 All right. Good morning, everyone, and welcome on Shuffling Landscapes. This is a webinar on the impact of landslides on topographical evolution. The webinar is part of a series on modeling topics hosted by CSDMS, which is the Community Surface Dynamics Modeling System. And hence, unsurprisingly, we will be talking about modeling landslides today. And I'm going to give a quick overview of what we're going to talk about today. So we will start with the question why we actually study landslides in the first place. Next, I'm going to tell a bit about the numerical Landscape Revolution model, the Highlands model. That will be followed by a small demonstration where I show some of the new components we developed in LandLab. I'm going to talk a bit on how you can run all of this yourself on your own computer or in the cloud. And finally, I will get back to my original question as to what extent such Landscape Evolution model can be used to understand better what landslides do in evolving landslides, landscapes. And then we will end with some questions. So if you have a question, please hold it for the end. And we will have time to discuss all of these. So let's start with the question, why do we study the impact of landslides? And to answer that question, I'm going to take you to what I call two end-member landscapes. On the left-hand side, you see a part of the Northern Appalachian Plateau in Ohio. And this landscape experiences very little to zero months of surface uplift, resulting in, quote unquote, old landscapes with very well-developed dendritic fluvial networks, moderate rolling topography, not very steep slopes. And as a result of that, you don't really see a lot of deep-seated betroth landslides in such areas. So with betroth landslides, I mean landslides that are rupturing through the betroth. What you do see in areas like Ohio is shallow landslide activity, where part of your soil is going to move over the interface between your soil and your soil and betroth at that interface. On the other hand of the spectrum, you have landscapes such as the Central Mountain Range in Taiwan that do experience very high amounts of surface uplift, resulting in steep topography, deeply incised river gorgeous, and hence a bunch of deep-seated landslides. And you can, for example, see one here in the middle of your screen. So this is an example of a relatively recent deep-seated betroth landslides. Now, we also plotted the drainage network for these two landscapes. And if you look at the drainage network and you zoom in on part of the network, what you see is for the Ohio case, which is the upper figure in the middle row, you see this indeed nicely developed dendritic fluvial network. Whereas if you look at the Taiwanese network, what we see is very steep quasi-planar slopes with narrowly spaced, almost linear channel features. So both of these subplots are made by taking a fixed threshold value for the drainage area. So by looking at these drainage networks, we clearly see that there's a difference between both. And one of the questions is how are landslides actually influencing the shape of these drainage networks? Another observation we can make for those two end-member cases is that if you look at the slope frequency distribution, where you just plot all the slope values or the frequency diagram, you basically see for both, but especially for the Taiwanese case, that the slope distribution shows a right-tailed distribution. So what you see is that there's many slope patches exceeding what is often referred to as a threshold slope or a theoretical angle of repose, which we often consider as a value where landscapes evolve if you have a lot of petroglyphalic landsliding events. But if you actually look to the measured topography, we see this nice tail. So both plots are the same, left and right. The right-hand plot is plotted in a semi-log space, so the y-axis in log values to accentuate all of these slopes that actually exceed what is often considered to be threshold slope values. So let's say that in highly tectonic active regions with strong lithology, we often consider a slope of 1.2 meters per meters as a threshold slope. But if you look to existing topography, you see that there's a bunch of slope patches that actually exceed those angles of repose, those theoretical value for the threshold slopes. So that's two observations from measured topography, the drainage network and the slope distribution that learn something about the role of landslides in landscapes. Now, there's another issue. So one of the questions or one of the current debates really is what is the status or the shape of steady-state landscapes? And there is debate whether landscapes, if they evolve towards equilibrium, evolve towards a fixed or frozen equilibrium, where the morphology of landscapes remains kind of fixed through time and where erosion would be uniform over the entire landscape. So that's one end of the spectrum, whereas the other theory says landscapes evolve towards a highly dynamic equilibrium where riches and rivers are constantly on the move and where topography can be inverted in such a way that existing riches become fluvial valleys and vice versa. So like in the two pictures you see, the left hand is actually a sandbox experiment where they show that landscapes actually never evolve towards this frozen equilibrium states, but rather evolve towards a dynamic equilibrium where you see all this lateral channel mobility, lateral rich mobility going on. And the right hand figure is a theoretical study that recently came up showing that lateral fluvial channel dynamics indeed force a landscape to go towards this dynamic equilibrium where neither riches nor valleys are fixed through time but keep on moving constantly. And our question is, can landslides actually explain such a persistent dynamism in the landscape and can these landslides be used as one of the processes to explain