 Welcome, everyone. This is Marty McCann. I'm the chairman of the committee on geotechnical and geologic engineering. We are a standing committee of the National Academy of Engineering and Sciences, and we are under the board of Earth Science Resources. And this is one of our regular webinars on geotechnical and geologic engineering topics. So I want to welcome you all today and thank you for participating. Before we get started, I wanted to first thank our director, Sam Magsino. Her email address is on a number of the slides that you've seen rotating here. Sam is our director, and she can be your main point of contact for the committee about future webinars, getting on mailing lists and things of that sort. And also wanted to thank Remy Shepada, who is our program assistant who makes all the technology work for us. So Sam and Remy, thank you both for your assistance in putting this webinar together. And with that, I am going to turn it over to Scott Anderson, who is a member of the CAGA committee, and will be moderating the session today, introducing our speaker and facilitating the Q&A session after our talk here. So Scott, with that, I'll turn it over to you. Thanks very much, Marty. And thanks everyone for joining this afternoon. The screen that's showing right now talks about our in-person meeting coming up at the end of the month, but of course you're here today to listen to a webinar titled Regional Scale Landslide Risk Assessment, Methodology, and Application. And that's going to be delivered by Dr. Joe Wortman. And just as a little background for Dr. Wortman, he directs the Natural Hazards Reconnaissance Facility, which is known as the RAPID, at the University of Washington, where he's the HR Berg Professor of Civil and Environmental Engineering. He's a former editor of the ASCE Journal on Geotechnical and Environmental Engineering. He's the author of over 100 professional articles on geologic hazards. And a special note, I think, is that Joe is very active, more so than many of us anyhow, in publishing essays and op-eds, and they've appeared in places like the New York Times and the Seattle Times and other mainstream media venues. He's the recipient of research honors, many of them, and including most recently the Geologic Society of America's Burwell Award in Engineering Geology, which I was fortunate enough to be working alongside Joe. So, with that, I want to introduce Joe and turn it over to him. Okay, thank you, Scott. And thank you, Sam, Scott and Marty, for coordinating and facilitating the webinar this afternoon. And to all of you who are joining remotely, it's my intuition to say thank you for being here. But since I'm doing this from a small audio-visual room in Washington, D.C., that seems like maybe it's not appropriate. So I'll say thank you for being there wherever you are this afternoon. I'm going to be talking about some work that we've done over the last several years on regional scale landslide risk assessment. And I'm going to briefly review the methodology, and then I'm going to step through the application in a very systematic way to offer an example. And that example is going to be for the country of Lebanon. And I'll say a little bit more about why that's the focus of the study I'm going to be presenting today. I think all of us are aware of the rapid advances and really what GIS or geographic information systems has enabled for the geographic sciences community in general. But I think it's worth pausing to just think about where we are today and what this has enabled and where this really leaves us and what opportunities it presents to us in the geologic hazards community. So I see us today in the age of ubiquitous digital mapping data. And you only have to reach into your pocket and access your smartphone to probably find one of several applications that are using mapping data, Google Maps, Apple Maps, Yelp, amongst many other mapping applications. And increasingly that data is becoming not only open, but at least freely available. And this whole notion of scale that we've that we've had for a number of years, I think is also changing scale is whatever the resolution you decide to zoom in or pinch into your screen. And I've illustrated that my middle diagram here where we're visualizing LiDAR data at two different scales. And that resolution, the ability to zoom in has been enabled not just by the remote sensing technologies that have come around in the last 10 years, but also some pretty significant advancements on the platforms that have hosted those remote sensing technologies. And I'm just going to offer one example. I have a photo of Jake Daphne, who is the operation specialist for the rapid facility, and he's piloting a high resolution LiDAR system on a on a drone. And this allowed us to develop bare earth LiDAR of an active landslide site in Oregon, just a couple days and something like that, of course, would have been really unthinkable just a decade ago. So I think we're seeing some really important advancements there. But that's not really the motivation for this work. Those are the tools. I think a lot of the motivation here stems from experiences with the 2014 Oso landslide, which occurred in Washington state and very tragically killed 43 people. And if you look at the LiDAR image that I'm projecting here, you can see that the the side slopes of the Oso Valley are flanked with with with large landslides, many of those long run out landslides. And just a cursory depiction of the kind of risk that was present at the time of the Oso landslide is shown in the diagram on the right. And perhaps unsurprisingly, we see that that this is above what has largely been seen as acceptable using some international standards for for loss of life. And so Oso was a pretty risky place to be. And the issue is that a lot of that information doesn't make it into the way we traditionally map landslide hazards. And so just below, I've shown the landslide hazard map geologic hazard map actually that was available for the region at the time of the Oso landslide. And what you can see is that the the dark box depicts the location of the steelhead Haven community that was inundated by that landslide. And you can see that falls outside of the landslide hazard zone, which is depicted with the red cross hatching. And while there were zoning ordinances that required setbacks from those landslide hazard zones, this really doesn't convey the the risk to those who lived in Oso at the time of that tragedy. And that's a problem that's not just limited to to Washington state. If we look at another well known landslide site in the United States, the community of La Conchita in California, if you make reference to the available landslide hazard maps for that community on the lower left, I've shown the California Department of Geology co seismic landslide hazard zones. And you can see that the community falls outside of that zone. The zone is depicted in blue. And so we're not capturing the