 So we're going to begin with our next keynote speaker and we have Elwin Yeager who's going to be from the University of Idaho is going to be talking to us about predictions of bed load transport in vegetated channels, uncertainty, and steps forward. So I'd first like to thank the session conveners for inviting me to talk today. Also thanks for everyone to come to this talk in your post-lunch food coma. Hopefully this will be understandable. And also I'd like to acknowledge my co-authors. This is kind of a compilation of a number of studies that have tried to cram into one talk and Mark Schmeckli I collaborated with on the first half of this and then Ann Lightbody and Alex on the second half. So as we've learned a number of other talks during this conference, vegetation can affect blood plane deposition rates and so when we're modeling long-term landscape evolution of things like floodplains, consideration of vegetation is important. It also is used really widely in river restoration so we often add vegetation to increase bank stability or to decrease stream temperatures which provides aquatic habitat for a number of endangered and threatened species in aquatic ecosystems. But as Mariela pointed out in her excellent talk just before this, vegetation is not all good so particularly downstream of dams we can have invasive vegetation that takes over like tamarisk, salt cedar, and that can significantly alter channel morphology, sediment transport, and flow. Although there's been a lot of work on the influence of vegetation on flow and turbulence, there hasn't been a lot of research devoted to understanding how vegetation actually impacts bed load transport and so the fundamental mechanics of how vegetation interacts with flow turbulence to then interact with bed load transport hasn't really been investigated. Oftentimes we just assume okay we add vegetation and increases drag therefore it reduces flow velocity, re-reduce sediment transport rates, and have deposition. So that motivates our two basic questions that I'll try and at least partially answer today. The first is how does vegetation affect flow and sediment transport in river systems? And again I'm focusing on bed load not suspended load here and then is there a way we can accurately predict bed load transport through vegetation? So I'm going to focus on the first question first and that's going to be through some highly simplified flume experiments. So to answer that first question we conducted a set of experiments in which we simulated vegetation as rigid emergent arrays of regular cylinders and you're probably looking at this thing okay well this looks nothing like real vegetation in the real world, Elwin what are you showing me? And I know there are three reasons why we did this. The first is because so little is actually known about vegetation effects on bed load transport we wanted to start simple. So starting simple means just saying okay vegetation are rigid cylinders. Second this could be used to simulate essentially flow and sediment transport through trees or reeds. And finally because of the measurements I'm taking we couldn't actually add stems and pronating vegetation because it would interfere with our measurements of turbulence and bed load transport. So acknowledging this I'm asking you to stay with me at least for this part and then I'm going to add complexity in the second half of the talk and you'll actually see that this isn't such a bad simulation of some vegetation. So in these experiments we hold the mean flow velocity constant between runs so we're forcing that mean flow velocity to be constant by adjusting the channel slope and discharge and then we're transporting uniform 0.5 millimeter sand through these arrays and we scale this using Reynolds and Froude numbers to natural conditions. And in between experiments we vary the vegetation density by area so from an experiment with no vegetation so just a plain sand bed to four percent vegetation which is a pretty dense array. And for each of these different vegetation densities by area we had a range of flow velocities. Then we use particle imaging velocimetry to measure the detailed flow field so we did this to measure the downstream and vertical velocities. So what you see here is a laser that's illuminating neutrally buoyant particles that are in the flow field. And then we use a high speed video camera to record this at 250 frames a second. So this gives us essentially turbulence in space and through time. And coupled with this we also use high speed video again at 250 frames per second to measure the detailed sediment transport field around these cylinders. So our video camera is here. It's looking down onto the bed and this is a zoom in of what you would see with a video where these white circles are the cylinders. So we can take a look at. I'm not a Mac user so I don't actually know. It's not working. Oh it's playing. Oh right. It's not playing on here. Okay. So if you look at this video you can see individual sand greens moving in the intervening area between these simulated vegetation rods. You see relatively low transport but immediately adjacent to the rods you see higher transport particularly downstream of the vegetation. We have this recirculation zone and there's actually sediment being transported in the upstream direction. So we have high spatial variability around this simulated vegetation which is similar to what if you're an engineer you might see with bridge pier scour except of course we have many more cylinders than what you would have with a reasonable bridge. If you designed a bridge that looked like this you would be kicked out of school probably. So the reason