Remote Sensing For Landslide Science And Disaster Planning

Speaker 1: Welcome to the RSET training, Landslide Monitoring and Risk Assessment Using NASA Earth System Data. We're delighted that so many of you are joining us today. My name is Sean McCartney. I'm an RSET trainer based at NASA's Goddard Space Flight Center in Greenbelt, Maryland. It's a pleasure to welcome you all to the first part of this three-part webinar series. For those that are new to RSET trainings, the following slides provide a brief overview of the programs. RSET is an acronym which stands for the Applied Remote Sensing Training Program.

Speaker 1: RSET is part of NASA's Earth Action Capacity Building Program. RSET provides accessible, relevant, and cost-free training on remote sensing satellites, sensors, methods, tools, and applications. The goal of our training is to increase the use of earth science, remote sensing, and model data in decision making through training for professionals in the public and private sectors, environmental managers, and policy makers. On the right of the slide, you can see all the different thematic areas in which we conduct trainings.

Speaker 1: Trainings are offered online and in person and are targeted to beginners and advanced practitioners alike. Trainings are freely available and conducted either live, instructor-led, or self-guided, such as our newly released Fundamentals of Remote Sensing on RSAT's Learning Management System. Since 2009, the program has reached over 145,000 participants from 189 countries. RSAT trainings are cost-free with materials offered in bilingual languages and are freely available to use and adapt for your curriculum.

Speaker 1: If you use the methods and data presented in RSAT trainings, please acknowledge the NASA Applied Remote Sensing Training Program. If you want to learn more, please visit our website. The following slides provide an overview of the three-part webinar series, Landslide Monitoring and Risk Assessment Using NASA Earth System Data.

Speaker 1: By the end of this three-part training, participants will be able to identify the core concepts of landslide risk mapping, including geophysical and meteorological drivers, and how satellite data can be used for this purpose. Select appropriate satellite data and model data to support landslide science and disaster preparedness associated with landslides. And recognize how to map where landslides have occurred using optical and radar data and understand how automated tools can be used for this purpose.

Speaker 1: The prerequisite for this training is our Fundamentals of Remote Sensing course. We recently relaunched a new interactive online self-paced Fundamentals of Remote Sensing training on the RSET Learning Management System, and we hope you will check it out and share it with others. A link to this new online training is provided here.

Speaker 1: Over these two weeks, from March 11th to March 18th, there will be three one and a half hour sessions, which will include presentations, demonstrations, and question and answer sessions. All materials and recordings from each session will be available from the training webpage. If you are not able to attend one part, A recording will be made available within 48 hours on the RSAT training webpage. Homework opens on March 18th and will be due on April 1st. You will be able to access the homework from the training webpage on March 18th.

Speaker 1: A certificate of completion will be awarded to those who attend all live sessions and complete the homework assignment by the given due date. Part one of the training is focused on remote sensing for landslide science and disaster planning. By the end of this part one, participants will be able to identify the core concepts of landslide risk mapping, including geophysical and meteorological drivers, and how satellite data can be used for this purpose, as well as select appropriate satellite data and model data to support landslide science and disaster preparedness associated with landslides.

Speaker 1: Please put your questions in the questions box, and we will address them at the end of the webinar. Feel free to answer your questions as we go. We will try to get to all the questions during the question and answer session after the webinar. The remainder of the questions will be answered in the question and answer document, which will be posted to the training website prior to next week's training.

Speaker 1: It is now my pleasure to introduce the guest trainer for today's webinar, Dr. Robert Emberson. Robert serves as Associate Program Manager for NASA's Earth Applied Sciences Disasters Program Area. In this role, he helps administrate a range of projects funded by the Disasters Program, as well as supporting the disaster response activities by providing satellite data and insight to a range of stakeholders before, during, and after disaster strike. Alongside this programmatic role, Robert is involved in several research projects.

Speaker 1: He leads research that seeks to understand how satellite observations can be used to inform estimates of agricultural soil erosion, as well as how to medium-term weather forecasts can be used to predict extreme rainfall events. He is also part of the NASA Landslides Research Team, where his current research focus is to better understand and model the exposure and risk faced by human settlement due to landsliding. The primary aim of his work is to better model the economic and social impact of landslides globally and in near real time using satellite data.

Speaker 1: Robert, over to you.

Speaker 3: Okay.

Speaker 2: Well, thank you, Sean, so much for the introduction. Once again, my name is Robert Emberton, and I'm excited to be here to present to you on behalf of NASA and the OSSET program. I'm part of NASA's Disasters program, as mentioned, and yeah, this is going to be a great opportunity to teach and to learn.

Speaker 3: A bit more about some landslides.

Speaker 2: So I'm going to start talking today about the core concepts of landslides and basically introducing to you from the ground up the science behind landslides and what we can learn how to use satellite data to understand these systems more efficiently and effectively. So let's start by talking a little bit about what is a landslide. So this is a term that can be used quite widely around the world in different contexts. But in general, we define this as the downslope movement of a massive rock debris or earth.

Speaker 2: And that can be all sorts of different sort of volumes of material. This can be anything from more amounts of debris falling down a small hill slope to enormous amounts of material that can be up to cubic kilometers of material. rolling down hills or flowing down hills. And these can be in the form of rock falls, in other words, falling, toppling over slides that can be, you know, large scale movements at high speed, spreading material, and this can be anything from highly fluidized material and all the way up the kind of flowing debris flows, which is something that may describe a mixture of water and sediment that can flow for a very long distances.

Speaker 2: And so this is a whole range of processes that we're talking about here. And landslides, the term is often used to describe this. The picture you're seeing here, for example, from Taiwan, shows the scar from a landslide on a hill slope, but also potentially some material that has run out as debris in the river below. And so you can see that there are multiple different elements that could form part of this, both on the hill slope and in the river channels as well. So let's talk a bit about what causes landslides.

Speaker 2: So ultimately, when a material falls down a hill slope, it's undergoing gravitational failure. And that's when the downslope component of the gravitational force exceeds the body, the forces that are resisting.

Speaker 3: That downslope component. So that could be friction, that could be cohesion.

Speaker 2: Ultimately, this is when the force balance is, for a short amount of time or for a longer amount of time, exceeds.

Speaker 3: The downslope component, exceeds it, and that leads to movement.

Speaker 2: This also allows us to talk a little bit about a concept called failure plane. And this is the surface along which landslide material fails. And so that's, in other words, below that failure plane, there is no material movement, and above that failure.

Speaker 3: Plane, all the material moves.

Speaker 2: Now it's a simplified assumption, and typically we would see maybe kind of more complex systems where some materials fail below that sort of initial surface.

Speaker 3: And some material above may remain static.

Speaker 2: These systems are highly complex in shape often, But in general, we're thinking about the forces operating on that.

Speaker 3: Sort of failure plane.

Speaker 2: And so what we're thinking about really are the longer term and short term balance of factors, of forces that can affect this overall force balance. And so on long time scales, we might think about how that sort of the angle of these failure planes might change. But on short time spans, there may be triggers, and we'll talk more about what that means. and that will dynamically and rapidly change the overall force balance and lead to this gravitational failure. So thinking about this in a kind of general sense is easy enough, but we want to talk a little bit about the specific triggers that might be involved here.

Speaker 2: So water infiltrating into soils and rocks can reduce the cohesive and frictional forces in a landscape and the hill slope.

Speaker 3: And that's because as water infiltrates, it may reduce friction.

Speaker 2: It can change the overall buoyancy of the amount of material and the weight of that material. It can also increase the weight of the overlying material.

Speaker 3: And that can then potentially fail.

Speaker 2: And so you can think of rainfall as a very important trigger of landslides and increasingly intense rainfall or extremely long duration rainfall can be something that can trigger landslides.

Speaker 3: And generally, this is one of the most important factors that we consider.

Speaker 2: And this is particularly relevant because the flow of water and the movement of water can also further mobilize sediment and lead to the formation of downstream debris flows. And again, those systems can be significantly impacting the people and populations who live long way from the initial landslide source. So this is a factor we'll come to in the future a lot is rainfall triggered landslides and we can think a lot about how satellite data can be used to understand these systems. Thinking a bit about that in more detail, it's important to note as I mentioned that the rainfall intensity and duration are relevant to determine the impact from landslides.

Speaker 2: So you can think about long, long-lived sort of low-level rainfall that can increasingly saturate a hill slope and change the weight of the material.

Speaker 3: Overlying a failure plane.

Speaker 2: But you can also think about very, very short bursts of rainfall, very intense rainfall. Thunderstorms and hurricanes and so on can introduce this kind of rainfall. And that can lead to very short spikes in the level of water within a hill slope, and that can change the overall amount of off-balance and therefore lead to failure as well.

Speaker 2: Scientists have thought a lot about the thresholds for landsliding, and generally these are based upon, particularly when it comes to rainfall, the intensity and duration in combination with one another. Here I'm showing here is a compilation of data sets. I'm looking at the intensity and duration of rainfall prior to debris flows, and this is used to sort of understand, in this case.

Speaker 3: Specifically for debris flows, but you can see.