kind of run out or inundation behavior that often occurs in these kinds of events. And to the right is a mapping that was put together after the Thomas fire of debris flow hazard maps. And this comes with the stipulations that does not predict downstream impacts and potential debris flow runout paths. And so again, that's very much the standard. And one of the things we were aiming at in this project was to provide a better depiction of the kind of risk to to people but also to potential loss of capital assets and infrastructure and so forth. So what I'm going to be talking about today is part of a larger effort that began in 2016. We've published a paper in engineering geology on the multimodal method of landslide hazard assessment. I want to just briefly introduce the research team. And I have a picture in the upper right as a stop during some of our fieldwork in Lebanon. That's Alex Grant in the in the leftmost position in the photo. Alex is a former PhD student who is now with USGS and behind Alex is grace up a job. Sitting across from grace is Will Pollock. You'll hear a lot about his work this afternoon. And Will is sitting next to Maria and track and I want to acknowledge two people who are also been involved in the work we're pictured here. Angela said is a researcher was formerly at the Lebanese American University and Chris Massey is with GNS science in New Zealand and he's been a close collaborator over the last five or six years. So what I'm going to focus on today is a is encircled in red is a our development of a platform for landslide risk assessment. And I'm going to make reference to version one here and at the conclusion of the presentation I'll talk about version two and what's to come. But this built upon the multimodal method in a couple significant ways we've made modifications to the geotechnical models we didn't we've included a precipitation induced landslide module. We have a module for modeling run out. And ultimately the product of this is not hazard but it's risk assessment and I want to say that one central theme that has run through our work is the idea of low cost high resolution mapping. And before leaving the slide I want to also say that this work has been supported by the National Science Foundation and and none of this would happen without that generous financial support and we also receive supplemental support from the United States Agency for International Development. So this work has a lot of relevance to policy there are currently two bills that are pending in Congress once in the Senate and once in the house I'm just going to briefly review the one in the house and this was been led by Susan Del Bene. It's come out as a couple different iterations but I want to note that includes provisions to develop maintain a develop and maintain a publicly accessible national landslide hazard and risk inventory and it also includes a provision for establishing the three depth or 3D elevation program. And the status of this bill is that it's currently out of committee and that's significant only about one in four bills make it out of committee and the current prognosis at least according to scope this labs which is a legislation analytics firm is that this has a 50% chance of being enacted and so if this moves forward it will really be quite significant for us as a landslide community and that bill has a a lot of really important stuff that captures the benefits of regional scale hazard mapping and risk assessments and there have been reports that have demonstrated that mapping is a very cost effective investment it obviously supports land use planning it serves as a transparent basis for making important decisions. It allows mitigation actions to be prioritized and I think really one of the exemplars is the USGS seismic hazard mapping program and what I've shown in the lower right is the way that information from from that mapping has now been converted to risk cross sign has developed an application called Trembler which allows you on a on a house specific basis to look at your seismic risk throughout the United States. I think that mapping also plays an important role in enabling citizens to make informed decisions it enables by making information openly available that enables land and housing markets to operate efficiently and then really I think unless people understand the risks they face it's very difficult to inspire action for risk mitigation because it often involves a very significant cost investment. So I want to say a couple of basic words about the the methodology that we've developed and the diagram on the left shows the traditional infinite slope model that has been used for regional scale landslide studies for at least the last decade. This is an excerpt of a paper from my colleague Scott Miles and one of the issues with this kind of modeling is that it doesn't capture the wide range of modes that we see in the field and so this is a photo that I took after the Kaikoura earthquake and what it shows is just in a single scene three distinct modes of cosysmic landsliding we see rock slope failures shown in yellow is a rotational slump we also see shallow disrupted failures using the terminology of Dave Kiefer the USGS and then off to the left we see lateral spreading and so what we aim to do when we originally developed a multimodal assessment was to capture these specific modes and then implement geotechnical models for each of these across the terrain and I'll show you an example of how that gets implemented. It's quite popular today to use statistically based landslide hazard models and I want to say a couple words about why we've opted not to do that for the kind of work I'm going to be presenting today we have done some work in this area around 2010 and 2011 looking at landslides from the Northridge earthquake and the idea with this is based on observations what's the combination of geology morphology and forcing in this particular case ground motion resulted in the initiation of landslides and the idea is that we formulate and train statistical models with landslide inventories and of course the key advantage to this approach and it's really quite powerful is that there is no need to quantify geotechnical properties over large regions that inherently comes out in the landslide inventory but there were some issues for us in moving forward and trying to implement this particularly in a place like Lebedon one is is that for Lebedon we don't have a pre-existing landslide inventory the model is only as strong as the landslide inventory training set and so it's very much a function of the quality of those inventories that are available I think at an intellectual level it's not fully satisfying because it doesn't tell us why it doesn't allow us to fully disaggregate really understand the results and there's questions about the transferability of these kinds of models to other terrains other regions and geologic settings and so forth I think it's also