why we're making this high speed video is to get really detailed bed low transport rates without using invasive measurements. So most of the time when we measure bed low transport we need to stick something in the stream and then we'll alter the transport field. In this case what we're doing is we're taking the video and we're differencing two successive frames and this is an example of the difference between video images where the white dots are where we have changes in the video and the black dots or black area is where there isn't any change. So then we add up all of those changes through time and that gives us total change in the image at each point. Now this doesn't actually give us a sediment transport rate just gives us changes in the images. So we have to go and calibrate this image change to visual measurements of sediment transport and we do this by staring at this video and creating imaginary transects through which the grains are passing. So like maybe right there I would stare at it and for the 1400 frames we have I would sit there and count each grain as it passes through that transect and then I do this at a number of different locations so that I can calibrate transport rate for a wide range of fluxes. And you're probably wondering why I'm not using like object recognition software to do this. It's because it's actually pretty difficult to get it to recognize individual sand grains. Sometimes they're moving together and in particular here you might have problems. So this can drive you a little bit crazy. I have to admit you wouldn't want to talk to me after doing this for a couple of days. So these images take about 30 seconds to measure but then it takes weeks to just sit there and count grains. But there is a payoff. I'm not just doing this because I like to torture myself. We get to have really detailed maps of sediment transport. So this is a contour map of the sediment transport rate around what we have our cylinder shown in brown and so this is the calibrated sediment transport from the image analysis where our high sediment transport is shown in red, low sediment transport is shown in blue. And what you'll know is there is about an order of magnitude or more variation in transport around these cylinders. And furthermore the transport that we're measuring corresponds to our visual observations. We have high transport immediately adjacent to the cylinders low transport and also downstream of them and low transport in the intervening area. So what does this mean for actual results and vegetation impacts on sediment transport? Well here we can compare two different sediment transport rates. So one on the left is a run without any vegetation and the one on the right is a vegetation density by area of 1.7% keeping in mind that we're holding the mean flow velocity constant between these experiments and they have the same scale of sediment transport right here. So if you're just visually comparing these and I have quantitative data to back this up I just didn't want to overwhelm you with plots. We see we have much higher spatial variation in sediment transport with the vegetation and also our mean sediment flux is much higher with the addition of vegetation than without and this corresponds to all of our different vegetation densities by area. As we increase our vegetation density by area we have higher sediment transport rate which is kind of contrary to what you would expect. You expect vegetation causes lower transport and deposition. So why is this occurring? This is happening in our experiments as has been shown in a number of other experiments because the vegetation is significantly altering the flow turbulence field. So these are two videos from the particle imaging velocimetry experiments where the white dots that you're seeing moving in the flow are the neutrally buoyant particles. So on the right we have an experiment with vegetation. On the left we have one without vegetation. Again these are for the same mean flow velocity. So what you'll know is you can actually see packets of fluid on the right that are interacting with the flow bed much more than you can on the left. So I mean these are both a fully developed turbulent flow but on the right we have much more turbulence occurring because of the vegetation and you can actually see individual sand grains that are moved when we have essentially turbulent bursts or sweeps interacting with the bed and these sand grains are being picked up into the flow column and then saltating downstream whereas on the left you don't actually see much sand movement at all. And that's because of the lower turbulence. And we can quantify this using the turbulence intensity. Here I'm not using normalized turbulence intensity so I'm just, it's RMS of turbulence or of fluctuations in turbulence, fluctuations in velocity. So here we have turbulence intensity plotted against the average downstream velocity and each different line represents a different vegetation density by area. So if we just look for example, average downstream velocity of about 25 centimeters per second we have a run with zero vegetation density and one with 1.7 percent we have a two-fold increase in the turbulence intensity. So that means we have much more turbulence, higher fluctuations and turbulence occurring when we add vegetation and this is causing higher bed load transport rates and this occurs pretty much anywhere you look as for the same mean flow velocity you increase turbulence intensities with the addition of vegetation. So at least in these experiments that we hold the mean velocity constant, we add vegetation, we increase turbulence intensities, this increases sediment transport