such mobility? So that's the science questions we want to answer and that's actually why we are developing these landscape evolution models that do simulate landslide activity. So like a brief recap, we have the differences in the shape of the drainage networks. We have the right skewed slope frequency plots and we have landscape dynamism, which we want to explain with landscape evolution models explicitly simulating these landslides. So that's a scientific background of studying landslides in landscape evolution models. And this is why we developed highlands and highlands is a hybrid landscape evolution model and it's hybrid in the sense that every pixel can turn from like a fluvial domain into a hill slope domain. And with a fluvial domain, I mean both incision and sediment transport and with a hill slope domain in this case, and typically this would be landslide erosion. So in the highlands model, we have two components. We have a fluvial component and we have a landslide component. For the fluvial component, we have both incision as well as deposition and we simulate the fluvial dynamics using the space model that Charlie Showby developed a couple of years ago. And in the space model, we can dynamically transition from detachment limited states to transport limited states. So and if you want to know more about the space model, I invite you to go check Charlie's paper in geo-scientific model development in 2017. So that's for the fluvial components. A second component we have in our highlands model are obviously the landslides. And to simulate landslides, we assume a more cooler stability analysis or we do more cooler stability analysis to calculate what the rupture plane of your landslide would be. So basically what it means is we're going to identify critical nodes in the landscapes, the blue dots. And from this critical nodes, we're going to calculate the rupture plane of your landslides. And the rupture plane is actually calculated as the angle bisecting the angle of internal friction of your materials, which is the lower black line over here and the actual topography of your current relief. And the rupture plane is calculated as the straight line bisecting those two angles. And it's going to start at the critical node and it's going to propagate all the way until the daylight right here. And all the pixels that are exceeding or the topography of pixels that exceeding the rupture plane are going to erode and they're going to erode up to the elevation of the rupture plane, the red material. And then it's going to be deposited downstream of this critical nodes. For the deposition of your sediments, we use a non-local diffusion algorithm developed by Sebastien Caratier in 2016. And it's going to help in like spreading out or calculating the landslide runout, just the runout of your sediments. And we use it to calculate the deposition rate of sediments in space. One more thing that's interesting to note is that the way we calculate the location of these critical nodes or the location where the landslides actually start forming by using a stochastic algorithm. And the stochastic algorithm has two components. It has a spatial component. The spatial component is going to be based on the gradient of your topography. So the steeper the higher the chance of sliding basically. And it has a temporal component in which we sample randomly from a Poisson distribution through time. And we combine those two probabilities to calculate an actual probability for failure. And then we use a stochastic sampler to define whether or not we will trigger or initiate a landslide. So of course in reality, the triggering of landslides is often associated to earthquakes or storms or human activity in some cases. So that would represent our stochastic triggering of landslides. So that's how the landslide algorithm basically works. If you would draw that in three dimensions, it looks like this where we again have this critical note which is the blue square from where landslide planes, rupture planes start propagating upstream. The red dots are the cells where we are gonna see erosion. And then the eroded volumes is gonna be spread out following the green arrows here in the bottom. And to spread out the sediment, we use a multiple flow direction algorithm so that you can basically go from one cell to its eight neighbors. So that's how the landslide algorithm works. And we published the landslide algorithm earlier, no, in 2020, summer 2020 in a GMD paper called Highlands, a hybrid landscape evolution model, et cetera. So if you wanna find out more about the algorithms and the implementation of the method, you can find it in this paper. Now, what you will find interesting is if you go and check this paper, you will see that we implemented our first version of Highlands in MATLAB, in the Topo Toolbox Landscape Evolution Model, which is originally developed by Wolfgang Schwanhardt. And over the last year and a half, basically, we moved from, it's still there, you can still use it, if you like in TTLAM, but we re-implemented our algorithms in Landslap so that we can use it in our open source software packages. So what is Landslap? Most of you on the call probably know, but Landslap is a Python engine or a package of numerical modeling tools to simulate earth surface dynamics. And it basically does several things. First thing it does is it has a grid engine and we use a grid engine to basically represent our spatial domain and we use the grids to calculate our numerical equations or to calculate our earth surface processes. Another very important ingredient of Landslap is the model components. So we have a bunch of components to calculate all kinds of earth surface processes. And then there's some utilities we use for plotting, for example. So there's a bunch of functions to do plotting. And so we recently developed this Landslap Highlands components and basically the Landslap Highlands, you