 All right, we're at the top of the hour, so I'm going to get started. Greetings and thank you for attending this month's science seminar presented by the NSF's National Ecological Observatory Network, which is operated by Battelle. Our goal with this monthly series of talks is to build community among researchers at the intersection of ecology, environmental science, and neon. We are excited to have a dual presentation today by Adriana Oskanga and Nayani. Oh, I don't have the last name in front of me. I'm going to let our next presenter introduce their names very perfectly. Nayani and Adriana, our dual presenters, we're very excited for their talks today. But before we turn it over to those fantastic speakers, I want to go through a few logistics. So we have enabled optional automated closed captioning for today's talk. If you would like to use it, find the CC button in your Zoom menu bar. The webinar will consist of a presentation, actually two presentations, followed by Q&A. And we're going to do a short Q&A after each presentation and then try to leave a little time for more of an integrated Q&A, maybe reaching across themes that touch both talks at the very end of the session. So as you think of questions, please put them into the Q&A box. We also have a meeting chat, which you're welcome to use to share links or other items of interest, but add questions for the speakers into the Q&A. We'll facilitate discussion of those questions at the end of each talk and also at the end of the presentation. And there is also an opportunity to ask questions over audio by using the raise hand feature if you prefer. Neon welcomes contributions from everyone who shares our values of unity, creativity, collaboration, excellence, and appreciation as outlined in our Neon Code of Conduct. This applies to Neon staff as well as anyone participating in a Neon event. The full code of conduct is available via a link that I will share in the chat in a moment and is also embedded in the Science Seminars webpage, which I am showing on my screen and I will also share that link with you in a moment. This talk will be recorded, these two talks will be recorded and made available for later viewing on the Neon Science Seminars webpage. If you scroll down here to the list of talks, you can see that once a talk is passed, we soon thereafter make this recording available link appear and those will take you to the recordings of the seminars. To complement our monthly Neon Science Seminars, we host related data skills webinars on how to access and use Neon data. Registration for those is available on the same Science Webinars webpage. If you just go to the bottom beneath the list of talks, you come to the list of data skills webinars. Later this month, we're having a very relevant webinar on remote sensing of wildfire impacts. So if that is of interest to anyone, again, I'll drop the link in the chat in a moment. We would love to see you at that data skills webinar later in March. And lastly, if you have ideas for a talk for this seminar series, nominate yourself or a colleague today by filling out the form on our Science Seminars webpage. It's here near the top. Nominate a seminar speaker we would love to hear from you. So that's it for me. I'll turn it over to John Musinski to introduce our fantastic lineup of speakers. Thank you. Thank you very much, Samantha. I'm John Musinski. I'm a remote sensing scientist with the Airborne Remote Sensing Group. And I'd like to, I'm very pleased to introduce Dr. Adriana Uskanga, who will be our first speaker. Adriana is a post-doctoral researcher working with Dr. Kyla Dahlin at the Ecological Remote Sensing and Modeling Lab in the Department of Geography at Michigan State University. Adriana is a landscape ecologist who integrates field and remote sensing data to analyze patterns of vegetation change over space and time, investigating their impact on carbon dynamics and biodiversity. Adriana earned an undergraduate degree in biology and a master's degree in ecology and evolution at the National Autonomous University of Mexico. She went on to earn a PhD in geography from the University of Oregon, where she analyzed the interacting effects of environmental gradients and land use intensity on tropical forest structure and field diversity. Currently, her interdisciplinary work seeks to improve our understanding of the influence that disturbance, land use, and land management have on forest structure, composition, and function across scales. So without further ado, please welcome Adriana Uskanga. Hi, everyone. Thank you very much. I will share my screen with you all. There, I hope you can all see that. So my name is Adriana, and today I'm going to share with you part of the project I've been working on with Dr. Kyla Daling at the Ecological Remote Sensing and Modeling Lab in the Department of Geography at Michigan State University. And the main focus of this project has been to characterize vegetation change over time or disturbance trajectories to improve forest structural models that in turn help us to make better predictions of forest carbon uptake. And for that, we've been using neon sites and Landsat data. So let's begin with some definitions first. Forests, when we think about forest structure, we're referring to the 3D arrangement of trees and other plants. And we measure forest structure with variables such as leaf area index, or which is the area of leaves per unit ground, or variables such as canopy height, which is the height of the trees, canopy rugosity, which is a measure of how complex is a canopy in a forest. And there are many other variables of forest structural attributes that we can, we can measure in the forest. And we care about these forest structural attributes because they are relevant in for different forest ecological processes and dynamics. So for instance, researchers have found that a higher complexity in cannabis is related to higher biodiversity in the forest, for instance. Understanding forest structure is also important for understanding the cycling of water or