Speaker 2: That at low durations, so this is on the far left of the x-axis, you require a higher intensity of rainfall.

Speaker 3: To trigger a debris flow.

Speaker 2: And that's that kind of these blue points. But then as the duration of rainfall increases and goes on for longer, you might require a lower intensity of rainfall overall. And that's observed in this empirical data set. And so, again, when we're building these kind of more statistically based analyses we'll talk about later on in the training course, we want to think both about this threshold of both intensity and duration. So these are key factors to consider side by side. So many of you will also be familiar with earthquakes as a trigger for landslides.

Speaker 2: This is a really significant element in many parts of the world. Mountain belts where landslides occur are often highly active locations for earthquakes. And so the relationship between this earthquake activity and landslides is often very kind of interlinked. And so we know that increasing amounts of shaking as a result of seismic activity can lead to, again, maybe a reduction in the cohesion in a hill slope or a change in the overall frictional state in a hill slope. And that, again, can lead to landslides.

Speaker 2: And an important note here is that when we think about rainfall, as we mentioned in the last slide, the amount of water that infiltrates and saturates a hill slope tends to, because of gravity, increase towards the bottom of a hill slope. And that means that you have this greater saturation of water typically lower down in hill slopes. However, seismic activity tends to increase the amount of shaking as you go higher up in a hill slope. And that's because the seismic waves get amplified the further you are up a hill slope and the kind of overall hill shape is amplifying those seismic waves.

Speaker 2: And this means landslides triggered by earthquakes tend to occur higher up slopes than landslides triggered by rainfall. And so this is something to bear in mind when thinking about the overall population of landslides around the world and in specific areas. And so we do often think a bit about those two different types of triggering processes and how they may lead to different populations of landslides. So the image here is showing shaking intensity as a result of earthquakes and linking that to the potential likelihood for landslides.

Speaker 2: This is the USGS's ground failure landslide model, which is now operationally deployed around the world. And this is something that as they increase their understanding of the earthquake source and the earthquake triggering, this will lead to an increased, a better understanding of whether landslides will occur. We'll talk more about this in upcoming days. It's important to note as well that rainfall and earthquakes are not the only factors that can immediately trigger landslides. Certainly the action of water occurring through freeze-thaw processes, so in other words where cracks in the surface are filled with water that freezes that can then create increasing those gaps and expand those cracks that can lead to fracturing and failure of those hillslopes potentially.

Speaker 2: And so that kind of like for free store processing, particularly for rock falls is a relevant factor for landslide failure. And then as many of you will be aware, humans are an important influence on the landscape in many places, both by deforestation, urbanization, or for example, loading.

Speaker 3: Of water due to pipe leaking.

Speaker 2: These are all processes that ultimately still can affect the overall force balance in a hill slope. And so human-driven processes and, for example, road building is a good example that can, on both long and short timescales, affect landslide triggering. So we must consider those when we're thinking about urban and peri-urban settings for landslide hazard analyses. Volcanoes are obviously also an important trigger of landslides in many places where those are prevalent hazards. Many of you will be familiar with Mount St.

Speaker 3: Helens. The Mount St.

Speaker 2: Helens volcano, the devastating impact of this was mostly due to a large landslide that occurred as the entire flank of that volcano failed as a result of the eruption, which led to an enormously large debris flow. And so these kind of associated hazards with volcanoes, these landslide associated hazards, again, are a really important factor. And it is also worth mentioning on coastal settings, erosion of cliffs or nearby surfaces as a result of waves can also trigger destabilization and failure as a result of kind of debris flows top of the landslides.

Speaker 2: And so again, this is not something limited to inland settings. This can certainly happen on the edge of coastlines as well. So moving on, let's talk a little bit about the effects that landslides have. Here I'm showing an image that you'll probably see a few times in the next few weeks. of landslides in Taiwan, and this is a catchment in southern Taiwan, where an enormous number of landslides occurred as a result of this typhoon called Typhoon Morakot in 2009. This was a very large rainfall event, perhaps one of the largest ever in Taiwan's history, that triggered thousands and thousands of landslides all across the island.

Speaker 2: And you can see here this image from Google Earth, where multiple landslides in these high hill slopes are obviously evident. And the run out of debris and materials was significantly impacting population centers further downstream from these landslides. So again, these are very large systems potentially that can influence human systems and population. I'm almost talking a little bit about that next. So on a very basic level, landslides in many steep mountain belts are the source of sediment. And so Obviously, rainfall can act to move small amounts of sand and grains of clay, for example, from very high upstream places all the way to downstream places.

Speaker 2: But that process is typically not as important as this large-scale movement of landslide material.

Speaker 3: Through gravitational failure.

Speaker 2: And so in very steep mountain belts, it's been shown that the dominant process through which material sediment and erosional material is both eroded and transported is is through landsliding. In other words, in these very steep settings, landslides are the most important process through which mountain belts are shaped and changed. You can see in this image, this is a sediment plume in the southwest coast of Spain. Extreme rainfall in this setting drove intense amounts of landslides, and all of that sediment ends up in the rivers, which then ended up out into the ocean as well.

Speaker 2: So the source of the sediment is typically from these landslides. and that's something we can observe space using exactly this kind of methodology, using MODIS and others to observe.

Speaker 3: That off the coast.

Speaker 2: The material that landslides produce and the erosion processes involved can also influence downstream biogeochemical processes. So here's an image, this is from field work I was actually involved with, looking at bacterial communities.

Speaker 3: In water filtering through landslides in Taiwan.

Speaker 2: From the image I showed you before, now we're zooming into the actual catchment itself. And as water filters through this new material that has been excavated by these landslides, it acts in some ways like a fresh source of bedrock and rock that can then dissolve very efficiently as a result of increased rainfall and creates these very sort of chemical-rich outflows in some places. And that can lead to changes in the biological communities. And here you can see this red algal blooming or this algal growth in this outflow.

Speaker 2: And that's a direct result of this landslide based process. So again, this is a not just a physical process of erosion, but also a chemical weathering process and biogeochemical processes.

Speaker 3: That is influenced by landslides.

Speaker 2: So perhaps the most important thing to consider with landslides and the effect that they have is the effect and impact on human populations. We know from long amounts of extensive research that landslides cause thousands of fatalities every year and potentially billions of dollars of damage. Research including from the study whose image I've cited on the bottom here, we know that typically that occurs in the summer months in the northern hemisphere. But again, it's distributed all over the world and impacts and fatal landslides can occur at any time of year and around the world.

Speaker 2: So understanding how they interact with human populations is critical. Linear infrastructure is particularly vulnerable in mountainous regions as well.

Speaker 3: And so that includes things like pipelines, roads, power lines.

Speaker 2: And that's because whilst they may, individual segments of those linear infrastructure elements may be likely to be impacted by a landslide, over long distances, a single landslide can still cut those sort of linear sections. And as a result, the points connected at either end of those linear sections, maybe the connectivity between those sections may be reduced or impacted. So again, these kind of long linear sort of elements are something we really have to pay attention to when it comes to landslides.

Speaker 2: Downstream communities may also face elevated risks for many years. And that's because when a landslide is triggered, it can often sit as mobilized debris on a hill slope for many years. But when rainfall occurs further and tighter, like later on, that can then remobilize that sediment and form a debris flow that can then impact communities further downstream. And so understanding the long lifespan of these systems and how they may impact downstream communities is particularly important. And then finally, it's worth noting that peri-urban development is particularly vulnerable.

Speaker 2: And this means communities that are living on the edge of typically large urban developments, these may be less, maybe the socioeconomically disadvantaged people, who are unable to live in places that are more secure, often on steeper hillslides and without adequate support for landslide mitigation methods. So these may be very vulnerable to landslides. And again, these kind of human triggered landslides may be most important in those settings as well. And that's a subject of increasing research.

Speaker 2: So based on studies by NASA scientists, we have demonstrated that population exposure to landslide hazard is widely distributed. So the NASA's Disasters Program, amongst others, has supported research to explore how satellite data can be used to fill in gaps around the world in landslide understanding. And so, for example, here we're showing the number of people exposed on an annual basis to potential landslide hazards. And you can see that even though the mountain ranges where we might expect there to be increased landslide hazards, like the Alps, the Himalayas, are clearly shown, There are also parts of the global south, including Central Africa and West Africa and Indonesia, where there's extremely high levels of landslide hazards and exposure of populations.

Speaker 2: And these are the kind of data that we can only generate using satellite imagery. I'll explain why that's the case in coming slides. So let's talk a little bit about the factors that can influence the occurrence of landslides. So many factors may be relevant in leading to the likelihood of a landslide occurring.

Speaker 3: So for example, the.

Speaker 2: Most obvious thing you can think of is slope. This is, as you might imagine, the steeper hill slope, the more likely it is to fail catastrophically. So here on the right-hand side you're seeing real data exploring the pixels in a sort of, in a part of Zimbabwe where a lot of landslides occurred. And we're looking at the slope in that pixel based on a digital elevation model and comparing it to the average across that landscape. And in landslides, we can see on the top level that there is a distribution that is slightly, the slope is slightly higher than the landslide average, landscape average, I'm sure, I'm sorry.

