 I'm very pleased to announce Andy or Andrew Nichols he had this really impressive paper about a year ago on his h-star model and we started communicating with him a little bit and like tried to like invite him as an setup and highlight on his research and then he emailed back and said like hey we've got this really large nerd project funded and we seized that opportunity to invite him here and like sort of showcase his work as it's been going on at Exeter and like with lots of collaborators and it will be continuing and we hope to like hear from him in our community with CSDMS so Andy floor is yours. Okay thank you and thanks James for the invitation and thanks also for helping out with that grant application and providing the letter of support which ultimately helped us out there being successful so yeah I want to start I guess first by acknowledging the very large number of people you can see here who have been involved in some of the work I'm going to be talking about. I'm going to be thinking about that you can see the evolution of large river floodplains talking about some recent progress in that area some problems and some issues of uncertainties to try and fit in with the theme of the meeting. I want to think about a couple of different spatial and temporal scales okay so firstly just thinking about the smallest of these spatial scales I want to think about what's going on within individual channel belts okay so we've got a large meandering river here if we want to think about the evolution of that we need to think about sediment transport in the river channel migration overbank processes bar and island construction and sort of evolution of the system over at least I guess several hundred years okay that's one scale I also want to think about a larger spatial and temporal scale okay so we can we can then go outside the channel belt so we're looking at the Rio Bene there in Bolivia and we've got the active channel belt at the top of the image and then we've got the larger floodplain and once we start thinking about that sort of spatial scale we need to think about how sedimentation within the channel belt can lift the channel belt up above the floodplain create an alluvial ridge and therefore ultimately possibly drive evolutions and a relocation of the channel belt onto the floodplain okay and lead to those sorts of processes of compensational stacking that we heard about yesterday okay so those are the two scales I want to think about I just I wanted to point out before I go any further of course that you know I'm referring there to meandering systems of course there are a whole range of different channel patterns that we might find within the channel belt not just meandering so we've got braided systems here like the Brahma Putra Gemuna and then also some really depending on your perspective beautiful and actually horrible if you're a modeler situations you know where we've got this sort of multi-threaded meandering system we're not even really sure where the channel what's channel belt and what's floodplain but we know there are some complex interactions to try and deal with it okay final bit of context obviously there's a very wide range of sort of river and floodplain models that have been developed in the past and I just want to try and draw a distinction between them so on the left hand side there we've got a couple of models that are used for looking at very long-term river evolution channel migration floodplain sedimentation and and channel belt of ocean and in these sorts of really long-term models in the past they haven't explicitly included any representation of hydrodynamics on the floodplain because the because they need to run for such a long time it's just computationally not possible and so consequently they need to parameterize some of the processes that are driven by hydrodynamics such as overbank sedimentation and that's whether the sort of classic exponential decay law comes from which is commonly used in those sorts of models for modeling that the decrease in sedimentation is as you move away from the channel or the channel belt and then on the right hand side we've got some models that are typically applied of a much shorter time in space or smaller time in space scales that do have a hydrodynamic driver and think about processes such as overbank sedimentation or channel morphodynamics you know and if we start to put those sorts of ingredients together then we can think about the sort of feedbacks between channel and floodplain processes within the channel belt that I referred to earlier okay so those are the two contrasting approaches to modeling and I want to think to start with about that sort of what I call a higher resolution approach although actually having seen the the St. Anthony Falls modeling yesterday they are my ideas about high resolution have been slightly revised there certainly are millions of grid cells in any of these models but anyway so you know there are a whole range of models and I've listed just a few of them now on the right hand side but there are many of them that take similar approaches to these sorts of problems so they're implemented on structured or unstructured meshes maybe grids or curvilinear meshes and they solve partial differential equations for fluid flow and sediment transport to model patterns of erosion and deposition and therefore channel and flood plain evolution through time so I'm going to talk about some results from a specific model but I think what I'm trying to you know the main points I'm trying to make a generic and they run across models I'm not talking about this one particular model really but I'm going to show you some results from a particular model no equations in this I'm afraid so there are a couple of references at the bottom if you want to see the basis for this and so the model I'm going to talk about it involves a sort of coupling of a hydrodynamic component and a sediment transport component the hydrodynamic components based on the 2D depth average shallow water equations with a correction for secondary circulation the sediment transport component just involves two grain sizes so there's a sand component and sort of total sand transport rates that suspension plus bed load and modeled with the Englund Hansen law and the sediment moves in the direction of the mean flow but then the direction of sediment transports adjusted to account for secondary circulation and the sort of gravitational deflection of sediment on lateral side