have to see Landslap Highlands as a Landscape Evolution Model. So Highlands is a model and it exists out of several new components we developed. So basically in over the last couple of months, we developed three new components to create this Highlands numerical Landscape Evolution Model. A first component is the Priority Flood Flow Director. The Priority Flood Flow Director is a new flow rather we developed. And the reason for that is the following. So if you have a bunch of Landslites going on in the Landscape, you can imagine that all these Landslites generate a lot of sediments and the sediment is gonna block rivers and is going to create what you typically know as a Landslide Dam and is gonna invert your topography. So basically what you do is you build a lot of lakes in your Landscape and this is gonna complicate water flow calculations because if you have a lake and invert topography, the water doesn't necessarily know where to go. So what we typically do is we bridge the DEM or we fill the DEM to calculate the new location of your streams. And this is a pretty expensive operation. And this is why we developed this new Flow Director which is basically wrapping an existing Python package called Richdom. Richdom is a very nice Python tool developed by Richard Barnes which is especially designed to calculate flow routing and filling and bridging over large DEMs. And it's efficient because it uses, it parallelized the process of calculating flow directions. So that's the first component we developed. The second component we developed is Space Large Scale Eroder. Space Large Scale Eroder is basically derived from the original space code as Charlie Showby developed but it's designed so that it can be used on larger scales. It's a bit more of a boost against bigger time steps and hence we can increase the performance of our Large Scale Landscape Evolution models a lot by using this. And then the final component we developed is the Betrok Landslider which is obviously which contains obviously the Landslide Algorithm I just briefly explained. And then finally, I also added one new utility function which is called Imshow HS Grids and HS stands for Hill Shades. So this plotting function is gonna allow you to plot shaded relief and to add additional information on your shaded relief and we will do an example of that in the tutorials. So now I'm gonna risk something and I'm gonna do a small mini demonstration. So let's see if I can show how the Highlands Algorithm works. So I'm gonna exit my presentation and I'm gonna move to my browser. Is everyone seeing this? Yeah. Thank you, Mark. You're welcome. So this is a small tutorial I made especially for the Highlands model and it's gonna be available to you if you want so you can try to play with all these little notebooks. You'll see that we start with a short introduction on LandLab. So there's a bunch of information on the, it's basically the read the docs of LandLab. So there's basically information on how to install LandLab, how to run simple components. I also added a link to the tutorials where you can find a bunch of notebooks on how to run different components in LandLab. And then the first thing we're gonna do is we're gonna try to simulate some better of landslides on the landscape. So let's click the link, fingers crossed. There we go. So there's a lot of documentation in these notebooks and I'm not necessarily going over all of that today. I wanna do a quick demonstration just for you to get the ID and because all of this is gonna be available for you to play with and just take your time to go through it after this. So we typically start in Python by importing a bunch of packages we will use for our calculations. So let's look what we have here. So we have NumPy, NumPy is often used to do all kinds of areometric operations in Python. We have the matplotlib libraries which we use for plotting and then all of this is gonna be LandLab stuff. So what you see here is actually importing the components we're gonna use and the new ones are for example, the priority flood flow router, the better of landslider, channel profiler is another prof component we use for plotting that already existed before. Other stuff we do is read NATCDF we're gonna read in an existing DEM so we need to read the data. What else? The roster model grid is for example the component you use to create a grid in LandLab. Right. So a couple of months ago, Mark Publer, who you just heard developed a very nice package. It's called the BMI topography package and it's actually made to download SRTM data just by providing for coordinates. And it's based on the open topography API so we will actually download the SRTM data through their API. And to do this, it's a very simple task. So you're gonna initialize a topography components by using the topography function. I say which kind of DEM I want to download. So I said SRTM one and I give four coordinates listed over here and I say that I want to have the output as an ASCII file. And the DEM data is gonna define where I store my actual data. So when I run this cell, it's gonna initialize this topography components, say. Now, by initializing it, we're not yet downloading it. So we have to actually fetch the data. So if this sounds complicated, don't worry about it. So like all the explanation is here and also I added a link for more documentation on how this component works. And once we downloaded the data, you can also load the DEM in the data. It's gonna print some of the properties. But obviously what we want to do is we want to visualize the DEM. And to do that, we're gonna read it. So now we downloaded the data. Now we're gonna read this data into LandLab. And the way we do that is by using the read as we ASCII function, which is a LandLab function. F name is gonna refer to the name we gave to the data we downloaded. So you're gonna find F name over here. Let's see, F name. So when we fetch