important to recommend remember that the training inventory represents the predictor conditions at the time of the forcing events so if you have a rainstorm proceeding a an earthquake the the outcomes are going to be much different than if it was a dry season for example and again that's inherently reflected in the landslide inventory and then finally I want to say that it cannot at least yet and I've put yet italics here effectively capture complex cascading type impacts and risk assessments and and I think I've noted yet because I think that's obviously going to change as we continue to collect more and more data after landslide disasters but I don't think we're there right now and so we moved forward with a physically based and empirically based modeling approach the landslide risk assessment equation is is relatively straightforward it's long it's a little bit easy to describe on a site specific basis before I talk about a regional scale implementation but I think many people who have tuned in have seen this before this has been around and used since the early 2000s the upper equation is for risk of loss of life and this can also be expressed in terms of loss of capital assets but the risk of loss of life is essentially a series of conditional probabilities given the probability that a landslide occurs what is the probability that it has the spatial reach to to reach for example in the example below a house and given that it's reached the house what's the probability that someone's home and given that they've been impacted what is their vulnerability what's the likelihood that they would be killed by that impact and then that gets multiplied by the number of people who are exposed and ultimately this gets integrated over a whole series of scenario events with different return periods and again we can express that same kind of equation just in terms of financial losses instead of loss of life which is what I've done below so this is the landslide risk methodology that I'm going to step through this afternoon and involves three distinct steps the first is regional scale hazard assessment using the multimodal method and again there's a two-step procedure to this the first is that we assess the susceptibility to each landslide mode based on topography in the sense reflecting the terrain and the second is that we implement a mode specific geotechnical models to assess that hazard the second part is we estimate run out when that's applicable and we use an empirically based approach for that and then the third part is the risk assessment and that's where we estimate human and or capital losses and then we repeat that process for other return periods I want to say that all of the equations and all of the background information and what I'm going to be talking about has has appeared in print the details of the multimodal method were described in engineering geology we've recently published just about a month ago a series of companion papers in Elsevier journals on all of the methodology and background equations for the risk platform and that is referenced in Pollock et al 2019 and I'll provide those references at the end of the presentation kind of information or data that's needed to feed this risk assessment is pretty basic information for the first step we need information on terrain geology we use satellite imagery as well as I'll describe for our case application in Lebanon we also need the intensity of the forcing that is the intensity of the the storm or the the ground motion intensity if we're considering both of those four scenes together for the second step for the runout estimation we need data on terrain and we gather that from the first step the the digital elevation model and then finally for the risk assessment and really this is very much a function of where we're working and where we're implementing the models but we typically use a combination of census data open mapping data and government databases and in some cases data from NGOs so I'm going to talk a bit about Lebanon and perhaps you might wonder why we've we've started this work with an application in Lebanon and of course it's not going to be a surprise to anyone tuning in today that there has been a major humanitarian crisis that's occurring in that part of the world and beginning in 2011 with the civil crisis that erupted in Syria and that's led to a massive influx of refugees into Lebanon which has raised the population of the country by 40 percent over just several years and so and I'll say that a majority of the population who are refugees are children they're people who are aged from from zero to 15 and so again this is I think one of the most significant humanitarian crises of our generation obviously that population expansion has occurred without the benefit of any kind of formal land use planning just because of the timeframes that are available and that's led to a number of pressing questions about Lebanon's refugee placement policies and so that ultimately motivated that work we were also trying to answer some some very basic and I think very quite multidisciplinary questions along the lines of how the regional crises humanitarian disasters and policy making affect a country's geologic risk profile and then finally from a technical sense I would say that this was a really challenging problem from a couple different aspects that the first is we were working in a data limited setting Lebanon is a middle income nation it doesn't have the same kind of data resources that we have for example in the United States and secondly it's a highly active landscape that's affected both by precipitation and seismicity so just briefly a little bit about the geology and landslides of Lebanon the geology of the country is dominated by two mountain ranges the lebedon range which is shown here to to the west and then to the east is the anti-lebedon range and that's separated by the becca valley the photos on the right show some of the photographs we took during our our two years of field work in the in the country and you can see a whole range of different styles of landsliding a number of debris flow hazards there are rotational slumps and there's also a number of rock slope failures that we observed in in our travels across the country specific data sets that we used for the analyses i'm going to step through or as follows we we adopted a national scale geology map and and use that to estimate the geotechnical properties of the underlying materials we accessed a 15 meter dem that was provided by the government of lebedon and then we use landstatt imagery to estimate root cohesion by way of nvdi and you'll see the way that that factors in just a moment we also use national precipitation maps to estimate storm intensity and then we utilized a probabilistic seismic hazard study that was done for the nation and that helped us estimate ground shaking across the country and then finally we use population data from the national census and also