slopes so that's the sand component there's also a finer silk component that's handled using an advection diffusion equation and all of these sorts of process representations are fairly standard I mean the models I referred to they vary in their complexity and their sophistication but they're all using these sorts of approaches within this model there are a couple of other aspects of the process representation which is in fact extremely simple so all of the grid cells are defined as either active channel bed or floodplain and there are very simple rules for how those cells can change from one class to another so floodplain cells which are vegetated can be converted into active channel cells by bank erosion and the rates of bank erosion are just modeled as a function of the sediment transport rate in the near bank cell the sediments removed from the floodplain cell okay but the floodplain cell is not actually lowered until sufficient sediment has been removed from it to bring it down to the level of the channel bed in the cell adjacent to the bank and that keeps a fairly vertical or fairly steep bank as the bank retreats there's a very simple vegetation component which basically just converts active channel cells into floodplain cells when cells have not been inundated by a small specified sort of threshold depth over a certain time period now those parameterizations are extremely simple but they're enough to give you some stabilizing effects for vegetation and to give you migrating bank lines which are necessary to produce some of the sort of dynamics that I'm going to show you so I'm just going to start by showing you three simulations and all driven initially from a straight channel the flow is from left to right flatbed although there are some small random elevations elevation perturbations there a series of hydrographs run through here although actually the movies are only made up of the low flow images because it gets quite hard to follow when the water level is going up and down and really all I'm trying to do here is illustrate that there's a range of channel behavior that's produced by differences in certain controlling factors so for these simulations there's differences in the rate at which the vegetations established differences in the strength of the banks and differences in grain size slope and bed roughness which control the mobility of the sand so I'm just going to run these things they all start off in a fairly similar way with sort of unit bars migrating through the channel but they quickly then take on different forms so up at the top we've got a simulation with with strong banks and slowly growing or vegetation which establishes slowly on bars and the result of that is that the channel doesn't widen there's no space for the bars to establish or for the water level to drop at least which is necessary for the vegetation to become established and so you maintain this sort of dynamic braided form whereas in the bottom simulation that's that's got more rapid channel widening and also rapid vegetation growth and that creates space for bars and flood plains to form and we end up with that sort of multi-threaded meandering system and also you might have noticed in that many branches forming so there's a tendency for channels for bifurcations to occur and for some of the bifurcates to ultimately be abandoned and the middle simulation contains aspects of both of those sorts of behavior okay now what's driving that it's partly those simple parameterizations of bank strength and vegetation growth but there's also an important effect of sediment mobility in this model so what we're seeing is that where sediment where the sand is more mobile the gravitational deflection of the sediment down lateral side slopes is weaker okay and that reduces the amount that sand that's being transported away from bar tops so the bars grow more rapidly and that tends to drive the production of these bifurcations in the channel and I'll come back to that that sand mobility mechanism in a moment okay just one other simulation from this I want to show which is not a movie but just six snapshots from a simulation where the only real difference here between this and those other simulations is that the vegetation grows very rapidly in this simulation so it starts off actually with a straight single thread channel which quickly then gives developed this sort of sinuous towel wig which translates downstream the the individual bends then start to amplify you can see that some sort of quite nice scroll bar topography on the inside of the bends and because of that scroll bar topography during the high flows the water is focused across those shoots on the scrolls and that then leads to cut-off and you can see there's a couple of cut-offs in there and also we see some sort of reactivation of channels on the on the floodplain okay so I guess what's the take-home message here I think we've got to the point and maybe not just in this model but with other two-dimensional morphodynamic models of this type where we're beginning to be able to simulate the full range of channel patterns or maybe that's maybe I go too far to say the full range but certainly a nice wide range of channel patterns that we see in nature okay so that's obviously a positive but there's considerable uncertainty still in the process parameterizations here and I try to pick out what I think are the key issues that need to be addressed and many of them relate to that idea about sand mobility and how that could be important in controlling bar growth and by bifurcation dynamics so I've picked out a whole series of things that are important in driving those processes so whether the sand is moving as bed load or in suspension how the direction of sand transport deviates from the mean flow direction which is controlled by how secondary flow and that sort of gravitational deflection mechanism how they've been parameterized another mechanism which is actually not included in this model but it's included in other models is the the adaption length of sediment so overall over what distance does it take for the sediment transport rate to come into equilibrium with the transport capacity of the flow that's potentially an important effect and then there are also all sorts of effects that relate to bed form so spatial and temporal variations in roughness and the fact that the bed forms are actually what's controlled in the topography the local topography and therefore they should be influencing