rates. Now of course in nature what will actually occur, the mean flow velocity isn't going to necessarily be constant when you add vegetation so we've also done these experiments holding the mean shear stress constant and we see that the addition of vegetation doesn't really impact sediment transport rates very much at all. So the question is what parameter do you think is going to change, what flow, what is the actual flow within vegetation that's going to impact your turbulence and your bed load transport rates but the main take home that I want you to look at in these experimental results is that we're seeing that we significantly change turbulence that means we significantly change bed load transport rates and this has really large implications for trying to predict bed load flux in fields of vegetation. So that brings us to the second question, how can we accurately predict bed load transport through vegetation? And of course we've seen a number of great talks today where we have really complex numerical models where we solve the flow field and we can couple that with momentum balances on actual grains to predict sediment transport and flow through vegetation so my collaborator Mark has done some of this, this is a video of his using simulated vegetation and solving the flow field around them but of course this isn't necessarily practical if we really care about modeling this over really large scale reaches or long term landscape evolution. So question is can we take this understanding at the grain scale or almost the grain scale of sediment transport and broaden that to use in a sediment transport equation that might actually be useful at these larger scales or longer time scales. So let's just start really simple using one of the simplest possible sediment transport equations, sort of a straw man we can set up here. So this is the Luke and Van Beek equation you could use pretty much any other sediment transport equation I've tried a bunch and I get up about the same results that I was that I'm going to show you here. So we can predict sediment transport rate as a function of applied shear stress minus critical shear stress where we often use the total shear stress in this equation. So of course as many of you know total shear stress isn't going to account for the fact that a large portion of that stress is borne as drag on the vegetation and it's also not accounting for the effects of turbulence. But since this is what's commonly done let's just try and see if it works in vegetation. So I should note so we're taking the total shear stress measured in each of those experiments we predict the bed load transport rate in each experiment we compare it to what we've measured using that high speed video. So here I've plotted the predicted bed load flux on the y-axis versus the actual measured bed load flux on the x-axis one-to-one line between predicted and measured is shown here in the solid line getting to within an order magnitude of the measured values are shown by the dashed lines and so normally when you have a really large roughness getting to within order magnitude of measured values could be considered relatively good in bed load transport predictions which isn't the best thing to be shooting for but we're going to try to get within that order of magnitude. So here I have our 12 runs from our experiments we use the total shear stress for systematically over predicting sediment transport rate as we would expect because we're not accounting for the fact that a large portion of that total shear stress is being borne by the vegetation and isn't actually acting on our sand bed. So that's really nothing new. What can we do now to really help these predictions? Well one thing is we could try and use what's actually acting on the bed of sand. So we could use something called the near bed rental stress where we measure it five millimeters away from the bed which we obtained from those high speed video measurements and we could take the mean of that near bed rental stress that's actually acting on the sand bed. So this gives us the counts for the effect of drag essentially because we're only looking at the stress acting on the bed and it isn't really an actual measure of turbulence but it kind of approximates some sort of turbulence effects in it. So what if we use that near bed rental stress measured in each experiment? So that's shown in green when you note there's only one data point here and it's not because I've removed all the data points it's just because the actual near bed rental stress is less than the critical shear stress of our experiments so we were predicting zero flux for 11 out of 12 experiments. So now we're systematically under predicting sediment transport so this really hasn't improved our calculations. So why is this? Well if we get back to thinking about that large spatial variability in sediment transport that we observed in our experiments that has to be driven by large spatial variability in turbulence and stress or velocity and if we can look at the distribution of near bed rental stresses that we see for let's say one experiment this is the average we have a lot of rental stresses that are much greater than that average value and we're putting a mean value of rental stress in a nonlinear equation so we're not really representing what's actually going on in the bed by putting a mean value in that nonlinear equation to calculate a mean sediment flux. What we need to do is put in a distribution of rental stresses calculate sediment transport rate for each of those rental stresses and then calculate a mean. So what happens now if we do that? So the yellow are what we get if we use a distribution