the water cycle in general. And perhaps more important for us, forest structure is very related to how much carbon forest can take up and sequester. So if we think about how leaves and trees are arranged over space, we can, for example, we can visualize how different, that will change how different leaves can intercept sunlight and in turn how those leaves can make photosynthesis and uptake carbon from the atmosphere, for instance. Just to give you an example of how forest structure is related to these very important functional processes that take place in forests. So really understanding the spatial distribution and spatial patterns of forest structure is really important for us to improve our understanding of this ecological process and specifically carbon dynamics. So these two forest structure and forest function are very related. And forest structure is determined by abiotic conditions, such as climate, soil and topography of a forest, but also by historical processes, by the history of that forest, such as the disturbance of a forest, for instance, if a forest had, there was like a fire in a forest, it will have a different structure than a forest with no disturbances. Land use and land management are also important for determining forest structure and succession, which is basically how vegetation grows after a disturbance. And in general, when we think about the models we have for modeling forest structure, we actually models make a pretty good job at including abiotic conditions, such as climate, soil and topography, but they don't, many models do not include these variables related to the history of a forest. So we don't have models in general lack this this part of like variables related to disturbance that are included into the models. So that's what we've been working on in this project. And part of this is because there's this general idea that disturbance and land use is only or are only relevant at very local scales. But actually, there has been a lot of research showing that disturbance and land use are also relevant, not only at local scales, but also at larger extent, such as regional, continental and even global scales. So we think about that there's about like 60% of the forest around the world has been modified by humans. Then it kind of naturally follows that we probably have a large effect on how on like this 3D arrangement of plants around the world. But the other challenge in including disturbance or variables related to disturbance and land use into forest structural models is the data, the type of data we have for doing that. So sometimes we use records of past land use or ownership and forest inventories, kind of to get that information of the history of a forest, but those are not always spatially explicit. So it's difficult to understand the spatial distribution or to model the spatial distribution of forest structural variables with that spatial explicit data. And sometimes we use things like land cover maps, for instance, but they tend to be cores and they simplify these very complex processes of disturbance and land use and succession. So what we really need is continuous data that represent the complexity of disturbance and land use processes with accuracy. And that's why we in this project, we've been using Landsat time series analysis. So for people that are not very familiar with remote sensing technology in general, Landsat is this mission run by NASA that they have launched satellites from the 70s. So we have actually Landsat is the longest record of the Earth surface that we have publicly available today. And each satellite has a repeating cycle of two weeks, and it has global coverage. So that means we actually have many images, or we can get many images from all the forests around the world. And in this figure, this is an example of one of these Landsat time series, where they basically stacked or they make a series of Landsat images ordered over time. They calculate a vegetation index to follow vegetation change over time. And in this example, you can see, for instance, that this forest in Oregon, there was a fire in the year 2002 represented here by these red colors. And then the authors could actually track how vegetation grew after the disturbance over time. So this is basically what we've been using for analyzing disturbance trajectories. So with that background in mind, the objective of this project is to improve forest structural predictions by integrating patterns of forest disturbance and recovery across time and space over the last around 40 years. So the idea is that using these disturbance trajectories and integrating these disturbance trajectories or data of these disturbance trajectories in our models, we can actually improve models of forest structure. So that's what we're testing in this project. And for that, we're using neon data, so the National Ecological Observatory Network. And in particular, right now, we're using, we're studying Harvard forest because Harvard forest has not only all the data that neon people have collected in that forest, but also data on previous land management and disturbances and things like that. So as many of you probably know, neon collects or like, yeah, neon people collect a lot of data in neon sites, field data on the ground. They also have flux tower in each site, but they also have an airplane with a LiDAR sensor and a hyperspectral sensor, which is great for us. So basically we've been mapping different forest structural variables using the LiDAR data from neon. And today, the results I'm going to show you are just for relief area index, but we've been also working with other forest structural variables such as canopy height and canopy rugosity. So we've been using LiDAR for mapping relief area index and for the disturbance trajectories, we've been analyzing Landsat time series using three different vegetation indices, the normalized different vegetation index or NDVI, the normalized difference moisture index or NDMI, and the tessell cap wetness. And to analyze this Landsat time series, we've been using