the steering of sediment and none of those bed form effects are incorporated in the model at the moment because the these sorts of two-dimensional models only represent the mean bed topography not the local bed forms so that all sorts of uncertainty in the process parameterization what that means is that although these sorts of models can represent a wide range of channel patterns if you wanted to use one of them to simulate a specific channel and you knew the hydrologic regime and the sediment supply regime and the slope and those sorts of things and you were to use the model to simulate those things you would probably find it didn't simulate the right channel pattern and you would need to tune the calibration to produce the right channel pattern okay now that's a significant issue but potentially more significant if you want to then use these sorts of models to investigate how rivers might respond to environmental change you want to know that the the sensitivity of the model to any sort of environmental forcing is similar to the sensitivity of real rivers and we don't yet know that because of the uncertainty in this sort of process representation that's an issue that needs to be addressed partly through model development aided by more high resolution filled datasets and also possibly by carrying out numerical experiments with high fidelity models with 3d CFD models the sorts of which we saw some of yesterday and that could help improve these parameterizations one other effect I think that's worth mentioning here as a problem with boundary condition uncertainty as well so at the upstream inlet of these sorts of models you need to introduce some sort of perturbation so if we look in anywhere say you know in a braided channel halfway down the reach we'll see there are bars moving around there are channels are moving around there's a certain amount of noise in the system that's just a product of the internal dynamics and you need to introduce that noise at the inlet to mimic that effect if you don't or if you introduce a very weak inlet perturbation as you can see in that sort of second braided river simulation there then what happens is the channel stabilizes and then we have a low braid intensity with large stable islands whereas with a strong inlet perturbation we get a much more dynamic channel with with smaller braid bars and there's a similar problem with meandering channels you can see the example there at the bottom okay so on the on the right hand side of that meandering simulation we've got lovely high sinuosity meander bends but on the left-hand side near the inlet we've got a really messy flood plain with all sorts of abandoned channels and what's happening there is that the the periods periodicity of the inlet perturbation is out of phase with the the translation rate of the meander bends and that's driving cutoffs and channel abandonment near the inlet you see exactly the same thing happening in experimental attempts to simulate meandering channels okay now what I want to do now is move on and think about things as a larger spacious scale so move outside the channel belt or think about the interaction between the channel belt and the larger flood plain so I just going back to that issue that I mentioned previously of these models that are intended to do that and to look at things like long-term flood plain evolution and alluvial architecture and the fact that they're typically not driven by flow hydrodynamics so an example that's relevant to what I'm going to be talking about the model of model of how Alan Howard that uses the Johansson and Parker meander migration model and then a very simple parameterization of overbank sedimentation not exactly this exponential function basic but basically an exponential function so you've got declining deposition rates with distance from the channel and of course the key question there is what's the decay rate okay and it should be a function of grain size but it should also be a function of something like the advection velocity of the water across the flood plain and highly uncertain what value you should use for that parameter so is it possible to come up with a slightly more physically based approach of doing that sort of thing so what we've done is we've taken a one dimensional model for the channel and a two dimensional model for the flood plain and coupled them together both the 1D and the 2D scheme solving the shallow water equations and an advection diffusion equation for sediment transport what we want to do ultimately is run this model over very long time period so thousands of years so that's pretty computation intensive so we need to we need to make the model more efficient and one way of doing that is implemented implementing it on a grid with a variable resolution so we use a quadtree grid and the other way is to simplify the numerics a little bit so whereas the other model was a was was second order accurate this is only first order accurate so that these things speed it up a little bit now we're applying the model on the Rio Benny which I showed you earlier to try model patterns of sedimentation and think about total sediment flux from the channel to the flood plain I'm just going to show you a very simple example so you can get an idea of the sorts of spatial scales that we're working at so this is a hundred kilometer section I'm just going to run a simple six month hydrography here's only a few time steps and you'll be able to see the sort of pretty coarse detail that we're working at so you see that those patterns are suspended sediment concentration so you see the flood the water moves out onto the flood plain via very localized breaches the sediment moves out onto the flood plain then the sediment concentration declines basically the flood wave has gone through there and the sediment concentration has declined because either the sediment has been deposited always drained back into the channel and then we've got water left on the flood plain now if we run that model over many floods we can start to sum up and get an idea of average average or mean annual rates of overbank sedimentation so there are a couple of examples there and there are all sorts of factors contributing to uncertainty in these sorts of things that I wanted to highlight and the first one in these sorts of environments is the DEM so we're working on a large tropical river where the only information we've got to define the DEM is SRTM which is pretty problematic because obviously that