Speaker 2: And then the middle slide is the average topography itself. So you can see that in general, the slope is lower than the landslides themselves. And then we can create a kind of probability ratio of these two, and we can see that landslides are much more likely to occur at higher slopes. So this is real data that bears out this, what we probably would assume. And again, topography is pretty much the most important element of why steeper slopes are more likely to fail gravitationally. And again, there are elements that go alongside this, including relief aspects and position in landscape that can also affect the influence, so the influence, the likelihood of landslides.

Speaker 3: And we'll talk a bit more of that in a few minutes. It's important to note, though, that slope.

Speaker 2: And topography are not the only thing that can, that affect landslides in long timescales. The type of material under gravitational stress is also a major factor that influences landslide potential. So here I've picked a couple of images from very famous landscapes around the world. And this is Half Dome in Yosemite in California in America and the Grand Canyon, also in the sort of southwest of the United States. These are two settings where you can see some very, very steep hill slopes and almost sheer in some cases hill slopes.

Speaker 2: These remain stable for long periods. And so there aren't landslides occurring every day on these hill slopes, despite how steep they are. And that's because either they're made of very strong material. And so on the top section, this is almost a solid block of granite. And so it's very unlikely that it's going to fall apart. You can imagine, conversely, an isle of sugar would not be able to reach that sort of steepness. But when things are all held together in these very strong rocks, by strong rock formations like this, that is certainly a way to reduce the likelihood of landslides.

Speaker 2: And again, you can see in the Grand Canyon here, these rocks are almost layered one on top of each other in a horizontal manner. And that means that there's a kind of much stronger likelihood that the failure planes could form as a result of the failure, the typical plane of, let's say, the rock strata, but with a flat. And so there aren't obvious failure planes within the Grand Canyon that could trigger landslides as a result of this layering of rock. And so typically you can now, as a result of that kind of horizontal layering, it is possible in some places to see much steeper hill slopes than you might expect.

Speaker 2: So again, this kind of lithology, either the type of rock or the overall structure of the rock can certainly affect the likelihood of landsliding as well. Not just obviously the rock underneath, but also the way in which the land is used is a really important way that determines likelihood of landslides. Particularly, the type and density of vegetation is a strong control on landsliding. Plants act in several ways to reduce or increase the likelihood of landsliding. Their roots are an important binding factor for hillslopes and for soil.

Speaker 2: They hold together large amounts of material. You can see here a tree that's fallen over. There's a lot of soil that's captured in that root ball. Trees that remain standing are quite an important stabilizing agent for hillslopes. Forested areas may be less likely to have a landslide than deforested areas. And similarly, roots and plants act to increase the rate at which water infiltrates and flows through subsurface areas. So in other words, the water moves quickly, more quickly through these areas, which reduces the potential for saturation.

Speaker 2: And as a result, typically, we would see a lower likelihood of landsliding. So again, there's lots of different ways that this can interact. And one thing I also note using this image here this tree has fallen over potentially as a result of extreme winds. And that is a way that large amounts of material can be then shifted in this case and potentially lead to failure of a hill slope as a result of what's called tree throw. And again, so there's lots of different elements playing together here, but understanding the way in which vegetation also acts to trigger landslides is certainly relevant.

Speaker 2: We should also talk a little bit about what we refer to as slow moving landslides. mostly we think about landslides as these catastrophic failures, these rapid motion from the hill slope all the way into the valley and material being moved very very suddenly leading to significant impacts and these often could reach 35 to 50 miles an hour. However some landslides can move very slowly at rates of millimeters or centimeters per year and this is typically when the balance of forces within a hill slope are quite closely matched but in general the perhaps the increase in saturation of water is slightly decreasing the friction or slightly increasing the weight.

Speaker 2: And as a result, there's a gradual motion of this material down slope. And this can lead to long-lived, typically very large landslides that can move downhill in very slow rates. And you can see here an image of a road that originally ran through unimpinged by landslides, but the slow movement of this hill slope from left to right has gradually broken that road apart and damaged it. So these are still damaging events. And it's also worth noting that these very slow moving landslides can still transition to very fast motion catastrophic events under the right kind of triggering conditions.

Speaker 2: So again, we should still pay attention to them, even if they're not moving quickly. And it's another factor that we can observe from orbit. So now I'm going to talk a bit about how we can use satellites to understand landslides and observe them. And this is something that, again, I alluded to in this kind of introduction. So let's dig into that in some more detail. So the Disasters Program and NASA more generally has spent many years supporting science to understand how landslides can occur and what leads to those failures and what it means for communities.

Speaker 2: And so we think about these elements across the entire disaster cycle. And so we think both on a kind of local to regional thing, understanding what has happened. We think about modeling of landslides to kind of understand what could happen. Then we move to think about forecasting and predicting landslides and really understand what is likely to happen. And then finally, we're thinking about exposure analysis and really trying to translate this into who might be impacted in the potential landslide occurrence.

Speaker 2: And so there's all kinds of different satellite data that can be used to inform this. And we'll talk more a little bit about each of these over the course.

Speaker 3: Of the training.

Speaker 2: But just to quickly highlight mapping on the right hand, top right, you can see landslides being mapped using satellite imagery. The bottom right shows an estimate of the changing hazard associated with landslides every month around the world. And then we're thinking about the bottom left, the potential changes in landslide impacts as a result of El Nino status. And that's something we can then predict several months in advance. And then finally, looking at kind of the top left, We're talking and looking at the potential for landslide impact at a county level and the population exposure in those individual counties in certain parts of the world.

Speaker 2: And so all of this is accomplished using satellite imagery and data. And we're gonna teach a little bit more how you can do that in the next few slides. So what are we thinking about in terms of observing landslides? So obstacle data, so in other words, visual kind of imagery that you would imagine that you often think about as the typical sort of satellite imagery has for decades formed the primary tool for landslide modeling and essentially observing where landslides have occurred from space.

Speaker 3: You can obviously go into the field with a camera.

Speaker 2: But in many places around the world, landslides occur in mountain belts that are highly inaccessible. And so satellites form an important element to get to those places and observe them when people even on foot are unable to. And that's particularly important when, as I mentioned, material can stay in the very high parts of catchments for many years before it becomes devastating downstream. So it really helps to use satellites to do this. And so for many years, systems like Landsat and Sentinel-2 have provided large-scale assessment with long archives at this kind of moderate resolution, around 30 meters per pixel, to understand where landslides have occurred.

Speaker 2: And this helps us categorize and map where the largest landslides around the world have occurred, And these tools remain useful even today, even although we do now have access in many cases to higher resolution commercial data sets that can be used to observe landslides at much higher fidelity and potentially observe landslides that may be smaller than we were able to observe with these older.

Speaker 3: Landslide data sets.

Speaker 2: It's important to note that cloud cover is a critical challenge. Obviously, seeing through cloud with optical imagery.

Speaker 3: Is not possible.

Speaker 2: We'll talk a bit more about how we can get around that. This is a critical consideration with landslides because, as mentioned, rainfall is a primary trigger for many landslides around the world. And so if we are looking for landslides in the short aftermath of a major rainfall event, the clouds associated with that event may block us from seeing that.

Speaker 3: So something to bear in mind.

Speaker 2: I also mentioned that spectral bands associated with those optical data sets, that includes, for example, the analysis of data from the red end of the optical spectrum particularly to understand, to build images of NDVI, or Normalized Difference Vegetation Index. This can be used to potentially look at landslides.

Speaker 3: In higher contrast imagery.

Speaker 2: So I showed just before that an image of individual landslides, a very small, I guess, sort of single landslide event. But the benefit of these optical data sets is that we can then observe large amounts of landslides that occur as a result of major rainfall events or earthquakes, and build large-scale inventories. And so this new image I'm showing is essentially an inventory of landslides. And all these red blobs are where landslides have occurred. And this is something that then allows us to build a much more comprehensive understanding of this at large scale.

Speaker 2: Again, as I mentioned, you could go into the field and take an image with a handheld camera, but you are gonna struggle to see this many landslides in one single image, unless you're very, very lucky or very unlucky, I guess. And so satellites allow this kind of large scale perspective that helps us understand landslides on a regional perspective. And we'll talk more about how we can do this and systematically and efficiently in the course of the training. I've mentioned landslides are critically often seen as a removal of vegetation.

Speaker 2: And so when we look at this normalized difference vegetation index, the red part of this image shows where trees are and the polygons here, the black outlined areas are landslides. And you can see particularly this very large landslide, it's exposed a lot of material, deeper bedrock material that is no longer vegetated. And so using that normalized difference vegetation index particularly is effective to understand the change between deforested areas or vegetated areas to the deforested and landslide impacted areas.

Speaker 3: So again, this is an.

Speaker 2: Effective tool to help us see where these have in fact occurred.

Speaker 3: As I mentioned, landslides.

Speaker 2: Often occur as a result of rainfall and may therefore be imaging of those using optical data.

Speaker 3: May be impacted by clouds.