this algorithm called LandTrender implemented in Google Earth Engine. So LandTrender is basically a segmentation algorithm. So it basically takes a time series and it splits that time series into different segments that are connected by vertices. And each vertex represents a change in trajectory. So a change in vegetation. Something that's really important for us and like that we really like about using this is that this algorithm and this method is that it's we get information pixel wide like by every pixel. So we get that disturbance trajectory per pixel. And each segment has a magnitude, adoration and a rate. So with that information, we can derive many disturbance metrics that we can later on use in our models. And disturbance metrics are things like the number of vertices in a time series, for example, or how many disturbances there were in a time series. The time since last disturbance, the magnitude of vegetation decline. And there are many, many other disturbance metrics that we've been using. And then we derive from LandTrender. So here I'm just showing three of these examples. On the left, you can see if NDVI for instance grew in every pixel along this like 40 years. In the middle, it's the NDMI magnitude change. So basically if NDMI increased or decreased over time over like the 40 years of time of time series. And also the standard deviation of NDMI over time on the on the writing in purple. But as I was saying, we have many of these. We derive 38 disturbance metrics per index. So we ended up with a pretty large dataset. So with that dataset, we've been feeding models of, as I said before here, I'm just showing models for live area index, although we have done it also with canopy height and other forest structural variables. And in these models, we include abiotic conditions. So information on soil and topography and solar incoming radiation as well. And we also include the disturbance metrics. So just to give you an example of how that looks on like maps. On the first row, there are some predictor variables related to topography and soil. Then we're also using the NDVI Tesla cap wetness and NDMI for the year where data was collected, which is to the year 2018. And then we're adding all the disturbance metrics we derived from land trender. And we've been doing two different approaches, modeling with ordinary least squares regression, which is basically just linear models. But we've been also feeding simultaneous autoregressive models, which are linear models that account for spatial autocorrelation, which is something that when you're analyzing these geographic patterns, it becomes really important. So basically SAR models are linear models that include a term for the distance matrix between pixels. And that helps for accounting for spatial autocorrelation in the models. And we've been feeding these models in three different steps. We've been feeding these models only with abiotic variables, then with abiotic variables and adding the vegetation indices for the year where data was collected. And then with abiotic variables, the year, the vegetation indices for the year of data collection, and the disturbance metrics. And we've been doing this to test whether this, including disturbance metrics actually improve the models or not. So here are some of the results for live area index. So in these scatter plots, I'm showing on the y-axis, the live area index, the observed values of live area index on the y-axis, and then the fitted values for live area index on the x-axis. And with, as I was saying, in these three different steps, abiotic, abiotic and vegetation indices, and also including the disturbance metrics on the third column. And with linear models and SAR models on the two different rows. So in general, there are two main kind of results or like take home messages that we can take from these results. The first one is that SAR models make a better job at explaining, in this case, live area index than just like simple linear models or ordinary least square regressions. And you can see this on the distribution of these points, but also on the r-squares of these different models. And the second, so actually like including this term of like correcting for spatial autocorrelation improves the models a lot. The other main like kind of like big result or take home message from these results is that actually model performance improves when adding the disturbance metrics. So as you can see for both ordinary least square regressions and SAR models. So for instance, for the ordinary least square regressions, the r-square improves from like 0.13 to almost 0.5 when we include the disturbance metrics. And for the SAR models, even if the r-squares between the model using only abiotic variables and the model using also the disturbance metrics is not that different. I want to draw your attention to this AIC value where you can see that AIC actually reduces or like declines a lot when we include the disturbance metrics. So our results so far have shown that data on disturbance trajectories that we have measured through these disturbance metrics improve forest structural models. And although the results I'm showing here were only for leaf area index, we have seen the same kind of the same trend with other forest structural variables. The predictors change with different forest structural variables, but it's kind of the same trend where SAR models work better than ordinary least square regressions and that adding disturbance metrics improve the models a lot. So we've been very happy with seeing these results. And in the future we would actually like to, so these models have been very helpful for explaining the spatial distribution of forest structural attributes, but we actually would like to make some predictive models to kind of use this similar approach, but actually to predict how forest structure could look like in other sites. We would also like to do these with planned functional traits using the hyperspectral data that Neon collects. And both for like especially for the predictive models, it will