we need to then remove the vegetation from the SRTM which is not straightforward and we can use various Landsat vegetation classifications and information either remote sensing information or field surveys that we've got of vegetation heights to try and remove that vegetation but there's many different ways that we could do it producing many different DEMs leading to many different patterns of overbank sedimentation so that's a potentially a significant source of uncertainty at least in the spatial patterns of sedimentation if not in the total flux of sediment to the flood plain possibly and one of the ways that we're trying to constrain some of the uncertainty is to compare the model results with some measurements of overbank sedimentation so we have a set of considering the size of the flood plain in hundreds of kilometers in either direction we're given that we're using sort of 1.5 centimeters diameter sediment cores there's obviously there's some slight issues there we've got we've got sort of several hundred cores we're using about a hundred for this comparison here and because there's so much spatial variability in these processes in the field what we've done is average sedimentation so that we can look at a sort of or average core estimates of sedimentation so we can see the sort of trend in the average sedimentation as you move across the flood plain that's what those blue boxes are on there and the numbers represent the number of cores that are involved in the averages and then we've got a sort of envelope on there which is giving you a very crude idea of some of the uncertainty when we run multiple model runs with different DEMs different parameterizations of roughness different parameterizations of sedimentation we get a sort of envelope or envelope of possible sedimentation results and now this is by no means a rigorous uncertainty analysis we need to do that yeah but it just gives you some idea that there is considerable uncertainty here kind of interestingly you can see the red line on there and the yellow line they represent model runs with different parameterizations of overbank sedimentation and what we find is that when you change the parameterization of overbank sedimentation not surprisingly it has a significant effect on the gradient of deposition across the flood plain and the implication of that is it would have a significant effect on the development of a topographic ridge over time it doesn't have a significant effect really on the total amount of sediment that's being deposited on the flood plain really what seems to be controlling that is uncertainty in the boundary conditions of the flood hydrographs the sediment load and also the representation of sediment exchanges between the main channel and the flood plain okay the last thing I really want to mention is we then try to take this sort of model and think about how we can go back to these models of long-term flood plain evolution and ideas about topographic ridge construction which might ultimately lead to evulsion so what we've done is taken our hydrodynamically driven sedimentation model that I was just showing you the results for and we've then coupled that to a very simple meander migration model and we start these simulations with a flat flood plain and a straight channel and we allow our channel to meander and for sediment to be deposited on the flood plain through a series of floods because we're interested in the formation of an alluvial ridge we also need to enforce some aggravation of the channel so we're specifying a main channel aggravation rate that's just being applied to the 1D component of the model not on the flood plain and you see we've run this at the moment for several hundred floods and over time our alluvial ridge develops so what we then want to do is think about whether we can look at the results of this model and interpret them in the context of the sorts of simple modelers of flood plain evolution and alluvial architecture that have been used that don't include any sort of hydrodynamic component and think about whether this model tells us something about the assumptions of those models that maybe aren't right or need to be improved or the parameterizations in these models and there are three things I think that are worth mentioning at this stage this is pretty early stage of this work but worth mentioning at this stage in the context of that so as I said these models that typically don't include a hydrodynamic component they represent overbank sedimentation by assuming it fits some sort of exponential decay law and the big question is what's the decay coefficient is it a constant does it change through time what really controls it when we run this model what we see and I've just picked out results from one of the simulations here is that broadly speaking if you average the spatially distributed sedimentation patterns they do follow an exponential law however the decay coefficient changes through time so what is happening here is as the alluvial ridge develops the conveyance of sediment away from the channel to the flood plain becomes more efficient and the decay coefficient goes down and that would introduce a sort of negative feedback whereas the alluvial ridge when the alluvial ridge grew it would begin to grow more slowly so that's potentially an important effect which is not incorporated in these sorts of models at the moment but could be parameterised using these sorts of results possibly okay there's a second interesting effect if we look at the we take our alluvial ridges from a whole series of simulations and each one of the points on that graph represents a different simulation if we take our alluvial ridges we can calculate all sorts of topographic indices from them because those are the sorts of indices that are often used to predict evulsions things like the the super elevation of the of the I don't need to answer that I take it so the super elevation of the channel belt above the flood plain or the gradient the transverse gradient of the channel belt so what we got here is a plot of the transverse gradient after 600 floods for a series of alluvial ridges 20 different simulations and how the the ridge gradient varies as a function of certain parameters now what we see is that as you increase the suspended sediment load of the system we're moving from the green line up to the yellow line and you see an increase in the alluvial ridge gradient which is what you would expect more more suspended sediment