Monitoring Of Aquatic Invasive Species with Remote Sensing

0:00 Speaker 1: Specific to this section. By the end of Part 2, participants should be able to describe the extent and impacts of aquatic invasive species on biodiversity, ecosystem functions, and nature's contributions to people. Describe key considerations, benefits, and limitations of remote sensing of invasive species. Identify applications of airborne data for monitoring aquatic invasive species. Identify relevant NASA multispectral and hyperspectral data for mapping and monitoring invasive species. Compare remote sensing methods used to monitor aquatic invasive species.

0:40 Speaker 1: It's important to note that the definition of invasive species is context-specific. All species are native to somewhere, but when removed from that context, they may become invasive by outcompeting native species, exerting new pressures such as predation in the case of animals, and otherwise disrupt the balance of the native ecosystem. Finally, if you have any questions during this presentation, please put them into the questions box so that we can address them at the end of this webinar. We'll try to get to all of the questions during the Q & A session at the end, but any remaining questions will be answered in the Q & A document and posted to the training website about a week after the training, so don't worry if we aren't able to answer all of your questions today.

1:23 Speaker 1: And with all that information out of the way, I would like to invite Erin to share her presentation. I'll be back at the end of the presentation with some closing remarks, reminders, and things to look forward to in part three.

1:41 Speaker 2: Welcome, everybody. My name is Erin Hester, and I'm a professor of civil and environmental engineering at the University of California, Merced. I've been working with remote sensing and invasive species for a very long time and I'm excited to share the material with you that I've learned over the many years. Let's start with an introduction to aquatic invasive species. We see news like this headline from Mongabay from earlier this summer every year right around the beginning of the summer. where aquatic invasive plants are causing challenges to water resources.

2:23 Speaker 2: So this headline calls water cancer, this highly invasive species called water hyacinth that was choking waterways in Egypt. And in Egypt, just this past summer, it was estimated that this plant is causing the loss of around 3.5 billion cubic meters of water each year through evapotranspiration. So that's more water loss than an open channel. So for context, that's about 4% of the country's annual share of the Nile waters. And so wherever these water hyacinths multiply in a waterway, they're responsible for the loss of thousands of acres of farmable land and the loss of billions of cubic meters of water that could be used for irrigation and agricultural purposes.

3:16 Speaker 2: So when we think about aquatic invasive plants, they have a lot of impact to the world. And in the aquatic realm, there is this global trend of increasing spread of invasive aquatic vegetation. And this has negative consequences for both biodiversity and human livelihoods and well-being. So compared to terrestrial environments, freshwater aquatic habitats are disproportionately more vulnerable to invasive species and they're more likely to be negatively affected by invasive species. And these aquatic ecosystem invaders are sometimes called ecosystem engineers in these aquatic environments because they can affect environmental conditions.

4:05 Speaker 2: So they can change light availability and change water temperature and water chemistry, as well as affect water flow and the nutrient and carbon cycling and alter species and communities in these ecosystems. They also have impacts, as I said, to human well-being and likelihood. What we're seeing here is a picture of tugboats that are pulling a cargo barge that's actually stuck in a portion of the Pasig River covered with tons of floating water hyacinths in Manila. This was taken in October of 2020 and was featured prominently in regional and national news.

4:46 Speaker 2: Recent estimates for the economic cost of invasive plants to the United States have been conservatively estimated at $ 190 billion from 1960 to 2020, or about $ 2. 4 billion per year. When we think about the impacts of aquatic invasive plants specifically, we can think about examples such as what we see in the news here, where they impede commercial vessel travel and recreational uses They can also affect pumping for hydropower irrigation and water management. They also have negative impacts on the aesthetic value, and it's been demonstrated that places with these infestations of aquatic plants show decline in property values for homes around these waterways and has negative impacts on tourism.

5:40 Speaker 2: And finally, and perhaps most alarmingly for us, is they've been demonstrated to be bad for human health. And these outbreaks of aquatic invasive plants have also been traced to malaria outbreaks and have been documented as a concerning vector of spread for schistomyosis. Invasive plants are found throughout not only the United States, but much of the world, as we saw through some of those news articles previously. And they occur in a variety of aquatic habitats, including lakes, ponds, river streams, estuaries, and wetlands.

6:16 Speaker 2: The species of concern are very diverse in taxonomic groups and include not only angiosperms, but also macroalgae with plant-like growth forms. And there are three main growth forms that we can think about for aquatic invasive plants. There are floating plants that occur on the water's surface. There are submersed aquatic plants that are rooted underwater that potentially top out to form surface mats. And there are emergent aquatic invasive plants that have erect stems above or on the surface of the water or saturated soils such as those in wetlands.