be very interested to test the same approach at larger extents. So we plan to not only include other Neon sites also in tempered forests, but also use, we would like to use these methods with like other forest and especially we're very interested in seeing if these methods hold true or if these methods hold true for other understudied regions. So regions where we for instance don't have that much information on the disturbance history and previous land use and land management, but also where we don't have for instance airborne LiDAR and things like that. So we're very excited to be working with Neon sites where we have a lot of data to then being able to test these same approaches in understudied regions such as tropical regions for instance that we really need to be working more on those areas. And finally, we've been also working on this idea of disturbance syndromes which is basically like a classification of the disturbance trajectories we've been analyzing across across different ecosystems. So that's for future for yeah for the future future results. So finally, I would like to also announce that we've been running this spectral ecology summer school spec school is a school for grad students and postdocs that are interested in terrestrial ecology and remote sensing. And we use a lot of neon data in this school. So if you know someone that could be interested or if you're interested in joining, we will open applications by the end of this year. And we're currently running the second year of spec school. And the in person portion of the school actually takes place in the mountain lake biological station, which is also a neon site. So yeah, fun times if you're interested, please join or send your application and spread the word. So with that, I would only like to thank you for listening and also thank Kyla and people at the AirSum Lab and the Sevmian Landsat and the Landsat mission and LandTrender and Google Earth Engine for making this project possible. And with that, I think we do have a couple of minutes where I can take questions. Thank you very much. Thank you so much, Adriana. It was a great presentation. Yes, we do have a few minutes for questions before the next presentation. So please, if you didn't see the message in the chat from Samantha, add your questions in the Q&A box. Yeah, the first question we have is, have you noticed differences between sites or forest types in the influence of the disturbance on structure? So something important in this project is that we're actually not kind of studying how different disturbance have different effects. Most probably, I mean, if there's a fire in a forest that will have different effects than a pest outbreak, for instance, on like how forest structure changes. We're not, since we're working, like the idea is to work at landscape regional continental scales and make like a classification of these different disturbances. We want to see whether different forest react similarly, even if to disturbances, even if we don't know exactly which disturbance caused the, or what was the cause of that disturbance. So yeah, I don't know if that answers the question, but that's what we've been trying to do. And we haven't compared these results or like these trends with different forests. So I will look forward to do that, including like all other neon sites in this analysis. Okay, great. Hey, I think we have time for one more question. Hi Andrea, great talk. I was wondering if you use some sort of temporal interpolation for filling misting lens at images due to cloud cover? No, that's an excellent question. So what LandTrendr does is instead of using all the images for like the whole year, well, first of all, we do some like cloud screening and we do mask clouds from these Landsat images. But we also make a meteo image, so a composite image just for like a couple of months. So we use July, August, I think, for each year. So we don't use, so in that way we kind of remove the seasonal variation and we also remove the effect of clouds or like cloud covering. So that's what we've been using for the Landsat time series analysis. And I'm happy to talk more about that process later. Okay, great. We have a couple of more questions. But for the sake of time, what I'd like to do is save them to the Q&A session at the end of the talk and first give a chance to Neani to present and then we can open the floor to all the questions at that point forward. So to those of you who put these, send these questions in, thank you very much. We'll get to them in a few minutes. Thank you very much, Adriana. Okay, I'd like to introduce Dr. Neani Ilankakun. Neani is a research scientist at Earth Lab at the University of Colorado in Boulder. She earned her PhD in Geosciences from Boise State University in 2020. Her background is in remote sensing applications of ecosystem ecology and forest diversity. Her recent projects include post-fire recovery and carbon potential in western U.S. conifer forests using Jedi, spaceporn lidar, and unmanned aircraft systems, UASs. In addition, she works on model and algorithm development with Jedi data to estimate semi-arid ecosystem biomass density, implementation of UAS methods to understand post-disturbance vegetation regeneration, ecosystem transformations, and their impacts on carbon. Please welcome Neani Ilankakun. Thank you, John. Hi, everyone. First, let me share my screen. Hi, everyone. I'm very excited to talk to you today and thank you for this opportunity. So first, before I start my talk, I think I should acknowledge my funders, the NSF series and Earth Lab, and also all my contributors and collaborators for this research. So I'm Neani Ilankakun. I'm a research scientist at Earth Lab. So during this talk, I'm going to provide summary of some of the research that I did our past couple of years. They are all interconnected and mostly address the impact of forest diversity and ecosystem transformation due to disturbance. And I have used several remote sensing data sets from satellites, drones, neon data, as well as the drone data. So my