the data, F name is gonna be one of the output arguments. And it's gonna consist of the coordinates of your DEM, basically, and the location where you store the data. So we're gonna read the downloaded DEM into a LandLab by using the read as three function. And I'm gonna plot the function. I'm gonna plot the data. So the data you see here is this nice DEM. And you might wonder where this is. And actually this is gonna be Boulder. So if you look to the DEM, you basically see if you're familiar with the Boulder area, you basically see Green Mountain over here. This is gonna be Bear Peak and this is gonna be South Boulder Peak and then you see the Flatirons in front. All right. So as a nice intermezzo, I just want to mention that obviously I present from somewhere here in Boulder and we are, I'm presenting on the former grounds of the Arapaho, the Cheyenne and the UT tribes. And I was wondering as a small break because I'm thirsty, where you all are from. So you can feel free to enter in the chat from which place you are zooming in today. I'm curious to see that from Bellingham, more Boulder people, people from Europe. That's nice. Good evening, across the stadium, more people from Europe, Austria, from Brazil, from Iran, people all over the world. I love it from Columbia. Cool, Michigan. Excellent. So we have our data folder for the Boulder area. We have our flat iron sitting there. There's only one problem. So if you look to the plots, you see that we actually downloaded the data in a geographical reference system. So that means in decimal degrees. And obviously when we want to calculate landslides, we want to have it in metric units so that we can easily calculate slopes, et cetera. So we have to reproject our data. And now we're going to do like a small little trick and we're going to assume that our data is actually spaced by 30 meters, which is more or less what it is for SRTM-1. So we're going to make a new land lab grid. So I'm going to make a new grid by using this roster model grid function. And then once I have actually initialized the grid, I'm going to ascribe the elevation data to that new grid and assume that the spacing of your grid cells is indeed these 30 meters. In reality, when you want to do this, you actually have to use a better projection function, something like GDAL to do it more correctly. But for now, we're going to use that trick. So we're going to make a new grid, assign the values, and then plot the data. So one thing that's interesting to note here is that here I'm using this new utility. So it's called ImShowHS from HillShades. And it works like the input arguments are pretty similar to what you used for ImShow, the original land lab function. But now this time, if you plot it, you're going to create this nice shaded relief below your actual topography. And you see how the flat irons come out nicely and how these different peaks I mentioned before are nicely visible in the picture. All right, so the next cell is basically a cell that I wrote to do all of this plotting. You can look at it if you want. It's a function we are going to call to plot stuff. For example, if I call the plotting function, which is, so I create a function here, it's called plotting with a bunch of input arguments. If I call the plotting function, you're just gonna plot the same topographical elevation. DA is false, equals false means that I don't want to plot the flow accumulation. The reason why I'm not plotting the flow accumulation is because we haven't calculated it just yet. So to calculate flow accumulation, we're gonna use this new priority flood flow router, which is going to take a bunch of input arguments. And something that I always find very helpful when you're working in a Jupyter notebook is to look at the function you're actually calling. So you can do that by going behind the function and then hit shift tab. And then if you click the plus sign here, you see all the input arguments with the default values. And you also get access to the documentation of the function. So if you scroll down, you're gonna see some text explaining what the function does. And you're gonna see some additional information on all the input arguments and output arguments. All right, so we're gonna run this. Something that's important to know is that in this new flow director function, you can calculate simultaneously a single flow direction map and a multiple flow direction map. And the reason why we developed it like that is because for some components such as highlands, you want to have both single flow and multiple flow. So if you're making a model, such as highlands, you want to calculate fluvial incision using the single flow routers. And you want the multiple flow to spread this landslide derived sediment when you're running your bedrock landslider. And by combining it in one flow router, you can again win some performance because you don't have to calculate the flow routing over the filled areas over the lakes twice while running in one time step. So you can do it the same time, the multiple and the single flow routing. So to do that, I actually have to say, calculate a separate hill flow. So separate flow directions over the hill slopes. I said it to true. And also I want to accumulate our flow over the hill slopes. So if we run this component, you're gonna see that we, this is like the topography as we probably did before. And now we do plot our flow accumulation. And the first plot is the steepest descent algorithm routing over the neighboring cells. And the second flow accumulation is a multiple flow accumulation plot where water can go in several directions in the meantime. All right. So now we have our DEM. We protected the DEM and we calculated the flow directors. The next thing we're gonna do is calculate the landslides. So before we do that, we have to add a fields to our land lab grids. So I'm using a lot of terminology. If land lab is entirely new