6:58 Speaker 2: How do aquatic invasive plants become invasive? Well, the major invasion pathways are through human-mediated introductions. What you see here is a picture of water hyacinth. That's that water cancer example that we heard about from that news article in Egypt. And we can see that it creates these very beautiful purple flowers. And so a lot of times people want to use them in aquarium trade or water gardens. And so they intentionally ship them around the world to try to make something beautiful without understanding what the consequences are.

7:35 Speaker 2: They're also spread through shipping and navigation canals, through the practice of aquaculture, and through boating and fishing, because these plants and their propagules are transported by these boats and fishermen. The ability of an aquatic plant to invade a habitat is the invasibility of the plant, and often these occur in degraded aquatic ecosystem habitats. So these are aquatic ecosystems that have an excess of nutrient inputs or nutrient pollution into the waterways or those habitats that have undergone extreme hydrologic alterations or are seeing increasing temperatures that are favorable for these aquatic invasive plants.

8:23 Speaker 2: And then we need to think finally about the mechanisms of invasion, and those are the characteristics of the plants themselves. And so these are things like the genetic traits of a given aquatic invasive plant, their ability to spread clonally and the propagule pressure of these plants, as well as the biological interactions that they have with other species. In recognition of the threat that aquatic invasive plants have to biodiversity, the Convention on Biological Diversity has set an action target to specifically reduce the invasive species and the establishment of them by 50% around the world by 2030.

9:12 Speaker 2: So in light of these impacts of invasive species around the world, And in light of the policy goals that we're setting to try to reduce these, I want to share with you a case study in using remote sensing of aquatic invasive plants in California's Sacramento-San Joaquin River Delta and how we can use remote sensing to help us better manage these aquatic invasive plants and think about how we can meet those targets to reduce threats to biodiversity. In October of 2020, California's Governor Newsom outlined a comprehensive and results-oriented agenda to expand nature-based solutions across California.

10:00 Speaker 2: through his Executive Order N8220. This Executive Order called for restoring nature and landscape health to deliver on climate change goals and other critical priorities, including sustainability, securing food and water supplies, and achieving 30% conservation of California's lands and waters by 2030. A very key opportunity in these nature-based solutions is the use of restoring wetlands. Wetlands provide a fantastic opportunity for climate-based solutions. as well as delivering water and food for people in California.

10:39 Speaker 2: And so there's a big effort to try to restore wetlands in California to meet these goals of 30% of conservation of California's lands and waters by 2030. So enter the case study site that we are going to look at, which is the San Francisco estuary and the upstream component of that estuary, the Sacramento-San Joaquin River Delta. So here we are looking at an aerial view of the San Francisco Bay, and you can see the Golden Gate Bridge in the background. This San Francisco Bay is located in the western United States off the coast of California.

11:17 Speaker 2: It is the largest estuary in the western Pacific Ocean, and the upstream component of it, the Sacramento-San Joaquin River Delta, is a very interesting and special place. It has been called one of the most invaded ecosystems in the world. In addition to being one of the most invaded ecosystems in the world, it is very important to California, and it provides fresh water for over 27 million people. It is the hub of California's water infrastructure that's fueling California's $ 3 trillion economy.

11:56 Speaker 2: And this upstream component of the estuary, the Sacramento-San Joaquin River Delta, is also home to over 600,000 people who live in the region, primarily in rural and agricultural communities. It's also a global biodiversity hotspot, and it has over 750 plant and animal species, many of them endemic. And it is highly vulnerable to climate change and aquatic invasive species. There are currently over 50 native species that are listed under the Endangered Species Act. The federal Central Valley project and the state water projects were designed to regulate and protect against floods and provide water for irrigation and domestic use and generate power for California.

12:43 Speaker 2: Over the years, these water projects increased the water exports from the estuary to agricultural and urban communities across the California state, and these rose to record levels in the earlier part of this decade. The water projects proved to be very successful at expanding agriculture and urban development in California, which is primarily a very arid location in the western United States. But those water exports that build up our agriculture in urban areas in California also came at a huge cost, which was the collapse of California's once famous salmon fishery, the loss of wetlands, the loss of wildlife habitat, the damming of rivers, and the loss of many fish and wildlife species.

13:32 Speaker 2: In recognition of the importance of this place for both supplying water to fuel California's economy, but also the degradation of these ecosystems that are important for the functions that provide those services to people. In 2009, the state passed the Delta Reform Act, which was a legislated mandate to create co-equal goals of providing a more reliable water supply for California and enhancing the ecosystem at the same time. So we are faced with a legislated mandate to both improve the ecosystems for conservation as well as continue to deliver water supply for people.

