data type um and you can set up a wet tolerance as to when it will trigger whether something's wet for the first time or not and when we answer this in writing um i'll share a link with you to an example data set that we have on that yeah look thanks very much for that uh for that offer to to continue this conversation after the webinar chris that's really good bill greg that's great um look let's just ask can two flow output as a tiff in future releases to include compatibility with qgis um yeah the answer that is it's on the list um to provide or you know a wider range of um output formats so but not not at present but it's definitely on our development list that sounds good okay the last question from vector test interesting question how is the recent advancements helping how will that help in agricultural water management probably another another couple of webinars there yeah look i think um yeah i'm not quite sure the exact answer there but uh um we are exploring the possible with the benefits of quality and sgs on the whole of catchment modeling and we also have quite a range of infiltration models and some basic we're looking at building some basic groundwater functionality potentially we can start looking at smaller flows as being accurately modelled but early days yet on that well look there's been an absolutely a feast no no kidding it's been absolutely brilliant i appreciate so much for taking the time on this bill chris greg he couldn't possibly put it in a book it's too much but a big shout out to joel ortman at australian water school and to jess burgess really appreciate and you can't possibly have these webinars without two people doing the backup work that they're involved with you can see here the on-demand is coming from the australian water school 1d hec-ras 2d hec-ras and modflow in the middle there three webinars soon coming up in hec-ras hardware selection and python scripting and not just trevor so greg greg's giving the hardware selection one so if you really want to learn about uh hardware craigslist that's the one to watch 21st of october and in the webinars the middle one that'll be greg's well done yeah and can't just one more plug there trevor in about mid-november i'm giving one that was on let's focus on 2d cell size selection there's a few questions on that topic today is that question in mid-november uh hopefully yeah in the show size probably after the fourth um yep i think it'd be the 18th or 20th thereabouts fantastic twentieth look out for that once again thanks very much bill thanks very much chris greg uh jess and joel it's been a really pleasure to work with you all and i hope we can do it again one day thank you everyone for joining us the many people from around around the world thanks so much keep tuned on the australian water school website for future webinars and training lovely to have you all together bye for now thanks trevor thank you you

Channel: Australian Water School

Views: 2,217

Keywords: TUFLOW, BMT, Australian Water School, hydraulic modelling, Bill Syme, software development, computational approaches, simulation speed

Id: eJJtsS91XNo

Length: 61min 48sec (3708 seconds)

Published: Wed Sep 16 2020