to you and you don't know what fields are or what grids are, in the interest of time, I cannot talk a lot about it today, but we do have like other webinars where we go back to the basics of land lab where you can just go into the documentation of land lab and try to read up on all of these things. So but in land lab, we create a grids and we assign a bunch of fields to the grids. Fields contain values for the elevation, but also for other properties you want to store on the grid. And with landslides, we don't just want to store the elevation, we also want to know how much soil or sediment there is over the grid because we calculate erosion and the position of the landslides. So what we're gonna do is we're gonna add a field called soil depth, which is going to represent sediment thickness. And once you've done that, you can actually initialize, you can initialize an object of the Betrok landslider class or component. That's again, some terminology, don't worry about it, but you have to initialize the landslider first before you can actually run it. So that's what we're doing in these lines. There's a couple of input arguments. So for example, we give the angle of internal friction, which is a material strength property, and we set it to a pretty low value in this case. So open five meters per meters is actually too low for the kind of materials we have in the boulder area. But I set it that low deliberately to show you what potential landslides could do in this area. So all of this is purely hypothetical. And for real situations, you wanna bump up this value of internal friction. Another parameter we need is the return time for landslides. So when I was explaining the algorithm, I said that we have a stochastic time sampler and the amount of landslides we sampled through time is gonna depend on the return time for landslides. Now, think of this parameter as the return time for example, big storms or big earthquakes. And the longer it takes for a big earthquake to occur, the higher the value for this parameter will be. And then finally, we have a cohesion value that's some other material property that's gonna be needed to run the sliding algorithm. And once we initialize the component, we will also run it. So every landslap component has a run one step function that's actually going to execute a sample over a given amount of time. And in our case, we say run it for 15 years. All right, this landslide function, once you run it, it's also going to return something. Those are two values. It's the amount of sediment that is mobilized and the amount of sediment that's being evacuated out of your grid. So you don't have to worry about it for now. So let's see if we can visualize all of this. And let's start by plotting the magnitude frequency distributions of your landslides. And this is pretty interesting. It's, if you look at landslide literature, the magnitude frequency distribution is a property of landslides that's very often plotted. It's gonna give the area of the landslides and then the amount of landslides you have for a given area. And typically in the field, you always see like a rollover where you start with a little amount of small landslides, then you increase and then the amount of bigger landslides is going to decrease again. And we can quite successfully reconstruct this magnitude frequency distributions by running the highlands algorithm. And now this is like a simulation with relatively unrealistic values for, for example, the angle of internal friction. But if you would look in this GMD paper I mentioned earlier, you can see how we run the highlands algorithm for a region in Nepal and Namshebara region where we also reconstruct this magnitude frequency distributions. And then finally, let's see where we have actually erosion and deposition in our landscape. So the first spot is showing the erosion rates. So there's a bunch of erosion going on at where the mountains are going up. And then obviously you also have this deposition of sediment. And to plot this a little bit nicer, you can also plot it on top of this shaded relief where you see how the landslides typically occur on the steeper parts of your landscape. And then the deposition is going to be downslope of the actual landslides. And then one final trick you can do. So there's a bunch of code. And this is all about this new like Imshow hillshade grid function that's going to help to plot you various values on top of shaded relief. So this last thing is actually going to plot your topography. And then on top of that, it plots both the landslide erosion as well as the deposition. So you get to see this nice landslide looking shapes and then the runout of your sediments in blue, showing where the sediments will be deposited after a landslide events. So that is a very quick introduction to the bedrock landslider basically. Now, obviously you want to combine this bedrock landslider with all the functions we have in a land lab or other components we have in land lab. So I'm going to go back to the index. So what you see is there's two more tutorials that are basically on the use of this new priority flood algorithm and that are going to show how much more performance this algorithm is in comparison to the older version. So I'm not going to do all of them but I can quickly show how it's going to operate on real topography. So what we do in this tutorial is we download an SRTM DEM as we did before and then we will run the space component or the new space large-scale eroder component on this downloaded SRTM DEM time, how long it takes to execute and do that for the existing functions and for the new priority flood flow director. So if we execute this entire notebook, I'm not going to go through it cell by cell but what happens is we download the data, we plot the data, we plot it in three dimensions and then here in this function we are running the existing space component together with the existing