Speaker 2: And so we can also use a different type of data called SAR or synthetic aperture radar This uses pulses of radio waves to create two- or three-dimensional surface images.

Speaker 2: The science behind this is extremely clever, and I'm delighted that landslide scientists have been able to use this now to effectively characterize where landslides have occurred. And I would strongly encourage those who are interested to take the additional asset training available. And the biggest advantage here is that SAR can see through clouds. It's not limited by cloud cover and so we're able to image the entire surface of the earth day.

Speaker 3: Or night with this data set.

Speaker 2: And so what we would typically do is use this kind of data set to image the surface prior and or take an image before a landslide triggering event and then compare it with one that occurred after that triggering event and compare the surface change and then we can look at the difference between those two Surfaces is a way for us to understand the landslides. And we'll talk a lot more about that in part two of this training. But just in short, we can say that there are two different approaches in general.

Speaker 2: And that's to use the change in the coherence.

Speaker 3: Of the cell signal. That's the coherence between the radio waves that.

Speaker 2: Are sent out and returned. And then others use the difference in the amplitude. So just essentially the amount that's received back at the satellite. And so this is a really novel method that's been highly effective in recent years. And we'll talk more about that in the next part of this training. Again, we can also use certain other types of data sets that can either combine optical or star imagery. Often thinking outside of the orbital context, this is mounted on unmanned aerial systems or manned aerial systems.

Speaker 2: So in other words, aerial photography or drone photography is typically widely used. in emergency response settings where you need to get there rapidly and get underneath the clouds. You can see here an image of a debris flow in Colombia, South America, taken from, I think this was helicopter imagery. And again, this is very effective for this kind of short-term analysis. But one note here is that these are typically harder to geo-reference. In other words, harder to tie down the locations to existing imagery.

Speaker 2: And so it can be more challenging to use aerial imagery in combination with satellite imagery.

Speaker 3: Just a note there.

Speaker 2: We've also looked at using, or scientists have looked at using, LIDAR, which is laser imaging protection and ranging. In other words, this is a way in which we've looked at using laser, I guess this is pulses of laser beams that are sent out and come back to assess the change in texture as a result of landslides or built models of the surface. For more information on that, we encourage you to check out the resources provided. And then Again, the texture that associated with landslide deposits is often described as hummocky, and you'll probably hear this term at some point.

Speaker 2: What it means is that there's this kind of bolder field style surface, often these large landslides, and we can often look for that in surface maps as a way to assess where landslides, historic landslides have often occurred and build estimates of those ones, even if they're now forested, old landslides that we haven't got before and after imagery.

Speaker 3: For, we can use this texture and where they have occurred.

Speaker 2: And then finally, I mentioned that new satellite data sets of hyperspectral data. So in other words, imagery that has data sets across a huge number of spectral bands can also help us understand the location of landslides.

Speaker 3: This is very new science.

Speaker 2: And so there's a lot of potential methodologies that could be applied here, but particularly looking at changes in subsurface on the surface mineral structure or mineral composition as a way to assess where have occurred is potentially promising science and I suggest that that is a novel path forward and NASA has a number of hyperspectral satellites that are either already in orbit or are soon to be launched and we're excited to see how we can use that in.

Speaker 3: The land flight science community.

Speaker 2: So just a word to the parameters that you should consider when it comes to the satellite data sets. The primary one that most people pay attention.

Speaker 3: To is resolution.

Speaker 2: So you can think about this both in terms of the spatial resolution, in other words how small or how large are the pixels associated with that analysis. Obviously smaller pixels may be better to assess smaller landslides. And we'll talk a bit about why this really matters, but this is, it leads to reduced, a higher resolution image can often prevent amalgamation. And what that means is when you're mapping landslides or looking to sort of pick out where landslides are on an image, if you have a very small, a highly, highly kind of coarse image of one or two, even if there are many, many landslides occurring within those pixels, you'd probably only be able to say this one pixel is one landslide.

Speaker 2: And so you would amalgamate a lot of individual landslides. that matters for hazard and impact assessment. We'll talk about that later on. Of course, there is a trade-off between the resolution, higher resolution data, and the sort of overall footprint of that image. So in other words, you might be able to get very coarse resolution data that could cover a very large area. But finer resolution data may typically be over.

Speaker 3: A smaller band swath width.

Speaker 2: So again, if you're looking at very large landslide events, it might be useful to have both the synoptic scale assessment as also a kind of a local scale and high resolution.

Speaker 3: Analysis as well.

Speaker 2: The next thing to talk about is latency, low latency data. In other words, data that's available on an extremely short amount of time after a landslide triggering event, and let's think maybe within a day or so, is particularly extremely important for emergency response analysis. There's been a lot of use of this by many teams but including NASA's scientists through the Disasters Program to support the analysis of where landslides have occurred in the very short aftermath of triggering events. And that often means we need data that can be captured within a day or so.

Speaker 2: And so, again, this is also relevant in places where there is increased amounts of cloud cover. For example, in the Himalayas where the monsoon occurs, there is often cloud cover for months. And finding images that can capture the small windows where there may not be clouds may be much more likely to do that if you have low latency, high.

Speaker 3: Repeat rate imagery to improve that coverage.

Speaker 2: And then finally, just to mention that spectral bands, in other words, the types or the bands of the wavelength of the imagery that is collected An important note here is that if you're doing pre and post-event analysis, it's very useful to use the same kind of spectral bands to ensure that you're comparing like with like. And again, as mentioned, landslides typically remove vegetation. So spectral bands that are sensitive to biomass are likely highly beneficial. You can also think a lot about the triggering factors that lead to landslides and how we can observe those using satellite data.

Speaker 2: So as we mentioned, rainfall is one of the most important factors for triggering landslides. And for many years, NASA has observed rainfall.

Speaker 3: From orbit using the GPM, or Global Precipitation Monitoring Mission. And this has been extremely.

Speaker 2: Beneficial for science, and we'll talk about in part three of this training course, to understand where landslides have occurred or are likely to occur as a result of extreme rainfall. Here I'm showing an image of the rainfall due to Hurricane Iza from a few years ago in 2020. This hurricane impacted Central America significantly and led to a large number of landslides and using this GPM data of rainfall, our team was able to assess where the landslides are likely to occur and which matched up very effectively with where the actual landslides did occur.

Speaker 2: So we'll talk more about how to build those kind of general models in due course. We can also use soil moisture data measured by systems like SMAP, Soil Moisture Active Passive Mission. to assess the kind of precursory soil moisture conditions. In other words, if soil is very saturated, there might be a much higher likelihood that landslides could occur. And then finally, we mentioned that earthquakes are a primary trigger for landslides as well. This is typically an assessing earthquakes as a trigger is conducted using terrestrial seismometers based on the ground.

Speaker 2: But the post-earthquake assessment of how much shaking has occurred and how much deformation has occurred and how much landsliding has occurred is highly now reliant upon optical and far data typically in the emergency response and recovery community. We also talked a bit about the factors that lead to landslides or the static factors that can contribute to an increase in the likelihood of a landslide occurring in a given location. Things like topography, the type of vegetation or type of land use.

Speaker 2: These are all elements that can be observed from orbit and using a variety of different data sets. So for example, Land use is widely characterized using MODIS or Sentinel or Landsat data sets. These are really critical and form a lot of the analysis of landsat prediction. Vegetation and the loss of forests, again, can lead to a high likelihood of landsliding. And Landsat has been used effectively to characterize this over many years. And this has been widely incorporated into landsat analysis. And then fundamentally, digital elevation models built through either from satellite sources globally are perhaps the most important data set used in landslide science.

Speaker 2: So we have maps of slope from all over the world. We typically rely on the Shuttle Radar Topography Mission as data sets, but the newer NASA DEM data sets are also highly used.

Speaker 3: In more modern research.

Speaker 2: And so this is a fundamental element where satellites have contributed to the most important part, one of the most important parts of landslide science. It's important to note that many of these datasets may exist at different resolutions or the observation time. And so the combination of these different datasets certainly requires care and consideration. And we'll talk a bit more about that when discussing landslide susceptibility in the part three of this training course.

Speaker 3: So in this next section, I'd like to talk a bit about ancillary data.

Speaker 2: And what this means really is other datasets that can help us build a more comprehensive understanding of where landslides and where they impact. To do this, we need to think about not just the landflights as a kind of individual component, but how they fit within the overall understanding of hazard risk and so on. And so I'm going to introduce a few terms here to help illustrate this. So we often talk about susceptibility, and that would be essentially an assessment of where a landslide is likely to occur.

Speaker 2: Hazard, however, is not just where it's likely to occur, but where and when. So there's a temporal component involved in that. And as mentioned, this hazard will consider the triggering factors like an earthquake or rainfall.

Speaker 3: We often think about exposure.

Speaker 2: That is the who or what is in the hazard zone.

Speaker 3: And so that could be population or infrastructure.

Speaker 2: And understanding what that means and where it's located in comparison to where the landslide hazard is located is critical. The next element we want to talk about is vulnerability. And this is a system basically thinking about how likely is damage or injury to occur for a given level of exposure. And there's a lot of factors that build into vulnerability, and we'll talk more about that. But generally, that's something that is more challenging to do from satellite data.