outline of talk, like, I will briefly go through the Western U.S. forest ecosystems and the forest disturbance, and then some of the research highlighted, I did on the disturbance and the impact on the disturbance. So these two maps shows the forest distribution in the Western U.S. and the first one is once we talk about the different forest types. And I mentioned the right from the land change monitoring assessment and projection data set, as the map data set. So particularly that data set shows that distribution of all the land cover classes. And this map is from 1985. So we have that data set from 1985 to today. And I also overlaid the fire data set. That's one of the major disturbance in the Western ecosystems. So you see, like, there are a ton of fires happening in the Western ecosystems, especially in the forest ecosystems. And so the overarching question is that how Western U.S. forest ecosystems respond to disturbances. So in addition to fires, there are a ton of other different disturbances happening in the forest ecosystems, especially droughts, fires, beetle kills, and also climate warming, and also, like, the invasion as well. So the figure in the right, just to show that a conceptual figure that how we see, what is our assumption on it was in the disturbance and the recovery. So the state variable could be anything. It could be the carbon. It could be the land color. It could be a different index, or it could be anything. And so that all these lines shows, like, what we considered at the stable states, what's the transition can happen during the disturbance, and what if it's, if that ecosystem recovers, what's the recovery trajectory look like? For example, this green line with the uncertainty shows the recovering and then becoming to stable states after that. And also this yellow line on the bottom of the figure shows that if the ecosystem is not recovering, and maybe changing into an alternative state. So I will talk more about fire today, because most of my research lies around fires. And so the first thing is that post-fire recovery in the western US. So the figure here is just to show, like, how many fires happened in last 30, 40 years in the western US forest ecosystems. So this record shows that both the number of fires and also the total burned area increases in the western US. And so when we have that, and it continually increases the frequency and the intensity, everything increases during the last couple of decades. So the question is, like, what actually happens after a fire? So there are a couple of ways that ecosystem can undergo after a fire. So if the fire, see, if the severity of a fire is not that high, it's a low severity that ecosystem can quickly recover back to its original state. But if the severity is really high, there could be a mortality phase and also that even with the mortality, the semi-co-systems are really resistant that they could recover back to the original state. But it could take some time, but semi-co-systems may not recover into the original state. So it could maybe during the recovery phase, they maybe have additional disturbance and also the available resources may not enough for the recovery back into the original state. So it can go different alternative states. So the first thing that we did to understand like the post-fire recovery trajectory in the western US ecosystem, so we did this study for three different ecoregions. I'm only showing one here. So we did this one for the southern Rockies, Pacific Northwest, as well as the northern Rockies. So I'm only showing here the southern Rockies. So in the southern Rockies, the dataset that we used to understand the post-fire recovery trajectory is the JLi data. So the global ecosystem dynamic investigation data. So it's a space-point LiDAR data. And the reason that we use that one is that it provides structural data near global scale. So we thought there might be a good dataset that we could understand the post-fire recovery in a regional scale. But the JLi data started delivering the data from 2019. So if we want to have that recovery trajectory, then we need to go back in time. And so this study from all the fires in the forested ecosystems, southern Rockies, northern Rockies, and the Pacific Northwest, and the fires that we considered from 1985 to 2017. But we don't have JLi data for the 1985 to 2017. So what we did was like we used the space for time or the chrono sequence approach. So our assumption was like if the ecosystems are the similar conditions, the topographic conditions and as well as other forest conditions are the same, we could use the space. We can replace the space for the time. So our results show that in the southern Rockies, the recovery trajectory, the figure in the right shows the person can have a change compared to the unburnt area nearby for the fires across different fire severity gradients. The red line shows the high severity, blue line since the moderate severity and the green line is the unburnt of the low severity. So it shows that if the fire happens at low to moderate severity conditions, it's easy to recover in the southern Rockies. But if the fire happened in high severity, then it's really hard to recover. And even like after 30 years, it's very low the recovery. But it shows like based on the trajectory, it shows like like if let that ecosystem recover for maybe next 30-40 years based on the extrapolation of this trajectory might recover back into the original state. And what other thing that we found is that not only the time since fire, there are other factors also contribute to the post-wire recovery. Some of those factors are the fire size, the distance to the seed sources, drought conditions. This PEI is the mentioned in the drought conditions and also the topographic conditions. And so as I mentioned that based on our trajectory, it shows like it's in the southern Rockies if time allows the ecosystem may recover back into the original state. But this data that we analyze our structural data, it doesn't show anything about the species composition or