flow routers we have in land lab or the first generation flow routers we have in land lab and then in a second instance or before we go to the second instance we plot erosion. So from the start of your, from the starting DEM to the DEM you get after running the space fluid incision and deposition algorithm for a bunch of time steps. So we see it's pretty steep topography so you dominantly have erosion in this case. And then in the second part of the notebook we do exactly the same but this time we use the priority flood flow router and the space large-scale eroder components to calculate exactly the same processes basically erosion and deposition using the space component. And again, you see a map with the erosion in space. And then if you compare the performance of both algorithms you see that we gain a lot by using this priority flood flow accommodator combined with the new space components. So this is interesting especially when you're simulating larger domains if you're interested in simulating a large domain I would totally recommend to use these components. So we did a small performance test and you see how it's increasing. So the blue line is with the default flow routers and the orange line is with the priority flood flow routers. You can gain like several orders of magnitude of time especially when using larger grids. And the reason why is because the priority flood flow router is based on the rich them package. And as I mentioned earlier the rich them package uses parallelization which is especially interesting here in this case because large DEMs otherwise just fill up your memory and slow down everything by a big amount of time. All right, so going... This is the danger of doing stuff live, I guess. I can just fill a little bit of time until Benjamin comes back. Benjamin worked for a long time on this and actually it's kind of neat the pull requests for to put all of this software back into LandLab took a long time in order to make it all really good. And so there's lots of tests, lots of performance checks. So he spent a lot of time on this and it's really cool to see his work get there. One other thing I can talk about maybe just as a quick... Oh, I think there's a bench, but okay. All right, I'll shut up. I'm back. Sorry about that, my internet dropped. Can you hear me fine right now? Yeah. So I'm gonna go back to this tutorial I was showing. All right. So this run our model to a steady state and I was saying that we use classical theory here. So we evolve towards a fixed landscape through time. So basically you run your landscape for a number X amount of time, you run it to a steady state and then on this steady state topography we're gonna insert landslides. Something interesting to look at is like the shape of the landscape before we insert landslides. So we see a kind of fixed topography and more or less uniform sort of thickness. And then on this topography we are going to add landslide activity. So we do that by initializing the petrog landslider component and by running it in a for loop. So this is gonna happen over here where we say update the flow rather, run the space mall to calculate erosion and deposition and then run the high lands or better the petrog landslider component for 200 years. So we run 10 time steps of 20 years. So 200 years in total. And then in the end to visualize the output the magnitude frequency distribution of your landslides. You plot of your landslides you can plot erosion and deposition as we did before. And if you combine that in a figure as we did before you see that you have your relief with the red cells indicating erosion and the blue cells indicating deposition. So that's about what I wanted to show you and now I'm gonna go back to my PowerPoint slides. Oops, there we are. So one question you might have is how can we run all of this on our own computer? Marc, do you see the slides now? No, slides are not. They are not? No. Sorry. No worries. That should be better. Yeah. Perfect. So I'm gonna share this slide deck so that you can just go there and click the different links. Basically you have two options. You install the LandLab software on your own machine or you use our CSDMS Jupyter Hub and to install or to, if you wanna use a CSDMS Jupyter Hub Marc is gonna say a couple of words about that. Thanks, Benjamin. Yeah, so the Jupyter Hub is a really cool thing. It's an always-on-computer resource. So not only can you run notebooks like Benjamin's, for example, but it's really a Linux machine on the cloud that you can use. Anyone who's a CSDMS member can get an account and just go through the link that Benjamin has provided. It's kind of neat. Several professors around the country have already used to teach classes, which is really cool. We can help you get it set up. It has all of our software as well, so that makes it really easy to use. So just in general, I think it's a really cool community computing resource. Right, back to you, Benjamin. Oh, no. Maybe not back to Benjamin. Right, back to you, Benjamin, if you're there. I am. Okay, okay, awesome. Take your internet connection. That doesn't usually happen to me and of course it doesn't now. So let's see, we're gonna go back to this presentation. Oops. Okay. So that's a Jupyter Hub. If you wanna use, there's like one more thing I wanted to say. That's like, on this link, you can actually find the GitHub repository for the stuff I've been showing today. You can just navigate to that GitHub and you're gonna find a direct link to actually load all of these notebooks in your own CSDMS Jupyter Hub space. So you can just go there and click that link to try it all yourself. So I just want to wrap up now, basically, and I want to quickly come back to this question we asked in the beginning. How can we use all these numerical tools to actually learn something about the impact of landslides or landscape evolution? And to answer those questions, what we did