Speaker 2: And the reason for that is because vulnerability typically is a factor of not just the, let's say, for example, for buildings, it's the type of structure, it's how solid are those buildings, But for populations, it's even things like the healthcare outcomes. It can be things like vulnerability can be considered, including the socio, the cultural history of understanding of what hazards look like and how they can impact your community. It can even be what the emergency service practices. So a lot of that tends to be not assessed from orbit and ground-based analyses and more kind of qualitative estimates of vulnerability typically are important to build these systems.

Speaker 2: And all of this builds towards risk, which is often what we want to know most. Risk is hazard, multiplied by exposure, multiplied by vulnerability. And what this gives us is an estimate of the total impact of a given hazard. And so building all those elements together is the fundamental part of designing a kind of overall impact model for landslides. And so we're going to talk a bit about now, like how we can use satellite data and other sources to inform exposure, and potentially even vulnerability to build a more effective risk model.

Speaker 2: So to kind of illustrate this and to think about how to combine.

Speaker 3: These data sets for landslides, let's think.

Speaker 2: About this as a layered model. So at the hazard end, the landslide hazard layer is typically derived from a model or an observation. So we could say, well, we know a landslide occurred here. Here's a picture of this, and here's our basic model of this. I've used here, as an example, a version from team at NASA, which is the Landsat Hazard Assessment for Situational Awareness, or LASA.

Speaker 3: We get a hazard layer from that.

Speaker 2: Exposure, in this case, could be population density or building footprints, critical infrastructure. There's a variety of different sources for input there that we can introduce. So here on this image, the exposure layer is showing a town, an hospital, and some roads. And then finally, coefficients of vulnerability. something again harder to kind of capture as a visual but essentially this could be assessment of what are the policies designed to mitigate disasters and for example for a for a hospital, you could say for a given landslide, is it a given impact, or sorry, a given hazard, an exposure level, would you get between zero and one, where one would be total destruction and zero would be sort of negligible destruction.

Speaker 2: A vulnerability coefficient could describe that. And so you'd multiply all those things together to produce an estimated impact, which could be either a damage cost or population for each of these individuals. So that's kind of how we think about this overall combination of datasets. So briefly, I want to touch on some of these other datasets that we can use to assess this kind of population and exposure datasets using satellite-based imagery or NASA-based data. Here, for example, we're talking about population, and the greatest population of the world is a NASA source data available from the Socioeconomic Data and Application Center, or CDAC.

Speaker 2: It's available at one kilometer resolution. It combines data from census sources, which is an important consideration, considering that some sources of census data can be incomplete in some countries. And so population assessment may be variable in some places. And this is available at five time intervals.

Speaker 3: From 2000 to 2020 in five year increments.

Speaker 2: And this has been used for many years as part of a variety of landslide assessments to assess where population impact might occur. A more modern counterpart of this is WorldPOP. This is a global open source data set available that uses not just offensive data, but also a variety of data streams, including.

Speaker 3: Satellite data of night lights, all of.

Speaker 2: Many of which were sourced from NASA inputs that have then been used to build this more comprehensive assessment at 100 meter resolution of population density. There's also additional demographic data available within this data set that can be then used to assess the potential population or the type of population, the number of children, for example, the number of elderly people. So this is a more advanced data set that has been used more widely recently and is typically, again, quite robust system to build with these more high resolution models that again may be beneficial, benefit from a hundred meter rather than one kilometer resolution data set.

Speaker 2: As we talked about infrastructure, we often want to think about where we can get data to describe that infrastructure, and that can be buildings or roads or pipelines or critical elements like hospitals, schools, fuel stations.

Speaker 3: These exist all over.

Speaker 2: The world, and finding data on these elements can be challenging to do comprehensively or consistently.

Speaker 3: For example, datasets that are available in the.

Speaker 2: United States, the Homeland Security Infrastructure Datastat, or HIFLD, is widely available in the States, but there is not necessarily counterparts in other countries. And so if you want to build global models and incorporate satellite data, open source data, for example, OpenStreetMap has been widely used in land-side modeling. alongside for more localized studies, again, those kind of localized data sets. On the right-hand side here, I'm showing an estimate of US infrastructure, talking about power transmission node density, amenity density, and road density.

Speaker 2: On that top-hand side, you can see the kind of dense population centers of New York and East Coast cities. And we can use that kind of data sets derived from OpenStreetMap to, again, incorporate that with a landslide hazard model to assess the likely infrastructure locations that.

Speaker 3: Might be impacted by landslide hazard.

Speaker 2: Again, also important to remember that the typical type of infrastructure elements can be multidimensional. So we can think about this as either points, that can be just the location of an individual hotel or hospital, that would be very dimensional. We can have roads, which would be one-dimensional lines. We can have polygons, which would be building footprints, for example. And so you'd have the entire building polygons outlined two-dimensional objects, or sometimes even three-dimensional objects, which in kind of high resolution localized studies could be thinking about the three-dimensional shape of buildings as well.

Speaker 2: So each of those kind of types of data would either have different kind of volume of data involved in it, and certainly would change the file size characterization, but also the way in which that would be used with landsliding datasets would be important to consider. For example, the two-dimensional polygons, you might have the building footprint available at very high resolution.

Speaker 3: You can look at the individual corners of a building, but if your landslide.

Speaker 2: Data set is only available at one kilometer resolution or coarser, it may be irrelevant to look at that two-dimensional data when all you would need is a point to compare it with the landslide assessment. So again, thinking about the relative scale of your landslide hazard versus your infrastructure data is certainly a consideration when incorporating ancillary data like this.

Speaker 2: So thinking about satellites, combining this kind of ancillary data with satellite data, we often think about doing kind of raster-based analysis. And so in other words, we're thinking about pixels, individual pixels, and where the infrastructure is and how that can be combined with a satellite hazard. This is typically very effective at large scale. Gridded satellite data, either land use or rainfall, any of the kind of data sets we've mentioned are typically available as pixelated data. or raster-based data sets.

Speaker 2: And so constructing raster assessments of exposed elements is also relevant. So here, for example, I'm showing some images from Northern Italy in the Alps. The top estimate is the raster-based estimate of population exposure with redder colors showing higher exposure to landslides. The middle image is showing the road exposure. And this is, again, a raster-based assessment with a rasterized road network. And then finally, we're looking also at the bottom, infrastructure exposure to landslides. And again, you can see in the alpine areas, there's high population, higher road, and higher infrastructure exposure in the setting.

Speaker 2: And this again, useful at regional scales, but at very fine quarters of small resolution, so very kind of localized settings, the use of this may break down. I've mentioned if you have a very poor status set for Harvard, it may be irrelevant to compare that with And we'll talk about that in a second. So, and the reason I mentioned this is because the alternative to a raster-based assessment is a vector-based assessment. And so that's when the, there's a scale invariant data. And so any zoom level, if you like, you will still see the same points because the point, let's say the edges of a building, for example.

Speaker 3: Are known at any zoom level.

Speaker 2: To think about specific infrastructure elements. And here, what I'm showing here in this image is building footprints derived from OpenStreetMap overlaid on a landslide hazard estimate. You can see that the hazard level here is either moderate or high. This is a qualitative estimate, and I won't go into too much detail on how that's calculated. But in short, we can look at, for example, the area surrounding a given building footprint and estimate the potential hazard associated around that building footprint.

Speaker 2: as a way to potentially characterize or qualitatively characterize the difference in relative level of hazard in the setting. So again, thinking about not just the buildings, but maybe the building, the buffered area around it is certainly a way that has been.

Speaker 3: Used for these more localized analyses.

Speaker 2: Whereas at much coarser scale, at one kilometer resolution, we would have to look at the kind of density of buildings rather than the individual footprints to.

Speaker 3: Get an answer here.

Speaker 2: So again, this kind of trade-off between understanding things at a fine accuracy locally and doing things more comprehensively at a global scale that tends to be a local scale, we think more on a vector basis, and then globally on a raster basis. The only exception here would be potentially linear infrastructure, which can run for very long distances, and combining a kind of vector analysis of that would be a rational way of approaching that. The other thing to mention at the very fine resolution, and this is something we'll touch on more in parts two and three.

Speaker 3: Of this training, is that higher resolution.

Speaker 2: NLCs like this should also consider the source and run-out zones of a landslide. So in other words, where a landslide is triggered from is the source region, and where material ends up is the run-out zone or.

Speaker 3: The deposit zone.

Speaker 2: Obviously, if you're living in a building, you wouldn't want to be in a building that's sort of situated in either of those zones. But of course, if you're thinking about the downstream impacts, you want to make sure that you have a clear understanding of where material is likely to end up, as well as where it's sourced from. And so for this kind of finer scale analysis, that would be an important consideration. So let's give an example of how we can combine all this together with different kinds of data sets.