anything. And so one thing that we thought about is that when we talk about that post-fire recovery the images in the right side show two different things. One is the seedling, the top one that shows like ton of seedlings in one of the fire that we visited last year. And the bottom image shows that the same fire at high elevations. So it doesn't have any seedlings. So but when we measure those conditions, especially the structural conditions, both would show the similar pattern. But necessarily the first one actually shows the recovery and regeneration, but the second one is not the recovery regeneration. That's just a ton of shrubs that may be potentially this region is not recovering, may be transition into something else. So and so this is one of the study that I saw on Davis at all in 2023 paper that talked about the seedling generation and after a fire. So it shows that in high in the figure on the left shows like in high severity scenarios, what's the post-fire recruitment probabilities. So if the fire severity is high, it's really low probability for the seedling recruitment. And also some of the other factors that can control the seedling region treatment probability shows in the right. So it's the distance of seed sources mean to recover fire severity and some of the other climatic controls like the climate water deficit. Likewise, so there are a ton of other conditions that site-wide conditions that can potentially impact and the post-fire seedling regeneration and the seedling recruitment probability. However, this study was done with ton of in situ field data seedling counts plot level data, but we thought like it's really hard to like get ton of like field information, especially in the high elevation forest, like it's really hard to get into those forest and collect those data and have like information on where the seedling regeneration happens, what areas that we need to help for the regeneration of this. Anything that we could probably provide the right framework, like if the ecosystem is not recovering, what else we could do. So it's really hard to get into some physically get into some of those locations. So we thought might be a drone remote sensing might be a good option that could potentially collect more information and can probably go align with the in situ field data and that can cover large areas dancing in little field plots and can help to better understand regeneration after a fire in different ego regions. So in this one, what did it we again this one basically happened in the Southern Rock East Eco region. So we actually visited 10 fires in the Southern Rock East Eco region that happens around 20 years ago. We chose 20 years ago because we want to let them because those ecosystem to show some signs of recovery or transformation. So we could find that factors what helps for the regeneration, what does not help for the regeneration. So we collect the drone data. So the RGB images from the Phantom for drones, these drone images for like three centimeter solution images that could with the photogrammetry, a structure from motion we could delineate individual shrubs, even like smaller shrubs likely with height about 50 centimeters. So we could we were able to generate those individual trees or shrubs and also its spatial locations. And then what we did like we actually like use information from the drone images to classify into different categories. So we have eight different categories. So deadwood, ground herbs, live trees, rocks, seedlings and shadows and shrubs. So why we had like all these different categories, because like our assumption was like in different site, micro site conditions might also control the seedling equipment and the region version. So most of the studies consider like the macro scale factors, the climatic factors, the and elevation and top of victory in like regional factors. But we thought like maybe there are some other site level micro site conditions that might also control the post fire regeneration. So this is our workflow in the right that we develop to understand that micro site conditions on the post fire seedling regeneration in the southern Rockies. So in other study that it shows not only like other climatic and topographic gradient factors, but also like the sharp density, deadwood density, standing density also like impact the seedling regeneration in the, especially in the high elevation areas. So this is one figure that I'm showing the very wide that higher the sharp density, higher the shielding density in the southern Rockies, especially at the high elevation forest. So in this feature, so this is still undergoing. So our next step for this one is to evaluate the post fire configuration strategies against the micro topography and the site conditions. So in this one, like we are evaluating the sharp density, standing dead density and also some other micro site conditions. And so the all these first research that I talk about considered the fires that only burns only one times. But when we take a look at the fire data set, this is from the reality, the reality combined fire data set. So if we consider the fires from that data set that combine like different data sources with the fire records, and I only consider here the fires from 1985 to 2017. And it shows like some of the ecoregions in the vast fires that burn more than seven times. So what's the recovery if the region burns more than one time? So our preliminary, so this very minute, excuse me, this I'll show that some of the ecoregions in the western US likes burn new areas, but some of the ecoregions in the western US like repeatedly burn. So this could be like for us to turn into grass and the grass cycle helps more fires could be that reason or maybe there are some other reasons that potentially from it repeat fires and the new fires in those regions. So in this study that our next step is like evaluate the carbon recovery with repeat fires. So in this one we are planning us Jedi and I said to combine the biomass density data and evaluate that repeat fire carbon recovery potential