from NGOs and the NGO focus was largely on the refugee crisis and capturing the the changing populations across the nation over the last five years so i'm going to zoom in into one particular region and i'm going to step through this in some detail the the area that i've shown here is just about 25 square kilometers and so this represents obviously a tiny tiny fraction of the overall area of lebedon the area of lebedon is in excess of 10 000 square kilometers so this is a far less than than one percent of our study area but it allows us to take a look at what's really going on under the hood with this landslide risk assessment platform and and i'm going to show you examples for two different scenarios i'm going to show a 50-year storm event and then about a thousand-year seismic event and then ultimately i'm going to show risk results that have been integrated across different return periods and so this is the area of hamat and if you look at a satellite image you can see that this is a populated region that also contains a number of infrastructure systems including the the major north-south highway across the country so the first step in our hazard assessment using the multimodal method is to divide the terrain up into different landslide susceptibilities and we do that by slope we take slopes that are in the range of 15 to 50 percent of 50 degrees and judge those as being prone or susceptible to shallow planar slides slopes that are between 15 and 35 degrees are prone to rotational coherent slumps in soil and rock materials and then slopes that are in excess or steeper than 35 degrees are prone to rockfall we don't assess any ground that is flatter than 15 degrees and that's for reasons of computational efficiency that we don't have to be running our models across areas that are unlikely to experience any kind of landsliding now obviously there's some overlap in these and in cases where for example we have a slope that's 25 degrees we would run several analyses for both shallow planar slides and for rotational slides and that with the lowest factor of safety is ultimately what gets what gets mapped and so this is the depiction of those different landslide terrains the the green shows the location of shallow soil slides and slumps the the orange or I guess that's a yellow shows the location of rockfall I'm sorry of of disrupted failures which are basically very shallow failures and then rockfall is shown in brown and so I'm going to step through those individual modes first for our scenario storm events and I'm not going to spend too much time on the equations I would get bogged down with those and those have been described completely in the journal papers I made reference to earlier so I'm going to just refer to those for right now and just give you a little bit of an overview but to say that we assess slumps across the region using a 3D spherical surface whose radius of failure is a function of local relief and that's measured within a moving window and obviously the issue here is we're trying to take a routine slope stability analysis and upscale this in a for a 3D world in GIS for hazard assessment of debris flow source areas we've adopted the shell stab model which couples hydrologic and limit equilibrium slope stability models to compute critical daily rainfall needed to trigger shallow soil failure and we add a modest amount of root cohesion and that comes by way of nvdi and so what I've shown here in blue are the locations of debris flow source areas from our analysis and then of course there's a second phase here which is if we have the potential for a debris flow we therefore have the potential for run out and we don't have run out for coherence slumps they usually don't run out they will obviously damage infrastructure and property that are on those but they don't typically become long run out events and so we don't do a run out analysis for those but that is the second stage that we use for debris flows and so we've implemented a routine that's based on flow lines from a digital elevation model and we limit the path to 750 meters and that's been based on a statistical analysis of debris flows in the region of Lebedon we also assessed rock slope failure or rockfall hazards and we can incorporate the destabilizing effect of water and a discontinuity we model this with a Coleman wedge like function and those susceptible locations are shown just at the top of the diagram here in in green and so you can see this doesn't this doesn't cover a lot of the area but there are slope rock slope hazards that exist and those are associated with a run out hazard as well so we use a view shed analysis that basically allows us to let these run out in an apron like fashion below the source area so this covers the hazards that are related to precipitation event we have also done this for co-size mcland slide hazards and the routines a little bit different the models are a little bit different the first four disrupted or shallow depth slides is that we do an infinite slope analysis and then we estimate co-size mc displacements using a regression that's based on a new mark sliding block model and there are a number of these are available we've adopted both one from Randy Gibson as well as Ellen Rathchie that have been published over the last decade or so and that gives us estimates of co-size mc displacement we then take those estimates and we established a threshold of five centimeters to identify source areas and so the idea is if you have less than five centimeters of displacement we don't see those as being significant enough to really unleash a full co-size mc disrupted landslides so these are the are the source areas that are have displacements in excess of five centimeters and we take that same kind of thinking and we extend it to rock slope failures and so we estimate co-size mc displacements we set another five centimeter threshold to identify those source areas and of course these events are associated with run out and so now this is the source area along with the run out for the co-size mc hazard so if i put all of those modes together for the two different scenario events i've mentioned these are the hazards so you can see the source areas and then where applicable the the run out paths that are associated with so that's the first and second step together and if you take a look with you know how these results translate to what you observe in the field this is a photo of the area that's encircled in purple and this shows the rocks fall zone that's at the crest of the slope this is the zone that's susceptible both the brief lows and disrupted slides depending on whether it's a precipitation trigger or co-size mc trigger and then this is the zone that's susceptible to slumps or rotational kinds of failures and you can see this kind of makes sense with the observed behavior of these slopes of what we see in the field the third step is the risk assessment and for this we use an inventory of populations