Speaker 2: So I'm going to use Rio de Janeiro, the city in southern Brazil, as an example. The bottom left here shows qualitative hazard estimates produced using a range of data showing moderate and high hazard for landslides. This is from what's referred to as the LASA Rio system. This was a tool developed in conjunction between the Disasters program supported Landslide Hazards team as well as the Rio de Janeiro city government. So we can estimate on a kind of short time scale what the Landslide Hazard looks like in parts of the city.

Speaker 3: So that's the bottom left.

Speaker 2: And then on the top right is a population density at three meter resolution. So very, very fine resolution population density.

Speaker 3: Estimate.

Speaker 2: So we can also build that we can combine those two together we can multiply those hazard and exposure together to estimate an estimate of exposure population and now we're aggregating this at a city barrier or a neighborhood level And so we would calculate the people exposed to moderate or high hazard as a.

Speaker 3: Number in these settings.

Speaker 2: And we can then aggregate that information to provide a confusable data to stakeholders who might then benefit from it. And so we've pulled an example from one specific barrier or neighborhood in Rio.

Speaker 3: This is the exposure level.

Speaker 2: We can estimate the fraction of a population exposed to either moderate or high hazard. And that information can be then used either on a short-term scale or a long-term scale to assess the change in either kind of how landslide management has occurred or supported better practices, or whether it can be kind of, we can compare different neighborhoods to understand the relative level of acid as well. As mentioned, we can also think about how we can use linear infrastructure data. Here, for example, is a road network map for a small section of.

Speaker 3: Rio de Janeiro.

Speaker 2: You can see these highlighted roads, either in red, green, or yellow, for example. those are the roads that are exposed to landslide hazard. And this is calculated as the potential, the relative length of the road segment exposed to the type of landslide hazard that has maps produced from the Rio de Janeiro model. So in other words, we're looking at roads that are like red colored roads or roads that are significantly exposed to high hazards here. And the critical thing here is that you can look at the road segments, particularly through the mountainous areas in the sort of center left of the image, that are the connect points that are maybe long distances apart.

Speaker 2: And high exposure to the landslide hazard might mean that anyone stuck on those roads are unable to make an alternative choice. And they are then, even if a single landslide occurs on a road segment like that, it might block that entire segment and make that no longer useful as a sort of connection. So again, considering how connections may be affected by this is a sort of further derived analysis that comes from this. But it's something that has been highly beneficial to incorporate this vector-based assessment of roads with a landslide hazard analysis.

Speaker 2: So again, this can help guide emergency planning processes, as well as community planning for these bottlenecks that might occur. So hopefully that's been an interesting introduction to some of the science that sits as part of satellite data and landslides together. In the next two parts of this course, we're going to go into much more detail on how we can use satellite data to delineate and map landslides, both manually and automatically. We're also going to understand more about how we can exploit that satellite data to observe landslides and really understand the best types of data and the way in which we can use it most efficiently.

Speaker 3: We're then going.

Speaker 2: To talk a bit more about how to build susceptibility models. In other words, estimates of where landslides are likely to occur, and we can combine different data sets from different satellite.

Speaker 3: Sources to build a more comprehensive understanding of that.

Speaker 2: And then finally incorporating all of that to build these kind of hazard models. And I showed an example here from Rio.

Speaker 3: That puts together different inputs.

Speaker 2: From rainfall and from topography and so on. And we're going to talk a lot about how we can put together in part three of this training to hopefully help inform some of that. So I would also encourage you to check out some of these resources, including some excellent RFET trainings on various types of data sets that were mentioned, as well as information on the NASA Landslides Research, as well as the NASA's Disasters Program, which has supported a great amount of this research. And I'm grateful for your participation.

Speaker 2: I look forward to questions and your involvement in the next two parts of this course as well. And with that, I'll say thank you and pass it back to Sean.

Speaker 1: Robert, thank you for the wonderful presentation. The following summarized the concepts covered in part one of the webinar series. The term landslide encompasses dynamic processes ranging from localized rockfalls to highly mobile and fluidized debris flows. Rainfall-triggered landslides are a critical part of landslide activity. Earthquakes, freeze-thaw processes, human-driven processes such as deforestation and urbanization, volcanoes, and wave-driven erosion can also drive landslide activity. Topography is among the most important elements influencing the likelihood of landslide occurrence.

Speaker 1: Remote sensing observations, optical and radar, are critical for the global monitoring, assessment, and response to landslide events. Several remote sensing parameters, such as spatial, temporal, and spectral, are key for satellite observations of landslides. Combining ancillary data, such as population and infrastructure, with satellite observations allow satellites to model landslide risk at the global scale. Before we transition to the question and answer session, I want to remind you there will be one homework assignment which you will be able to access from the training page on March 18th.

Speaker 1: Answers must be submitted by Google form with a due date of April 1st. To receive a certificate of completion, you must attend all live webinars and complete the homework assignment by the deadline. You will receive a certificate via email approximately two months after the completion of the course. Below is the contact information for Dr. Emberson. along with links to the RSAT website and social media. If you enjoyed today's webinar, we hope you will sign up on the RSAT listserv to receive notifications of future trainings, and follow us on social media for other relevant announcements pertaining to NASA's Earth Sciences.

Speaker 1: We will now transition to the question and answer portion of today's training. Please, reminder, enter your questions in the question and answer box, and we will get to them in the order that they were received. We will post the question and answer document to the training webpage once we've finished answering and editing all questions. And thank you to everybody that's been submitting questions. We've gotten some really good ones thus far. To jump right into it, I know, Rob, it looks like you've been answering many of these.

Speaker 1: So the first question being, do soil properties influence the determination of failure planes in landslides? If so, would soils for the same properties have the same failure plane, assuming other factors are not considered? Rob, over to you.

Speaker 3: Thank you, Sean. And just to say, yeah, thank you for these questions.

Speaker 2: You're absolutely right to point out that soil properties are important and they certainly influence landslide likelihood in many locations. Typically clay soil, for example, may lead to more specific failure planes aligned with the clay particles.

Speaker 3: But it's important to consider this as part of a larger whole.

Speaker 2: Very shallow landslides, the triggering processes will depend very much on those soil properties if they're within the soil itself. But at larger scale, landslides may excavate as much as down to the bedrock.

Speaker 3: Point there where the soil profile right at the base of that.

Speaker 2: Might be another significant point of potential weakness and failure. And then even larger landslides might fail on bedrock faults or fractures. And those might be the things that influence that failure plane. So soil is not the only thing I think it's important to note. But again, yes, there are specific soil types and how that interacts with climate and vegetation will be a strong determinant.

Speaker 3: Thank you.

Speaker 1: Okay, wonderful. Question number two. How can geological material data be integrated with optical remote sensing data? Do you recommend using local geology or regional geology?

Speaker 2: It's a great question and one that we'll talk a bit more about in the part three of the training. But to say for now, I think it's very much important to consider the type of rock and the lithology and morphology of the rock types. Again, this has been a challenge to assess from satellite observations. the surface characteristics of certain specific rock type or minerals on the surface have been assessed with some types of satellite data, including hyperspectral data. I know there's an excellent offset training on that if you're interested in exploring more about that kind of information.

Speaker 3: But it's important to.

Speaker 2: Note that many of those assessments are only, let's say, skin deep. They can only see the very top surface of the material. And so understanding what lays below and what might influence the landslide potential for failure does often.

Speaker 3: Require, at best, field work. And so this is, again, regional data sets are pretty critical.

Speaker 2: We've used some global data sets, but again, that's relatively coarse in terms of the accuracy or reliability in terms of determining those landslide parameters. And again, one more thing I'll say that we will touch on more in part three of this training, satellite data is able to assess the kind of observations of the Earth from orbit are good at determining some parameters, but often what we're looking for with landslides is a factor that correlates with the physical of understanding of what's happening underneath the surface.

Speaker 2: So let's say we think about the failure plane of a landslide. We want to know something that on the surface that might correlate with the characteristics of that failure plane. So that might be to do with the vegetation or it might be to do with the soil type that we can observe at the top.

Speaker 3: Field-based studies and local studies.

Speaker 2: May be much better at getting those kind of like local, that physics-based understanding. That's something that we really, regional, for more specific physics studies, that would be kind of critical. We call that geoengineering studies.

Speaker 3: So, yeah, thank you.

Speaker 1: Okay, terrific. Question number three. Are the effects of landslides the same across different countries? Are the examples that you provided also applicable to Bangladesh, particularly in the Chittagong Hill Tracts?

Speaker 3: So again, a great question.

Speaker 2: So the scientific understanding and the physics relating to landslides applies everywhere. I think what we've talked about is certainly applicable across the world. But obviously, there are certain factors that differ across, depending on where you are in the world, that will strongly influence.

Speaker 3: The way in which landslides occur. So local lithology, the vegetation.

Speaker 2: One thing that particularly does correspond to maybe differences between different countries are things like land management practices and the way in which people use their their hillslopes, for example. You asked specifically about the effects of landslides, and so that certainly is something that differs across different countries, ultimately because different countries have different ways of approaching emergency management, mitigation practices, and that means that the outcome of landslides may differ depending on where you.

Speaker 3: Are, particularly on the population impact.