across ecoregions and also against non functional types like so this research was funded by series IRP. So that's another research that I'm doing. And so the other thing like first thing I'm talking I talk about the fire and recovery and the next one is that the fire and diversity how that fire changed the diversity in forest ecosystems. So when so when we talk about the diversity we can measure time usually like we measure diversity based on taxonomic species composition like those data. But also we can calculate the diversity in the same way using other metrics the remote sense metrics. So if you could derive some of the remote sent traits, then we could use that trait in the same way that we could calculate the different diversity matrices. So in this one we have three different matrices the functional divergence functional evenness and functional richness. So to calculate these three different diversity indices we use three different traits. One is the canopy height one is the plant area index and one is the foliage height diversity. So we calculated this one using airborne full wave form data. So and so once we have calculated this diversity this is a figure that shows like the distribution of diversity across the word shed. So the first figure is the functional richness, the second one is the functional evenness and then last one is the functional divergence. So you see like the distribution of this functional diversity across the word shed is different and so there may be some reasons why this diversity indices distribution varies within that eco region within that watershed. So like next steps are to see like what's the signature of this distribution of diversity across the word shed. So our research shows that the diversity functional diversity distribution within that watershed where it is based on some other factors especially the topographic gradient and also some other factors like when we talk about the topographic gradient especially the elevation and aspect change functional diversity distribution within that watershed and also we found that some of the other factors like distance to water that also like change that distribution of the functional diversity within that watershed and also we found that if there's a disturbance like fire happens that also change that functional diversity. So we compared like four different fires that happened during the happen two different time frames like one is like very recent to the data the airborne data that we collected and one fire happened like 18 years before that the data was collected still like we see like a difference between the functional diversity of the burned and unburned side neighboring each other within this all four different fires. So it's so we see the difference in the functional trace as well as the functional diversity and we see like the difference in the functional diversity is higher than the difference in the functional traits alone and so that was like a functional diversity we did in a one watershed but we thought like maybe we could use the same functional diversity idea to see like how that change the how can help that understand that first the invasion. So in this study we actually use the neon data so we use the neon field data and also neon airborne lighter data. So what we did like for all those neon neon sites that represent the forested ecosystems that we calculated from the field data we calculated like a different diversity exotic and native diversity matrices and also we from the airborne data data we calculate a set of diversity indices and so the first three that they had that the left image shows that the principal component analysis in the pc one the first and second principal component analysis and how this where how the native diversity matrix varies across the neon sites and the right figure shows that we calculated a ton of maybe 10 to 20 airborne lighter based diversity indices and see like how they varies across neon sites and also we use that's for the native diversity and the lighter diversity and also we we have here calculated from the field data neon field data like how that diversity the the various invaded was not invaded. So in this one it shows like there's a clear signal of the diversity variation in the native of the invaded and the not invaded. So what our goal here is that whether we could use structural diversity signature that could tell us the invasion potential in these sites using only the structural data. So we could potentially find out like based on the structural signatures whether this ecosystem is prone to invasion or not. So in this study so our next step for this one is to find that the signature of structural diversity on forest invasion in the western use forest systems using neon airborne lighter and so the last phase is that disturbance and ecosystem transformation. So all this like fast things I talk about for the after disturbance assuming that ecosystem is recovering like what happens and how much how long it will take and how we can recognize that recovered trajectory and how we can what other controls that might help or not help for that recovery trajectory. And the other thing is that there are other than the recovery some ecosystem may not recover they may come into an alternative state so that so in this one what we are trying to find is that using remote to sense data the the ecosystem transformation in the north central region. So in this one what we are thinking about two different things fire in ratio species and climate so they are like very interconnected and one reason so there are frequent the number of fires as I early mentioned the number of fire increases and then the area burning increases and the repeat fire there are tonal repeat fire also like some areas that have ton of repeat fires happens to and also like climate warming continues. So we want to see like how interconnect this one and see like where the invasion happens and can we recognize this invasion with using remote sensing data before it actually happens. And the other thing is that in