that are at risk and there's two kinds of populations that we disaggregated in this study we looked at urbanized areas or built or developed areas that pre-existed before the syrian crisis and then we've also looked at locations of informal refugee settlements which are basically settlements that have not been urbanized and so there's two of those there's one that's very small over here and the other one is depicted in red but all of the areas that are shown in in dark gray are urbanized regions and we also have a database of infrastructure assets throughout this throughout the nature so if we combine that with the with the hazard map you can see that there's some overlap with the hazard in the in the upper part of the region and then you can see along this other coastal area there's also overlap between the hazard zones and the population and of course that occurs through some of the other areas as well but these are some of the real hot spots and so ultimately when we apply the risk assessment we estimate physical vulnerability of the urban population based on data from nations having building types that are very similar to those to the styles that exist in Lebanon and again there's a bit more in background of the kind of fragilities and models that we've adopted in the publications but when we put those together what this shows is a mapping of annualized loss of life and this is for the scenario of having different precipitation induced landslides and so you can see that this area for example is free of risk or at least the risk is it's getting mapped up here at the color scale that we've adopted when that situation changes a bit when we look at cosismic landslides you can see that there's a cosismic landslide risk that exists here that doesn't for precipitation and induced landslides and alternatively these modes go away so ultimately we put these together and these are the results compiled across the entire nation so now we've zoomed out to the 10,000 square kilometer scale and one of the things that's really a nice about this kind of approach is it allows us to disaggregate risk according to different regions or according to the entire nation so i'm just going to offer a couple insights that we've gathered just by taking this risk data apart and the first is is that we see that the brief flows and associate run out are responsible for about 93% of the overall landslide risk in the nation of lebanon majority of those losses are from frequent and widespread low intensity events and that's in contrast to the more significant but less frequent intense storms for example and other types of landslides have a less significant impact and that's largely because they don't have the same kind of run out footprint as the brief flows and so this is the data again compiled across the entire country we could do this as a heat map as well it gets depicted at the scale a little bit easier with a mapping of annualized loss of life proportional to the blue symbol and these different color symbols refer to the different modes of landsliding if you take this data you can in order to get a sense of what the societal risk is in lebanon is to get an idea of where this stands to other nations if we plot those in an fn characteristic diagram which shows the annualized frequency of different numbers of fatalities you can see that this data pretty much plots in its on par at least with other landslide pronations and so it's not extraordinary but it is quite high and somewhat in excess of what is often deemed to be acceptable by some international standards I'm not going to have time to talk about this today but we've stepped through the same exercise for estimating capital losses to buildings and to infrastructure systems and that's been reported in a separate article by Pollock et al 2019c and basically it's the same kind of approach but now we're looking at the value of these assets I'm going to go back to the loss of life data and then try to disaggregate this by time because I think we see a lot of interesting trends related to the refugee crisis but I think this also highlights the way that risk varies in time according to population and to land use and so if we look at these diagrams that are shown in the left side of the article this shows debris flow, risk, precipitation due to rockfall and so forth all of the different modes in terms of annualized loss of life across the five-year time frame and so the first thing you can see is that in blue is the data for the Lebanese population that was there before the Syrian refugee crisis and of course they hold the majority of the societal risk nearly because of the population is greater than the Syrian refugees the risk to encamped or informal encampments for Syrian refugees is highly variable through time and that's shown in the green green diagram here and that's because of their largely transient nature and it's also quite high because of the increased vulnerability of often having temporary shelters that offer less protection against debris flows and rockfall compared to for example more routine built structures and then finally there's an interesting drop in the risk here in 1916 that reflects movement from rebel allied Syrians during a period of a particularly intense conflict in the Arsail area and basically a movement of refugees outside of that area during that time of conflict who then later came back and you can look at some of the other diagrams and get similar information there's a heavy refugee influx that into a into a rockfall prone area and you can see that that's changing with time and again that's reflecting the transient nature of those flows and then if you take a look at the cosysmic disrupted landslides and cosysmic landslides in general you can see that generally the loss of life is quite low for these compared to debris flows into the rockfalls and so these represent a relatively or a much lower hazard and as I mentioned earlier 93% of the hazards contained up here so there's a lot of information here and I'd say that overall what we found is that there's a disproportionate 75% increase in landslide risk in Lebanon since the start of the Lebanon of the Syrian crisis and that has really everything to do with the population increase but also higher vulnerability and exposure to those who are in informal settlements so the question is is how valid is this data and can this how do we check this data and we had a natural experiment that occurred in January 2019 we had a tense storm I don't know what the return period has not been documented but we know that it triggered a number of landslides very close to Hamat in the area known as Chechem it injured three people and closed the highway and these are some press reports so this is our our pre-events model and this is for a 50-year storm event and so we really can't compare apples with apples here because we don't know what the return period is of this particular storm event but I think one thing