Speaker 2: Specifically thinking about the Chittagong Hill Tracts, I know there's been a number of studies looking at landslides in that region. The NASA's landslide team itself is not supported specifically in that specific part of Bangladesh. We have looked at landslide assessment closer to the southern border near the Kutub Belong refugee camps.

Speaker 3: So there's been elements looking at that where we've been able to.

Speaker 2: Apply satellite data for landslide observations and susceptibility analysis. So yeah, if you're interested in that, the resources are available on the NASA landslides website, which I'll.

Speaker 3: Great.

Speaker 1: Thanks, Rob. Moving on, question number four. Is piping considered a type of landslide, or is it a distinct erosion process? How do the mechanisms of piping and landslides differ, and are there cases where both processes interact?

Speaker 2: So I believe what's being asked for here is piping, which is a type of erosional process that I understand to mean so the way in which when water flows through earthen dams, and specifically dams that are made of, yeah, dams that sort of, for reservoirs, water that filters through that dam carries soil particles away from that and builds these pipes through the dam itself. That would not be specifically considered a landslide, but as I stated here, theoretically, destabilization of hill slopes due to that kind of internal erosion may influence overall slope stability and ultimately lead to landsliding if that changes the frictional structure within a hill slope.

Speaker 2: So that's certainly a relevant consideration if there is a process of internal water erosion. I will say that anecdotally in field studies that I've been involved with, oftentimes where landslides have occurred, and you look at the large landslide deposit, you can see that there is water filtering through and that begins to sort of have this piping process maybe potentially after a landslide because there's this new in fresh material. And as a result, you may have continued destabilization of that landslide deposit.

Speaker 2: So again, leading to kind of further slope stability issues.

Speaker 1: Okay, terrific question number five. Is radar useful for looking at the stability of rubble piles?

Speaker 2: I think this is a really interesting question because you're asking about rubble piles, which typically are kind of human generated sort of hill slope. And the stability of those is very important for a lot of mine infrastructure and thinking about mine failures. Again, the underlying science and the physics behind how things fail is relevant to these same questions. Radar has been used to assess slow-moving landslides. So thinking about how very slow-moving landslides can change over millimeters to centimeters per year.

Speaker 2: And the same approach is called interferometric SAR analysis.

Speaker 3: Excuse me.

Speaker 2: I'm not personally aware of any specific analysis focused on rubble piles, although I'll make sure to research that before finalizing this question.

Speaker 3: In the answer.

Speaker 2: But I would say that the applicability of inter-processes to that kind of slow movement of these kind of potential hill slopes, or in this case, rubble slopes, may be relevant. So I think there's certainly an avenue to explore that. But as I say, I'm not personally aware of any research that has addressed that.

Speaker 1: OK, terrific. Question number six. Are drones with multispectral sensors better at detecting landslides at small scale instead of using Landsat and Sentinel data, the coarse resolution data?

Speaker 2: Yeah, so this is something I'd like to touch on again in the parts two and three of this training. Drones certainly have a very important role to play in field studies of landslides. We typically refer to these as unmanned aerial systems, or UASs, and they've been widely used in recent years to assess landslides. you may be able to use a number of different approaches to produce a landslide assessment, including three-dimensional structure from motion, which is what you take multiple images from around a sort of a location to build a 3D picture of a landslide, and then you can see how that might change over time.

Speaker 2: Critically, it's important to note that, yeah, this data is certainly higher resolution in most cases than the satellite data from Landsat and Sentinel, but to incorporate it in a kind of broader analysis, it is important to make sure that that is geo-referenced. That's typically done widely, but That is, again, an important prerequisite to the use of that data more generally.

Speaker 1: Wonderful. Question number seven. Typically for a risk assessment, the most difficult part to get is the post-event estimate of damage. It seems like we can derive hazard and exposure layers from NASA's satellite products or options like OpenStreetMap. Does NASA have an event reports database publicly available for landslides, How do NASA folks estimate social slash infrastructure vulnerability?

Speaker 3: This is a really fantastic question. I'm glad you teed me up for this.

Speaker 2: So I was halfway through writing an answer to this after you jumped into the Q & A, so I didn't want to kind.

Speaker 3: Of do too much on this.

Speaker 2: But let's say that, first of all, NASA has collected a large amount of information regarding the events of landslides around the world, both individual specific reported data as well as larger landslide inventories from major landslide triggering events, Typhoons or earthquakes. That data can be found at this link, so the NASA's landslides data page. We refer to this as our cooperative online open landslide repository, or COOLA. One of the exciting things about that data set is that it is a citizen science-based inventory, which means that we can incorporate inputs from citizen scientists around the world to provide their information.

Speaker 2: If they've seen a landslide around them in their vicinity, they can provide that information to us. And that helps improve our data whilst also kind of building a better inventory. We can then use that to improve our models as well. So that's something to note. I'll make sure that's listed in this answer. In terms of social and infrastructure vulnerability, that remains a major challenge. I think one of the critical elements, you know, I pointed out that the infrastructure also sort of social vulnerability incorporates things like healthcare outcomes, and certainly we can't observe those from orbit.

Speaker 2: So we're often relying on data from other sources that includes census data on population demographics to sort of begin to estimate some of those elements. We, the disasters program at NASA has just begun to support several projects looking at those, and those were selected for funding within the last several months. And so we've just launched some projects that are looking at specific in the face of different types of disasters persist. And so I encourage you to kind of look out for some of the research findings that come out of those things.

Speaker 2: And some of that information, I'll make sure that those links are provided in the answer to this question overall. But yeah, so that's certainly a really interesting.

Speaker 3: Question.

Speaker 2: And I think understanding more general models for social and infrastructure vulnerability around the world is a key topic of research.

Speaker 1: Terrific. Question number eight. Cloud cover is a constraint. Can we use precipitation datasets to map landslide susceptibility?

Speaker 2: Again, a fantastic question that we can talk a lot more about in parts two and three of this training. I think there are a couple of things here that are important to note. First is that cloud cover is a constraint for some landslide types, sorry, for some datasets. Optical data obviously cannot see through clouds. But as mentioned, satellite radar data can see through cloud. And as a result, we can or landslide changes using SAR data in some cases. And so that's one note there. Precipitation data certainly has been used as part of landslide mapping.

Speaker 2: And so NASA's team has developed a landslide accessibility and hazard model that incorporates satellite rainfall data. I'm going to happily talk about that in part three of this training. Essentially, what we're looking for is areas where there's intense precipitation that might be associated with landslides And we've done that both with real-time rainfall data as well as forecast rainfall data. And that has produced relatively good.

Speaker 3: Results in many cases. And so precipitation data is important.

Speaker 2: I've also been involved in research that has explored different types of precipitation data from different satellite sources. And just to say that there are strengths and weaknesses of each of those, kind of different varieties of satellite rainfall data. And many ground-based assessments will also use radar data. for rainfall, typically because that is more reliable and robust and produces better sort of short-term intervals, short-term estimates of rainfall.

Speaker 3: Than satellite data.

Speaker 2: I mean, Just to give a sense, satellite rainfall data at best can produce an instrument every 30 minutes, but if your rainfall is very, very short-term intense rainfall, you want to capture that. And so again, local studies using radar data may be important too. So yeah, precipitation, very important.

Speaker 1: Okay, wonderful, Rob. Question number nine. Could you speak more to the use of failure slope landscape slope probability ratio as shown on slide 24? Could you provide a reference for the frequency graphs shown?

Speaker 2: So I am going to have to go find where slide 24 is in these slides.

Speaker 3: Give me one minute. I'm just going to pull up the slides quickly. Apologies for that.

Speaker 2: If you wouldn't mind if we jump back to that question in a second, I'll have that.

Speaker 1: No problem. We will move on to question 10, and we'll come back to question 9. How can we use earthquakes to map landslide susceptibility? Is there any data for earthquakes apart from earthquake zones When we work on a city scale, the whole city might fall in one earthquake zone. How can we show an earthquake's spatiality on a city scale?

Speaker 2: So that is a really tremendous question. So one of the things that we think about a lot is satellite data for rainfall, for example, triggering landslides is something we've done a lot of work at at NASA. But earthquakes is something that we often look to other agencies to support. And the United States Geological Survey has some fantastic information on earthquakes, they produce a model called ground failure, which is, I would say, the most widely respected model that.

Speaker 3: Uses an earthquake. The earthquake process.

Speaker 2: And the potential amount of shaking is used to predict where landslides are likely to occur as a result of that. And that model, I will make sure, again, is linked in the answers to this question. I think your also question was asking about the spatial sort of extent of earthquakes.

Speaker 3: Within a city, for example.

Speaker 2: And again, I think that there are the seismic networks that are produced on ground-based data is widely used to assess the kind of.

Speaker 3: Like short-term shaking. NASA data has been used, again, the.

Speaker 2: Disasters Programme has supported research to look at defamation as a result of earthquakes more locally within cities.

Speaker 3: Again, I'll make sure that is linked in the study and, sorry, in the answer to this question.