the climate data we see like between past and future the variability of climate records also varies and so we want to if we want to forest management perspectives like we also need to adapt our forest forest management practices based on these variables to help the ecosystems best. So in this study that by the way yeah so in this study like we have like we measure this one state level ecology and management polygon scale and so our goal is to recognize in mission before it happens and also like how that invasion changed the carbon storage on those ecosystems. So this is the preliminary resource that we found like in the southern rockets in the left two figures so like the three annual three cover change in the southern rockets in the last 30 40 years and also the annual brush up covers in the southern okay so they are completely opposite directions one is going down and the one is coming like going up and then in the last 10 years like it's like a steady down in the tree cover. And so if we go in the right shows that the same thing but had been did for the seven states in the north central region Colorado, Kansas, Montana, North Dakota, Nebraska, South Dakota, and Wyoming. And so in this one like what we are going to use that first like the indicators of ecosystem transformations and then find the location that transformation happened and use the remote sense data to recognize the threshold that would tell us that if this ecosystem is transforming and then finally to and just evaluate the carbon change due to this ecosystem transformation. So we haven't identified several indicators from different remote sense data sets so especially we are going to use the modest data for this one. And we had a workshop in last December which shows with the researchers and also stakeholders to identify the locations of transformation and potential causes for the transformation in those locations. And we are going to use that information as also the remote sense information to identify the vegetation stability and also vegetation sensitive to fire and climate. With those data we are going to develop the ecosystem vulnerability index and that would help us to map in the regional scale for the whole north central region like which areas that are vulnerable to transformed in the next maybe next decade or so. And with the data we are going to we see like clear signals of the biomass change in the burned and unburned. And so then the right figure shows like the what's the current biomass in this state level. But what we are going to use that we are going to make a connection with the time series data of the modest and sentinel data and the Jedi biomass data to develop time series of biomass data that would help us to understand the change of the biomass as a result of ecosystem transformations. And also finally like we want to link these transformations to the management practices so whether it's treated untreated if it's the treated like what type of treatments happens and how that helps to recover the biomass in this areas. So in this one there are five steps like find locations where the one location is vulnerable to future transformation and what's the impact on carbon storage. So with that I want to conclude that my the research that I talk about these are the basic conclusions from all these things that disturbance frequency and intensity increases. And also the disturbance victories the carbon storage and most of the studies that I did are from the fires. But there will be other signatures that we might need to have more research and also some ecosystems that creep it by us some are not and why that happens and so it's also needs some attention. And you know so it's really critically important to identify areas that transformations happens and then why that happens and how we can help. With that I want to finish my talk but these questions for you that would probably help us to improve our research. Thank you. Thank you very much Nayani for an excellent presentation. We have two minutes left for questions. One came in during your talk by Nayani based on your experience using Jedi data for analyzing height changes was the noise in Jedi light are so high that it overshadowed the signal in height changes due to fire. Did you use only the nighttime data as daytime data may have more noise. And I was curious about how you dealt with ground detection in areas with dense canopies. Yeah so in this one we use the daytime data and also we use some of the quality flags that Jedi data comes with to have like only the high quality data. And so we didn't consider anything dense versus not dense. We only considered what it comes as is. But we instead of not so one thing we did we did not use only the canopy height. We use the other two datasets to the two other indices the Jedi data calculus, plant area index and the foliage height diversity as well. So our final conclusions we developed the trajectory with all three different indices and so our final conclusion based on combination of all three. So the foliage height diversity can probably recognize the different densities within the canopy. And the one other thing that when we do this research so because the Jedi cannot recognize some of the seedlings that are very short. So that was the drawback in this one so that's why we move into the drone data to see like how much carbon or the trajectory that we have missed because the Jedi cannot recognize that one. Great thank you very much. Well thank you both Anayani and Arianna for your excellent talks. We look forward to seeing you again and any further questions will be sure to send to you via email. Thanks everyone I put a couple of notes in the chat but we if you're interested in fire and remote sensing please join us for a data-skilled webinar later in March and then we've got our next edition of science seminar series again second Tuesday in April. We hope to see you for those. Thanks again for two excellent presentations. We appreciate you very much and have a great day everyone. Thanks for joining. Thank you all. Bye-bye.