that's encouraging is is that a lot of the modes that we had forecast in advance of this event in fact came to fruition and you see a lot of these kind of run-out landslides onto the highway and so forth so that gives us some confidence in the ability to forecast that hazard now when we compare the risk results the results are not quite as accurate and I'll say that we did a database search of recorded landslide fatalities in Lebanon between 1975 and 2015 using Arabic and English language nude services that revealed 146 fatalities that were conclusively caused by landslides and we believe that average is about four a year we know this is an incomplete inventory we know there's a strong recording bias in time with more recent events being recorded at greater resolution and more detail we also know that a lot of the injuries and fatalities to refugees have not yet been reported so it's likely to be an underestimation but nevertheless our projections are for close to 40 fatalities a year and that significantly exceeds the recorded data by an order of magnitude so why might that occur well one is obviously related to the under reporting of the landslide fatalities but there are some other reasons as well another is that our physically based models tend towards conservatism we have a 15 meter resolution that obscures small check dams and debris flow mitigation measures that otherwise don't get accounted in our run out analysis we don't account for short term variations in exposure we assume that people are there full time and we do not account for the ability of people to avoid impending or ongoing hazards that is to leave a dangerous region so what's next we have a second version of the platform that we've been testing for close to the last nine months or so it has a number of enhancements over the version one platform I've just described it has an optimized 3D failure surface routine it has an enhanced rock slope failure model we've integrated the triggers model that was developed at the USGS and this is a much more robust model that allows us to capture transient rainfall infiltration we've integrated a random walk 3D run out trajectory algorithm we've now included the ability to implement vulnerability fragility functions the code is now fully probabilistic so we can run a number of Monte Carlo type probabilistic simulations and we're currently engaged in some comprehensive testing against well documented landslide events in New Zealand particularly the Kaikoura earthquake that's based on the recently developed landslide inventory developed to GNS and we are applying this to several test bed locations where loss of life and financial losses have also been known and that is in Seattle, Portland and in the community in New Zealand and ultimately this code exists as a robust Python based modular type code and I say modular because it allows us to plug in advanced models as we continue to develop the results so I want to close with a couple thoughts and the first is that I think landslide risk maps provide really significant value over traditional hazard maps for about another 15 or 20 percent investment in resources you get much more actionable information but I also think it gets away from this issue of having the hazard maps covered with red with hazard zones and instead allows us to really focus and take some of the noise out and focus on the populations who are at risk I think the second point is that landslide risk mapping is realistic and achievable we got this to work in an area that has moderate quality data and I think implementing this in the U.S. as we've seen with our current studies for Seattle and Portland allows us to have a much richer data set and to have even better predictions and so I think this is within our grasp and then finally I'll say that the risk mapping provides important fundamental understanding in this case we've applied this to some international policy questions pertaining to refugees and the sheltering of refugees in urbanized areas versus camps and so forth and it also serves as a scientific reference for landslide policy so I'm going to wrap up here I'll make reference to a number of publications that have come out over the last couple of years again that detail this and at this point I think I will turn this back over to Scott and let him moderate any questions that might thank you very much Joe very interesting this is Scott Anderson I'll take for about the next 18 minutes or maybe 15 minutes or so some questions that have been posted in the chat pod excuse me not the chat pod but the Q&A function at the bottom of your screen if you have other questions you can post them there and I'll try to get them answered by Joe as well so I've got several questions here and I think maybe starting with the multimodal nature questions are for example what input goes into the inferred slope hazards like slumps and debris flows besides slope angles could you talk a little bit more about that Joe yeah I'll say that the outset of this work we we looked at a combination of both basic slope parameters as well as other kind of morphological parameters like for example convexity or concavity and what we found is really the best correlation existed with just slope when you started to add in some of the other slope morphology measures it added quite a bit of noise and so it is a it is a relatively simple screening procedure but I'll say that it's also based on a lot of statistical work that have looked at correlations between landslide moda failure and the slope that was in place before that landslide occurred and so it's again it's a it's a it's a very basic model and and we didn't see much in the way of additional value when we did our statistical analysis early on by adding in those those other kinds of parameters in fact we just felt that it was adding more variability and uncertainty to the process excellent all right there are a couple questions on the 15 degrees a threshold that you mentioned in the work in Lebanon and one of them had to do with you know how would you catch lateral spreading and and another had to do with that there are some certain geologic terrains where you know many slopes are unstable at at at less than 15 degrees can can you talk about that a little bit and then maybe about you know wrapping into that just some some thoughts on scale like that was a based as I understand it on some if you will landslide inventories and and statistical observations from the area of interest from Lebanon and so how would you calibrate that elsewhere yeah well let me begin with the the the point about the 15 degree threshold and and I'll say that in the original work on the the multimodal method that we published in engineering geology it includes a fourth module for liquefaction induced lateral spreading in fact that multimodal model was originally developed just for co-seismic landslides there wasn't a provision for precipitation induced events we didn't include it in