of rental stresses. Now we're getting to within that envelope of within an order of magnitude of the measured values so we're improving predictions because we are including that distribution but we're still only getting 70% of the predictions are within an order of magnitude so we're not doing the best job we possibly could and again my scale for success is still pretty low an order of magnitude that's not really what you know we want to go out to this community and say as we were just talking about discussions of uncertainty go out to the community and say yeah I can predict sediment transport with an order of magnitude I'm awesome I don't think I would get great reception on that one. So this brings us to can we accurately predict sediment flux in vegetation at least in these experiments if we use a spatial distribution of shear stress or possibly velocity we can improve predictions but even with that spatial distribution our bed load predictions still aren't all that accurate. So what if we try and apply this to more of a situation that's closer to the real world. So again we're going to focus on the first question first how does vegetation affect flow and sediment transport and we're going to do that in experiments in the outdoor stream lab at St. Anthony Falls which you've heard about earlier this week. In these experiments we hold sediment discharge and the flow discharge approximately constant between experiments and we run the loud to an equilibrium condition before each runs we have approximately the same bed topography and then we plant vegetation on a point bar just shown by that arrow and we plant the vegetation in different densities so we have zero percent at the top so just a sand bed we have two different vegetation densities by area shown in the bottom and we're planting juncus and kerricks which is sort of a clustered vegetation that pronates in the flow so it's much more it is natural vegetation and it simulates more natural conditions we might expect in a real channel. Then we measure the near bed flow velocities using an ADV so we get as close to the bed as possible in a number of locations on the bar and we use mini heli-smith sampler shown here to measure the bed load transport rate within the field of vegetation and then finally we measure bed topography so this is a map of that meander bend where each of these black lines are cross sections and then our point bar is here and all those different dots are where we planted vegetation in different runs and then the dashed box is where we did repeat sonar scans in each run so we ran the sonar scanner as often as possible and we did this to look at the migration rate of bed forms so we could identify bed forms in the sonar scans and then calculate how quickly they're migrating and get a sediment flux from the migration of those bed forms. So if we look at just some of the results from our experiments of course these are only three dots I'm not going to try and fit a line to them but we have zero vegetation density by area here and we have our two different runs with different vegetation frontal areas or densities. What we see is our bed form transport rate increases as we add vegetation so we have higher transport with the addition of vegetation and you might think okay this is similar to what I just showed you in those simplified laboratory flume experiments where vegetation is increasing turbulence so it's causing higher transport rates but that's not actually what's going on here. What's happening is that without vegetation or with more distributed vegetation a lot of the bed load transport in this flume is driven by the migration of sand dunes and those sand dunes are migrating over the entire point bar when we don't have much vegetation but when we add vegetation that's causing the dunes to no longer migrate over that bar and essentially they're bypassing the bar and moving over the side of the bar and in the thaw egg and so we have the same imposed upstream sediment supply and so since our dunes are migrating over a smaller area if we're assuming our channel is trying to be an equilibrium transport that means our dunes need to migrate faster to accommodate that imposed upstream sediment supply also we have flow acceleration around the bar that's likely causing the dunes to migrate faster as well so vegetation isn't just impacting flow and sediment transport in the patch it's also significantly altering the fundamental processes that are controlling transport and on the outside of the patch and if we look at the transport rate within the patch so these are bed load transport rates measured within the vegetation and these are from the heli-smith samples and I've plotted the bed form transport rates from the three different experiments as a reference all these dots and red are just from one experiment with vegetation and I've just chosen an arbitrary location and distance across the channel for those bed forms whereas these bed load transport rates they actually correspond to where they were measured in distance across the channel where this is the inner bank of the channel this is the outer bank of the channel so all of these are actually on the top of the point bar so we see a large spatial variability in the sediment transport rate within the vegetation and that transport rate is much less than what we would see with dunes so our bed form transport rate is much greater than our bed load transport rate through vegetation and this is because we no longer having those dunes migrate over the point bar and so we have much lower sediment supply through the vegetation that's causing lower transport rates and this has important implications for