14:20 Speaker 2: So California is in this place where they're trying to meet these co-equal goals for this really important water hub. while at the same time introducing new mandates to conserve 30 by 30, 30% of California's land and water by 2030 in order to enhance biodiversity and nature-based solutions. So one of the main ways that California is doing this is through its EcoStor investments. The California Eco Restore Program is a multi-agency initiative that was launched in 2015 to advance at least 30,000 acres of critical habitat restoration and enhancement in California's Central Valley, including the Sacramento-San Joaquin River Delta.

15:05 Speaker 2: The overall estimated cost to complete the current list of projects is $ 750 to $ 950 million. So that is about the same cost cap that is present for the surface biology and geology mission that NASA plans to launch in the next decade. So it's a very big investment that the state is putting in to enhancing these wetlands and creating more wetlands. There's a big problem with California's restoration plan, and that's that these biological invasions are going to threaten the benefits that these protected areas can provide, and they're going to cost a lot of money.

15:48 Speaker 2: This is data that is shown from Moodley et al. 2022, which shows that biological invasions are one of the main threats to biodiversity within protected areas worldwide. And the reported economic costs of these invasive alien species in protected areas amounted to $ 22 billion between 1975 and 2020.

16:13 Speaker 2: We can see here that these aquatic invasive species that are the most costly are the group in dark blue at the top. And we can see on the x-axis the total cost estimate of these species in U.S. dollars, and note that it is on the log scale. Of the aquatic invasive species, we can see that there is this Ludwigia genus, which is water primrose. and the Icornea, which is the water hyacinth that we were talking about earlier in the news. And we can see here an example of that Ludwigia or that water permos invading the Sacramento-San Joaquin River Delta and the wetlands therein.

16:57 Speaker 2: Providing better understanding of how these protected and restored areas are being impacted by invasions is critical for informing policy responses and to optimally allocate resources for both prevention and control strategies. So when we think about methods for mapping and monitoring invasive aquatic plants, we have a number of different ways that we can do this. The first thing we can do is we can get in a boat and we can drive around and actually collect very high precision data at very fine scale about the location and the presence of different species in these waterways.

17:39 Speaker 2: However, as we discussed before, one of the primary things that these plants do is impede boat traffic. And so it can be very difficult to access all of the areas that we need to access via boat. And it can be very costly to do this over very large areas and both themselves spread invasive species. The second way that we can monitor these are through remote sensing, and there are three main remote sensing platforms that we can use to map and monitor invasive aquatic plants. There are unmanned aerial systems or drones, and these provide high spatial resolution and on-demand temporal resolution, on-demand meaning you can go fly whenever you want to fly.

18:26 Speaker 2: However, because they provide such high spatial resolution, they cover a relatively small area, and you still have to be able to launch them from a point that is relatively close to the area of interest. So your access is still limited in terms of coverage. The second remote sensing platform that we can use are piloted aircraft. So these are traditional airplanes that we think about. These airplanes can be mounted with very high spatial and high fidelity spectral imaging spectrometers, which provide very high quality data for mapping species.

19:07 Speaker 2: However, it is costly to fly a plane over a large area, which means that typically we don't do this very frequently. And the reality is that it is dangerous. People still die in airplane crashes. And so we are asking people to take risks to collect these data for better management. The final platform that we can use for remote sensing is orbital data or satellite remote sensing. Satellites that are in low Earth orbit provide frequent predictable revisits over the same place in the Earth. But the downside of this is that they typically have low to moderate spatial resolution.

19:50 Speaker 2: So what we're going to do is I'm going to share with you some of the examples of remote sensing of aquatic invasive species in this river delta. One of the things that is very remarkable about this case study is that we have conducted over two decades of airborne imaging spectroscopy. So these are aircraft-mounted imaging spectrometers that have been flying at an annual cadence over the Delta since 2004. This work was pioneered by Professor Susan Houston at the University of California at Davis, where I was very lucky to work with her as a PhD student and have continued to work in this system on and off for many years.

20:33 Speaker 2: An imaging spectrometer is mounted to these aircraft, and they are flown over the Sacramento-San Joaquin River Delta. And what the airborne imaging spectroscopy does is it provides us with an image, and we get all of the spatial data that an image provides. So we can see here on the left this image, and we can see that there is this island in the middle of this waterway that has many different plants, including aquatic invasive species, floating, emergent, and submerged. All of these little white dots that we are seeing in the waterway are houseboats.

21:11 Speaker 2: This image was taken over the 4th of July, and this is a big boat party that's happening out at this wetland. So we can see that this is an important site recreationally as well. But in addition to getting all of that interesting high resolution spatial information from the images taken from the aircraft, because this is an imaging spectrometer, we are also measuring the reflected electromagnetic radiation in wavelengths ranging from the visible all the way to the shortwave infrared region of the electromagnetic spectrum.