the work that I presented today because we've been kind of waiting for some improved liquefaction and lateral spreading modules to be developed there's a couple projects that are currently in progress and we would like to ultimately integrate those into the model but but so when I say less than 15% we're not at this point including lateral spreads although that was again in some of the original model formulation and I think that that will come out in a in a future version of this platform you had a question also or a very good point about what about terrain not necessarily lateral spreading but terrain that's flatter than 15 degrees that also has a susceptibility to landslides and I'll say that that threshold really needs to be thought about in a kind of a region by region basis we don't see that kind of terrain in Lebanon which largely consists of sedimentary rocks and a lot of limestones and silt stones and sandstones and so forth and those kinds of materials have not produced those those flat landslides relatively flat landslides that you're making reference to so I don't know if that's a universal threshold there's no reason why we can't lower and in fact we we originally started to work with a much higher threshold and when we were submitting the engineering geology paper we've seen the very thoughtful review that that question that threshold and we lowered it the only cost to us is simply in terms of computational time and so we can easily extend that we're trying to watch our time because not necessarily the time for the webinar but we're trying to watch our computational time because now as we move more towards the mode of doing Monte Carlo simulations we're looking at the at the idea of something very similar to what's done for hurricanes where you look at different you know tens of thousands of simulations with different paths we're looking at different kinds of particular storm events that have different characteristic geospatial signatures and kind of better discretization of seismic events and so forth so we're trying to keep our eye on that but but that can be lowered and that 15 percent I think again was was was very applicable for Lebanon not necessarily other parts of the world so there are a couple other questions related to scale and one of them is computational time related is there any measure you can share with us about the you know time for computation in the region you showed in Lebanon in Lebanon so the the scale is is proportional as you might imagine to the pixel size of the analysis and so that analysis for Lebanon was done on a on a 15 meter pixel size basis and so that ran relatively fast we've been running this on a on a desktop Macintosh and we've been able to do those simulations over the order of single simulation in about an hour and a half and hour and 15 minutes or so now that changes significantly for our study areas in New Zealand and in Seattle and in Portland because there we have one meter DEMs and and really a vision of this work is to anticipate the adoption of the three-depth program and and looking forward to a day where we have this high resolution data that can improve for example our debris flow models and particularly the infiltration aspects of those so that significantly slows things down and and those those runs right now and and I'm going to say that this is about a probably a three-year-old desktop computer but those simulations at that finer resolution over the scale of metropolitan Seattle are taking something in the order of about six to seven hours per run and so our plan is ultimately move these to a Monte Carlo simulation on a much faster computer but that's where we stand right now so it's not really untenable but but this is it is it's not something that we run through in just a couple of minutes I see what are your thoughts on on the on downscaling of the results I mean you're you're showing fairly specific areas in fact you saw that is one of your one of the advantages that that large regions of susceptibility were boiled down to smaller ones of risk it sort of if you could talk to that a little bit and maybe what you see is the intended use of the products as you vision the future yeah and I'm going to answer that question with with reference to this slide I'm going to back up a bit because we looked very carefully at this and I want to say that that we have some some descriptions of our intended use of this kind of data and and and the role of scale and pixel size and so forth and how accurate these predictions are I don't and I actually should back up and really I prefer not even to use the word prediction but rather I tend to use the word forecast and I like to talk about verification because I really don't think we can fully validate these because we don't have the full range of conditions and storm events and so forth nevertheless the intention is is not to to to be able to discretize these results to the meter scale or perhaps even to the 10 or 20 or 30 meter scale but instead identify risk hot spots that might be might be worthy of doing a more intense site specific analysis having said that we have a test right here and again I said that this is for the 50 year storm event and I don't know what the intensity is of the event that occurred recently in Lebanon but but what you can see is that we are capturing the source areas and and some of the runout pass pretty close and then there's some other areas that are probably within about 20 or 30 meters or so that also initiated that we didn't capture with our forecasting here so I don't certainly do not want to oversell the ability of this to hyper localize the the hazard and the risk but I think really the best application particularly when we're we're applying this over the scale of of a of a nation like Lebanon nexus of 10 000 square kilometers is to identify those hot spots rather than specific pixels of failure and hot spots that might require some additional attention or might be worth moving populations from or implementing mitigation measures okay I have a couple of questions related to the root cohesion and the root strength and the use of the NDVI could could you talk to that a little bit yes and I'm going to back up a bit and I'll say that the only time we use that is for the debris flow module and I'm going to step back to that you don't mind me flipping through these slides too quickly I'll say that the the root cohesion that we add is really quite modest and it's on the order of three to seven kPa and it varies as a function of what the NDVI value is with the idea that that is depicting different types of vegetation of being that shrubbery or trees or low-lying grasses or or areas that are largely devoid of vegetation the the reason it's important to add that in is just a tiny bit of root cohesion plays a really important role in in in kind of taming down these models if they have no cohesion and and we know that you know there there's there's various sources of of