is simulating landscape evolution under different uplift gradients. So basically, if you have very low amounts of uplift, you're going to evolve towards a situation where relief or topography is not steep enough to actually see landslide activity. So that's what you see in the left column of this figure. So you've evolved towards a fluvially dominated landscape which is going to evolve towards a fully fixed steady state with uniform erosion and the second part is showing salt thickness. So uniform erosion, uniform salt thickness and no landslides. And by bumping up uplift rates, you basically trigger a bunch of landslides because you increase topographical relief. And I did a zoom of these figures in the next slide where you see that the left hand picture is the one with only fluvial processes going on the right hand figure is the one with a bunch of landslides going on. And interestingly, what we see is very similar to what we saw in the real landscapes I talked about in the very beginning of the webinar where you see quasi planar slopes with narrowly spaced channel features. So by doing this landslide erosion, we actually generated topography that is very similar to what we see in landslide dominated terrain. And there's a bunch of older topographic metrics we can compare and we see similar stuff. But one more thing I want to come back to is this slope frequency distribution. And very much like what we saw in our real topography we also simulate slope frequency distributions where a bunch of slope patches are exceeding this theoretical angle of repose. So in this model runs, we actually use a theoretical angle of internal friction of 1.2 meters per meters. And you see that there's a bunch of cells or pixels that are actually exceeding this theoretical value. And the reason for that is because we use a stochastic algorithm. So we allow slopes to over steepen beyond their theoretical angle of repose. So that's like regarding the questions of the morphology of your landscapes. Now we had this older question about landscape dynamism. So to understand what's going on there, I have this little movie where you see how fluvial profiles evolve through time. And in the left hand plot, you see a fluvial channel evolving under low uplift rate. So it's gonna be dominated by fluvial processes. And indeed it evolves to a steady state and it doesn't really move any. It doesn't really move a lot. Whereas on the right hand side, you have higher uplift rates, which is creating steeper relief, triggering these landslides and creating a bunch of irregularities over your fluvial channels. So what you see is, if you would pause the video like this, you're gonna see on the profile, you're gonna see blue spots which are basically landslide land lakes with orange bumps, which is a sediment deposited by rivers. You're also gonna see gray bumps, which represents the formation of epigenetic gorgeous, where your water is basically rerouted all of your former valley into what used to be the valley flanks. And all of this dynamics, all of this irregular dynamics are going to result in landscapes that are not really fixed in time anymore. And this is like my final slide of the day, where you see again these two end member landscapes. And on the left-hand side, you see a landscape that's evolving under purely fluvial conditions. So it's going to a fixed steady state where your valleys and ridges remain fixed through time. And in the right-hand side, what you see is the impact of landslides on the stability of your landscape in a dynamic equilibrium. And contrary to the left-hand landscape, where you evolve towards fixed steady states with landslides, you're gonna have ridges that are laterally mobile through time and they keep on being mobile. So this simulation is after 250 million years. So that's a very long time. It's longer than it took for African South America to get apart. So I can replay the movie. So you will see some dynamism in the left-hand figure in the very beginning, but then towards the end, you only see dynamism of the valleys and ridges in the simulations where you have landslides. So it seems that landslides are indeed a process that do explain this lateral mobility that we observe in both experimental setups as well as in theoretical derivations. So that's all I wanted to tell you. Thanks a lot for being here today. I'm really happy that there's so much interest in simulating landslides. And obviously I also want to thank a bunch of people that worked on this project over the last couple of years. So if you have questions, feel free to unmute yourself or you can also put them in the chat window. I have a question about the stochastic driver. Oh, sorry, when I jumped in before you said to please ask. No, no, go ahead. Sure. I guess I'm just wondering how much, like this is driven by a Poisson distribution, how much all of this work would change with a different distribution and whether that's sort of like a major driver that I don't know if it's an assumption we're making. Right, so for the temporal, so there's a spatial probability and there's a temporal probability and the temporal probability, we sample from a Poisson distribution, you can sample from another distribution. It's not really going to change a lot. What is mainly controlling the temporal probability is this return time for landslides or the return time for a shaking event if you want. So yeah, you can obviously use other distributions to sample from that would be a nice exercise, but it's permanently the return time for landslides that's going to influence the amount of events you see. Cool, thanks. Yes, so there's a question on how the landslide material is distributed downslope of the critical node basically. A question from Alison, thank you Alison. So we use a non-linear diffusion algorithm to do that. It's an algorithm published by Sybastien Karatier in 2016 and basically it's going to depend on a