Speaker 2: Typically, that looks like after an earthquake, we assess the changes before and after and show the kind of relative defamation. And that's something that we can do much more specifically within a city example And looking at, for example, building damage is something that has also been done using satellite data supported by NASA's Disasters Program. So yeah, I'll make sure those connections are made and we'll have those links in the chat. So I'm happy to go back to question nine.

Speaker 3: Now, so.

Speaker 1: Okay, great. If you wanted to go to question nine now, we can certainly.

Speaker 3: Yeah, yeah.

Speaker 2: I have this one in front of me again, so I close them before answering the questions. So the question here was about slide 24. So those frequency graphs that are shown actually come.

Speaker 3: From a study that I led a couple of years ago.

Speaker 2: More specifically, the detail of that, we'll talk a bit more about that in part three of this training, but I will ensure that a link to that open access study is provided in answering this question.

Speaker 3: Yeah, so happy to provide that information. Okay, wonderful.

Speaker 1: And we will now go to jump to question 11. In humid subtropical climates like Zimbabwe, there are pronounced wet and dry seasons. NDVI, the normalized difference vegetation index values, declined drastically in the dry season, depicting similar patterns as landslides, such as vegetation loss. How can we differentiate between both in the NDVI time series?

Speaker 3: So really, that's another excellent question.

Speaker 2: So what you're pointing out is that typically we have used changes in vegetation very extensively for landslide assessment. And that works very well in areas which are green for.

Speaker 3: Much of the year.

Speaker 2: And so we see little landslides, for example, we can pick those out. But in areas, yeah, when the vegetation dies off in dry season, it can be much more challenging to pick out landslides.

Speaker 3: I think one of the easier ways to answer this.

Speaker 2: Is that for smaller, shallow landslides, if the soil type is very difficult to see from NDVI change, then it.

Speaker 3: May be challenging to do that with optical data and NDVI data. If the.

Speaker 2: The land side is larger and excavates down to the bedrock, for example, that may still be visible, considering that the bedrock and the surface characteristics typically still differ enough in NDVI mapping. However, more generally, I would say this is a point where we would begin to look at SAR data, additional data sources like hyperspectral.

Speaker 3: Data, for example, to begin to think about how we can map these sources.

Speaker 2: The other final point I'll make, and we'll talk about this in part two of this training, is that the human eye is not always perfect at assessing where landslides have occurred. And automated methods have been used to do just this, and to pick out small but noticeable changes that we might otherwise miss with, for example, NDVI change. And so my colleague, Bukhara Mattia, who will present part of this presentation, will talk a bit about how we've used automated methods in this example.

Speaker 3: And he's beginning to have some success in research, too. to assess kind of dry regions and how that might change. So thank you. Okay, wonderful.

Speaker 1: Question number 12. I'm curious if there's an onus on international governments to mitigate risks if they are given risk data from another country.

Speaker 3: That's a great question.

Speaker 2: I am certainly not an international policymaker, so I don't have a specific answer to this question other than I think that NASA data has been used around the world to support disaster response information. And I, yeah, I think that's, I'm probably not the right person to ask specifically about international policy and essential governments.

Speaker 3: But yeah, I would be curious to know the answer to that question as well.

Speaker 1: Okay, great. Question number 13. I heard about NASA's disaster response efforts related to landslides in Rio de Janeiro before. Is there any repository where I can learn more about the methodology and how to assess the LASA model at a more local or regional scale?

Speaker 3: It's a great question.

Speaker 2: So all the NASA's landslides research that has been conducted is updated on our web page. So the link is provided above, but I'll provide it again here in the response to this question, which includes the work that has led to LASA and other systems.

Speaker 3: With respect to Rio, that.

Speaker 2: Research paper that describes that work is kind of in publication or like in the process of being published right now. So we don't have a specific link to that, but please reach out to me specifically. My email is available in the contact information and I'm happy to kind of like provide more detail on how the model works and how we can connect you with that specific data set. I'd love to be able to point to a general source, but at this stage, we're still pre-publication.

Speaker 3: Okay, wonderful.

Speaker 1: Question number 14. We can assess which street can be affected by the landslide. And from what I understand, to assess it, we also need the elevation or the slope gradient. Since digital elevation models, DEM data, that is freely available, especially in my country, is about eight meters, and streets could have a higher spatial resolution than that, such as have a smaller width than the slope gradient, how can we minimize the bias?

Speaker 3: Yeah.

Speaker 2: So I think this is when it becomes important to assess the the data that's available to you and which the data is not available to you. So if you want to answer questions about specific road segments, for example, and the data set that you have available to look at GEMs is not of the same, is of a coarser resolution than the typical size of landslide, you may be,

Speaker 2: this is not a kind of like scientific answer, but it may be pushing the data too much to like answer those questions. Again, you may be producing false results that would be kind of not supported by that analysis. So again, I think typically landslides occurring below the resolution of the GEM, for example, may not be appropriate to assess at that scale.

Speaker 3: So I think.

Speaker 2: Those very small landslides, typically smaller than eight meters, for example, would be typically best assessed using local studies. And again, I think like going to do geoengineering of those specific road segments that might be at risk would be more productive than I think trying to build models based upon course of resolution information.

Speaker 1: Okay, wonderful. Question number 15. What's the progress made by NASA in terms of understanding landslide mechanics and prediction using physics-informed machine learning?

Speaker 3: That is a tremendous question.

Speaker 2: To specifically provide some context, the person asking this question has talked about physics-informed machine learning. So that's a really critical distinction between what we typically do with satellite data to assess landslides at NASA, which has been statistical approaches. Essentially, we're looking from orbit at assessing the stability of slopes based on prophecies and correlations between different variables and parameters that lead to statistical relationships between the landslide occurrence and the parameters variables observed.

Speaker 2: The physics-informed machine learning would assess the specific, let's say, the friction structure, the gravitational weight of different slopes. That is something that we've not made a great deal of progress on.

Speaker 3: We have.

Speaker 2: However, there has been a disastrous program of supported research that is focused to use both satellite and ground-based data in combination to assess some of that element.

Speaker 3: And I will happily point you to the publications associated with that research.

Speaker 2: And so we've done And some work on that, but at global scale, we're still not at that point. I think, yeah, I'm going to, in the answer to this question, I'm going to provide a specific link to some studies, make sure that that's out there. But I think the best way to answer this, I don't know, I'm going tendentially around, is where we've made progress here has been incorporating both drone-based and UAS-based information alongside satellite-based data.

Speaker 3: And so that's really required.

Speaker 2: Us to kind of go into the field to do that kind of work at more high resolution to kind of inform the physics analysis of this.

Speaker 1: Great. And I think we can get through one last question, and then we'll have to start wrapping up because I do want to respect the time. But maybe we can get through.

Speaker 3: This one last one.

Speaker 1: Question 16. Is tidy cultivation on the mountain slope considered a cause of landslides? This was considered a chance in the landslide at Malin Pune.

Speaker 3: It's a good question.

Speaker 2: I am not familiar with that specific landslide event. And I do not want to speculate on potential land use that may be either involved or not.

Speaker 3: I think what I.

Speaker 2: Will do or suggest is I will have to go and look at some studies associated with that. But I think, again, I prefer not to speculate on a specific land use type and how that might involve. Certainly paddy cultivation is widespread across large parts.

Speaker 3: Of the world.

Speaker 2: And I'm not sure the extent to which that, I don't actually, yeah, I'll just say, I don't know for sure.

Speaker 3: So I'm going to have to go back and look some of that up. Great.

Speaker 1: Well, to respect the time, we are at the middle of the hour, which is the time that we agreed to be stopping. But as we wrap up, Rob, I did just want to know if you had any maybe closing thoughts or comments kind of looking ahead. And for all the participants that have been able to join us today, just wanted to maybe have any concluding thoughts from you. So over to you. Thanks.

Speaker 3: Sean, thank you so much. And thank you to everyone.

Speaker 2: And to those who've asked questions, thank you for such interesting and thoughtful questions. I think many of these have pointed out that we have more than the past two and part three that will answer help answer some of these thoughts. But yeah, I'm being very excited to present to you all today. And I look forward to further discussion. And certainly to if you want to reach out to me and ask further questions, please do so.

Speaker 3: So thank you all. Great.

Speaker 1: And Rob, thank you. We just knocked it out of the park today. Thank you so much for answering all these terrific questions and such a wonderful presentation. Certainly looking forward to the next one in two days. I did want to also thank the RSET team. You might not be seeing them or hearing them, but they are working very hard in the background to make sure that this training goes off without a hitch. That is Natasha Johnson-Griffin, Brock Blevins, Selwyn Hudson-Odoi, and Sarah Cutshall. So a big thanks to the RSET team.

Speaker 1: And lastly, I just wanted to thank all the participants, wherever you're joining from. We really appreciate. We hopefully you're getting a lot out of this. And I'm just going to say now there will be a survey at the end of this webinar series. So however you're using these data sets or hope to intend to use these data, we do hope that you will fill out that survey when it comes to your inbox so that we can kind of learn how you hope to be using some of the skills that you're learning from this training.

Speaker 1: But I just want to end everything now by saying thank you all. And we look forward to seeing you again in two days for part two of the training.

Speaker 3: Thank you.

The Story