apparent cohesion that closed that's that's close to the ground surface it's not just root cohesion but it's from partial saturation as well so unless we include just a bit we tend to have the entire area turn red we add just a little bit and again we we wanted to have some basis for doing it we didn't want to randomly kind of artificially add that in across the region but and we did this and I'm going to say the details of this have been described in the engineering geology paper that we published in 2016 the actual values and the calibration experiments we did using data from the Northridge earthquake which has a Mediterranean climate that's quite similar to that in Lebanon and so the the full details are there but basically what I'll say is we're adding very little and and and really the purpose is to to try to realistically capture just this tiny bit of resistance that exists at the ground surface because otherwise we make these very conservative forecasts of debris flows okay hey one thing I want to add here and it's not because Joe just said something that made me think of it it's because I just remembered to say it that it is important to note that the this work and the conclusions that Joe has has drawn here are his and they do not represent the national academy or or the the committee on geological and geotechnical engineering that's hosting this webinar I meant to say that earlier Joe and just getting to it now okay thank you Scott and of course that extends to the United States Agency for International Development and SF as well particularly with some of our findings related to policy related to refugee settlements yeah I've got a couple other sort of themes of questions here and one is about the fragility of of the buildings of vulnerability of people in of the occupants and so on the sort of the risk side of the equation here and and another one is about the computation but let's go to the the the fragility and vulnerability ideas Joe okay so what I say is that right now we're using step functions for the for the vulnerabilities and so it's really quite coarse that's one of the significant changes that we've implemented in version two of this platform is the ability to truly put in a fragility function a probabilistic function and so we can sample from that function basically that will show the vulnerability as an intensity of the of the landslide and that intensity is linked to the height of the landslide so I'm going to say that's that's the way it's changed but just backing up and focusing on this work for a moment because it is a very important factor the the temporary settlements that exist in Lebanon are are essentially tense and and offer really no resistance to debris flows and so our assumption there is that if you're impacted by debris flow and you're in a tense the loss of life is one so it's there's no step function there it's if you're hit we are assuming that that would result in a fatality if you're in a structure we've adopted vulnerability values from the published literature there's been a whole series of these have been published a lot of them have come out of British Columbia as well as a number of other places in Europe and so we've adopted those based on the building types that we see in in Lebanon and and those values are certainly less than one offhand I don't remember they those are described in the in the recent 2019 the specifics of what values we use but I believe there's something in the order of about one half that just a couple minutes left and maybe another sort of theme of questioning here is about you know communicating this information and and to decision makers and you mentioned the house bill at the HR 1261 at the beginning of this have you thoughts do you have some thoughts on how this type of information can be communicated other to other audiences besides like who you have on the call here today yeah I'm part of a project at University of Washington called M9 and there's been a kind of big theme through our project it's an NSF hazard seas project that is looking at the potential outcomes of a Cascadia earthquake in terms of tsunamis long period structures susceptible to Cascadia kind of ground motions as well as landslides and so this has been a big theme is how do we take this kind of technical information a lot of it very geospatial and communicate that to members of the public and to decision makers and so forth I would I've really learned a lot about this from working very closely with GNS science I feel that they have had some of the best risk communication products and if you Google up a landslide risk maps for GNS science or for the Christchurch area you can you can find some examples of those they have really taken it to a level of of a combination of techniques that involve things like the publishing maps and making those available in digital forms but also integrating those with stories of people who are within different hazard zones and so they have the story of a family whose house is now in a realizing a very risky area and the story of a family who lives a kilometer away who is in not in that kind of area but might want to take some risk mitigation measures I think it's a super important question I don't have a a great single answer for it except that I try to to emulate and I think what GNS has done and I think they've really been at the forefront of this and I think they've also really been fantastic in integrating social scientists with their technical teams to to help the the technical experts get those results to members of the public excellent hey Joe I just want to we're running out of time here and I want to thank you for a very interesting presentation including the answers here at the end and I want to thank the participants that were here with us this afternoon and many of you who asked questions that I was not able to get to but I did my best to compile some and hopefully we can it's quite valuable what you you've all offered so thank you and you'll see on the slide right now is is a announcement that I want to close with and that is that our committee is having a meeting on June 27th the subject of managing mind-waste risks practice limitations and needed research that is something that there is some available to join in person and or remotely and Sam Maxino whose email address is on the screen is the right person to talk to if you have an interest in supporting the committee's activities that way too with that I'm going to turn it back over to Marty or Sam if there's anything else that needs to be said again I thank you all very much no I don't have anything to add this is Marty and Joe I want to thank you on behalf of the committee and the academies for your presentation extremely interesting and valuable and thank you all for participating well thank you for the invitation I really appreciate this and thanks to the committee for coordinating and organizing this our pleasure thank you very good thank you all goodbye