more deposition a larger alluvial ridge as you increase the migration rate of the channel and that's moving from the red solid line to the red dash line what you see is that the the alluvial ridge is being reworked by channel migration and that's reducing the gradient the transverse gradient of the alluvial ridge the effect that's more interesting here is what happens when you change the rate of channel bed aggregation and there seem to be two things going on so for the highest channel bed aggregation rates what we're seeing is that as the channel upgrades more rapidly the alluvial ridge becomes steeper or its transverse gradient becomes steeper which is what you would expect but we're also seeing that at low low relative channel bed aggregation rates as that aggregation rate increases actually the alluvial ridge gradient is going down so there are there are two effects here one of them is that as the channel bed upgrades it effectively encourages aggregation of the channel belt but also as the channel belt as the channel bed upgrades it reduces the bankful discharge capacity of the channel which changes the way water and sediment are conveyed to the back of the flood plain in some situations particularly with low suspended sediment loads for the channel that effect actually dominates and we're seeing a reduction in the transverse ridge gradient okay now this is just one set of simulation but it seems to imply there's a much more complex behavior going on than we see currently in those sorts of simple parameterizations of long-term channel evolution alluvial ridge construction etc the last thing I want to mention in the context of those particular simulations this is a pattern of sediment deposition on the flood plain so you can see a pink line there and a blue line which represent the channel positions at two different points in time over the simulation so 230 and 280 flood events the black area in between is what's been reworked between those periods so I'm not looking at the sedimentation rates there the rest of the area is showing a sedimentation across the flood plain average per event and what we see is that sediment is concentrated as splays and the splays are located predominantly around the apex of meanderbends okay and we can look at several different simulations and over the course of simulations and we see similar behavior that sedimentation is dominated by these splays on the outside of meanderbends and there's a consistency there and when we look at the results from our coring on the rear benes okay so you don't to look at all the bars there if you just look at the blue bar showing meander migration rate and the red bar showing deposition rate what we tend to see is that deposition rates are highest on the apex of meanderbends that are migrating rapidly and what we assume is happening here is that rapid migration of the meanderbend is leading to destruction of the levee and creation of breach points which is where the flow is focused and where the splays are generated so there seems to be some sort of consistency here between model and filled situation and possibly we could use these sorts of models to investigate the relationships between the sediment load of the channel the migration rate of the channel and the frequency and the depth of these breach points which represent potential sites revulsion so we might be able to build simple parameterizations which could then go into our long term models of channel evolution or flood plain evolution sorry okay last thing I want to mention in that context we're now trying to apply the model on the Mekong and I've just got a landslide image showing you here about 50 kilometers of channel and you can see in the red boxes if I zoom in on those we've got a whole series of these lovely splays which seem to be the key to supplying water from the main channel to the flood plain so what we're now trying to do is change our scheme so instead of having a 1D network to represent the channel actually what we need is a 1D network to represent the channels on these splays and that's a much more complex proposition so just to summarize I've tried to look at things at two different spatial scales you know the I think 2D morphodynamic modeling has moved on to the point where we can simulate what look like really realistic channels but there's still huge uncertainty of the parameterizations that we need to address computational powers moved on to the point that we're now actually able to get hydrodynamics into some of our models of really long term flood plain construction and you know we can either do that to develop improved models albeit models that take a long time to run or we can use those models to try and build improved parameterizations that can still then be implemented in simple alluvial architecture models one thing I really haven't talked about is channel belt evulsion which to be honest is a complete nightmare and it's something that we could all do you know there's evulsion is a problem it's not just in rivers in flood in flood plains in fans in fan deltas in alluvial fans even gullying on hill slopes it's a community problem that I don't think we've really managed to tackle yeah so I'll leave it there thank you it's excellent talk I'm more interested in subsurface and what is what you're doing is preserved in a subsurface and have you looked or thought about inverse modeling and looking at geophysical data where I can show you geophysical data on my computer that shows these patterns in the subsurface is there a way maybe to use your parameterization do some inverse modeling and figure out what combination of parameters are explaining some maybe geophysical data yeah we're trying to do something like that we've we've taken some results from some of the sort of 2D high resolution models and tried to make comparisons between them and GPR data that have been collected on the Rio Paraná and we're also collecting on other large rivers and so it's definitely a direction we're trying to move in but it's really problematic because you know as you know you've got and hopefully maybe you can give me some ideas about how to deal with these patterns but we've got GPR data which is telling us one thing we've got a model which is telling us something else there are completely different spatial scales actually trying to decide what you can compare with what and what useful information you can extract from it is a challenge but you know something you might be able to help me with