channel morphology so if we just look at one example of a run where we added vegetation what I'm showing you is the difference in bed topography with from before we added vegetation to after we added vegetation this is the outer bank on the top the inner bank on the bottom so essentially our bar is from the inner bank to approximately midway up in this diagram and the contours again correspond elevation changes so blue represents erosion and red represents deposition and what we see is that on the upstream end of the bar we have erosion even though we're planting vegetation so again this is contrary to what you expect you expect deposition within a vegetation patch rather than erosion and we are hypothesizing that this is due to the fact that we've halted the migration of those dunes over the bar and so we're now no longer supplying sediment to the top of this bar and so we're getting erosion and also as a number of people have shown at the edges of vegetation patches you can have increases in turbulence that could lead to erosion on the side of those patches smaller morphologic changes were also observed so here are our individual vegetation stems we see distinct scour holes immediately adjacent to them and these long strings of deposition downstream and this kind of roughly corresponds to what we saw in the lab where we had distinct scour holes immediately adjacent to these cylinders and sort of these stationary bed forms or deposition areas downstream so it's possible at least in these two experiments that we're altering the flow and sediment transport field in similar ways even though here we have emergent rigid vegetation and here we have pronating vegetation that's composed of a number of stems that fluctuates between being emergent and submerged okay so finally how can we accurately predict that the transport through vegetation such as the one that I've shown you in this OSL experiment so we just tried essentially the same sort of procedure we took all those mean all those near bed velocities that we measured and we tried to add a turbulence intensity to each of those near bed velocities that we measured and then we put them into a sediment transport equation in this case we're using different sediment transport equation that's based on velocity instead of shear stress just because we didn't have shear stress measurements so we use the acres and white equation so even when we use the distribution of those velocities plus some sort of measure of turbulence to predict sediment transport which we would expect would work in these experiments because it worked in our previous experiments we get a prediction of zero flux and I showed you we had measurable sediment flux in these experiments so we're still doing something wrong even if we have the correct flow field so even if we measure the flow field correctly or maybe we have a detailed numerical model where we can predict the flow field and we put those measurements into a bed load transport equation we're still not predicting bed load transport very well and part of this I think is because of the fundamental way we've devised these transport equations so what's missing a lot of these equations are developed for reach average and not local conditions of course they're not measured developed for vegetation conditions either but all these empirical coefficients and exponents and everything that we have in these equations are all developed saying okay I have a mean flow velocity in my river I have a mean shear stress and so we fit data using those values instead of actual local values that include the effects of turbulence as well so a lot of transport equations don't really include the effects of turbulence at least transport equations at this basic level and a number of recent studies are coming out to develop more probabilistic equations where we have essentially velocities of particles and transport distance of particles and it's possible that these more mechanistic equations might actually perform better than deterministic equations and finally and I'm completing right on time because I am a speed talker we have larger scale feedbacks between vegetation and bed form dynamics that aren't really considered so we it's not just okay we need to understand turbulence level mechanics I showed you that a presence of vegetation is fundamentally altering the migration rate of bed forms and this feedbacks to change channel morphology even if we look at the fundamental level of turbulence we might not be capturing these sort of reach scale dynamics that we're observing so there's a wide range of scales that we might need to consider to improve predictions of battle transport in steep channels but at least at the basic level including some sort of turbulence effects like we showed earlier in this talk could help so thanks a lot yeah so I mean I think these average flux laws are supposed to take into account the non linearities in the averaging but what it's showing you is that there's more description needed than just the shear velocity so that doesn't you know that categorizes the mean stresses and even with all the turbulent fluctuations but there's length scales and time scales and you showed this very nicely with your work so I think this is a big step forward but you can still perhaps end up with a mean equation like that it's probably gonna have to depend on it'll have to depend on more things than just a friction velocity yeah I think I mean what you really need to include is some sort of level of maybe the mean flow velocity plus some sort of turbulence intensity and the fluctuations in that turbulence intensity spatially or like length scales of sediment transport