transport length scale. You're going to derive from the steepness of your topography. So the steeper the topography, the higher the transport length scale and the further your sediment will be transported downslope. So you can actually end up with, imagine you have a landslide in very steep terrain. Most of your sediments that's being distributed downslope of your critical node is going to go pretty far and almost all the way to your valley bottom. Whereas if you have like a gentle topography with a landslide somewhere in the middle, the sediments is the transport length scale is going to be shorter because you have like gentle topography and most of the material is going to be deposited pretty close to the critical node of your landslides. Does sediments completely evacuate the slide plane? Yes, it does. And that's one of the things we might be looking in for a future versions of the Betrok landslider component. So one of the ideas might be to have a landslide where part of the material is actually still sitting in the landslide scar so that it's not like completely evacuated downslope but like only partly. So that's a current assumption we make that all of the sediment is being evacuated. All right, there's a question from Miguel. In terms of runout, is it fair to think that the nonlinear mobile captures some characteristics of the runout in debris flows or other similar flows? So we've been thinking about debris flows because what we see in our landscapes, if we simulate landslides, the last picture I showed is basically that because we have this intermittent behavior of sliding erosion, sliding erosion, we are kind of mimicking a process that's like often associated with debris flows in the sense that it's gonna be a combination of gravity and fluvial incision that actually creates this linear closely spaced channels. So they are definitely related but that's something we have to figure out to what extent this mimics debris flow. Yes, can you hear me, Dr. Kanfortz? Yeah, thank you, Yvonne. Okay, thank you. And I just want to ask you if you have any thoughts about how these mechanism of like migrating divides translating to the submarine realm, that's something I'm wondering about a lot and I would like to just have your thoughts about it. Yeah, I didn't really get Artu translating into which I think. Oh, the submarine, the submarine realm. Oh, okay. Interesting question. That's a difficult one to answer, especially because in the submarine world, sediment transport is gonna be or might be pretty different and dependent on your marine flows as well. So basically a lot of the mobility in terms of like riches and valleys we see in our landscapes is because the sediment that's derived from these landslides is kind of pushing your fluvial channels out of their former valleys. So and I don't really know how that process would translate in a submarine world but that's a good thing to think about. Yeah. Okay, thank you. Okay, I think it's my turn. Hi, Sergio. Hi. Yeah, Benjamin, thanks so much for this lecture. I learned about this before but today I learned even more. So my question is the last few slides you showed comparison of identical landscape under two different conditions, one with uplift, one with no uplift. But I think another utility of this model is to understand and forecast what the impact of climate changes. And I think it has something to do with the stochasticity component of the highland if we can make some inference of how the landscape will evolve as we enter a different climate regime. I mean, is that something that we could do with this model? Absolutely, that's definitely one of the things we will be looking into in the future. So with climate change, we're gonna see increased storm intensity. So not necessarily more rain but a lot of like big storms. And of course these storms will increase your poor water pressure and thus trigger more landslides probably. That's the assumption we have for now. But there's a couple of unknowns in the equation. So the landslides we were talking about today are deep-seated betrothed landslides. And there's actually a recent work by Alison Duval that's showing that precipitation might be driving deep-seated landslides to a bigger extent than actually earthquakes do. So this is one of the things we will be looking into to what extent actually earthquake versus precipitation triggers are influencing landslide extent. And to do that, we have to think how to bring in this triggers explicitly in the algorithm. Because as you know, like now we have this stochastic algorithm that's like causing the triggering of an event. And if we move to earthquakes or storms, we have to bring that trigger explicitly into the model. But that's like for sure one thing to look into. And then a second component is, so we do mostly deep-seated landslides, but it's well known that rainfall triggers landslides are often shallow landslides as well because you're gonna see an increasing pour-water pressure. And especially on this interface between soils and betrocks, you're gonna see more sliding. And currently, we don't have algorithms to do this shallow landslide. So we have the gravitational sliding, but we don't have shallow landslides. So that's something else we should extend into if you wanna understand, if you want to fully understand what the impact of storms would be on sediment mobilization. And related to that actually, there's a nice paper on the role of sediment thickness in controlling shallow landslides. And like one of the current debates is whether it's sediment thickness or sediment availability, that's controlling shallow landslide activity rather than precipitation. So that's like a question tangential to the precipitation debates. Do we have any further questions? I don't think so. I think that's it for now, Benjamin. Excellent. So thank you all for being here today. That's really appreciated and see you all soon.