21:44 Speaker 2: So what this means is that for every single pixel, You not only have that spatial information combined in the image, but every pixel has the reflectance information as well, which allows you to make quantitative measurements of the environment using traditional spectroscopy techniques. On the right, we see an example of three different pixels, one pixel that contains pennywort, which is a native floating aquatic plant. And then we can see that there is also water primrose. That's that Ludwigia, which is a creeping emergent invasive aquatic plant.

22:22 Speaker 2: And the water hyacinth, which is again that water cancer that we talked about at the beginning of the lecture, which is a floating aquatic invasive plant. And we can see that even though the reflectance signal of all three of these species is follows very typical characteristics of vegetation, including strong absorption in the blue, strong absorption in the red, and strong reflectance in the near-infrared, as well as characteristic water absorption features along the spectrum. Because of the biochemical differences in the plant canopies and the structural differences in the leaves and the orientation of the leaves in the canopy, there are very subtle differences in the spectra that we can pick up when we have very high spectral resolution data captured from imaging spectroscopy.

23:15 Speaker 2: So through funding from the State of California, the Department of Boating and Waterways, and the California Department of Water Resources, airborne imaging spectroscopy has been acquired over this site for the purpose of mapping both these floating and these submerged aquatic plant species dating back to 2004. And over time, there have been a number of different sensors at slightly different resolutions that have occurred. From 2004 to 2008, the high map imager flown by the High Vista Corporation was flown at a three meter pixel resolution over the course of the summer.

23:57 Speaker 2: Then from 2009 to 2014, there was the Great Recession, and there was no funding. And so we were not able to fly these expensive aircraft over the site to map these data. Fortunately, data collection was able to begin again in 2015, and we used the avarice next generation sensor operated by the Jet Propulsion Laboratory. This was flown at a 2.5 meter resolution.

24:24 Speaker 2: From 2018 to 2019, the average next-gen sensor was unavailable, and so we went again with a commercial provider, the HiMAP sensor flown by HiVista. From 2020 to 2022, we went with a different commercial provider called Spektir, who flew an ASA Phoenix at a 2-meter resolution. And then beginning in 2023, we were very excited to start working with the Jet Propulsion Laboratory's newest airborne imaging spectrometer, which is AVERIS III, which was flown at a 2.5 meter resolution.

25:04 Speaker 2: So how do we map these aquatic invasive species? This work has been pioneered for a very long time by a large number of researchers, But the current operational method for doing this is using machine learning to map both species and life forms in the imagery. So on the left, we're seeing an example of an annual flight line mosaic shown in color infrared. And you can see this is a mosaic of about 64 flight lines that come together to image that entire Sacramento-San Joaquin River Delta. During the course of these airborne flights, we send field researchers out in boats and we use both motorboats as well as airboats as pictured here on the right that actually can glide on top of these floating and creeping emergent species in order for us to collect the geolocation of these different species as well as the composition and percent cover and phenological stage and other important characteristics about the plants.

26:15 Speaker 2: This map in the middle here shows the distribution from one of those campaigns of the different points that were collected that are recording the location of these different species. And we've color coded it just for plant life form. So you can see the distribution of emergent, floating, submerged aquatic vegetation and water points that record an absence of vegetation distributed across the image. We are then able to use a machine learning technique called a random forest classification model that takes the data that we have collected and allows us to build a decision tree that classifies the data into a given class.

27:04 Speaker 2: And so this decision tree takes a number of different inputs and it asks whether these inputs or these features resemble the point that we have labeled as a given species, and then it assigns it a class. We create many, many different trees, hundreds of different trees that all might come up with a slightly different answer, and then they vote, and that is our final class. The inputs into this random forest classification model include broadband spectral indexes that many of you will be familiar with.

27:43 Speaker 2: So these are things that are calculated from the imaging spectroscopy such as NDVI or modified NDVI. However, because we have the imaging spectroscopy, we can also calculate narrow band features such as the photochemical reflectance index or the cellulose absorption index. These are hyperspectral indexes that cue in on some of those subtle variations in the spectra that we can see with imaging spectroscopy data. We also put spectral mixture abundance rasters into this classifier, as well as spectral angle rule image rasters and continuum removal across three different locations in the spectrum.

28:32 Speaker 2: So what is spectral mixture analysis? Spectral mixture analysis is what we do to create those spectral mixture abundance rosters that we place into the machine learning algorithm. Spectral mixture analysis decomposes a mixed pixel into a collection of constituent spectra, or we call these end members. And then we can calculate the fractional abundances that indicate the portions of each end member that create the resultant spectrum. On the right here, you see an example of these end members showing submerged aquatic vegetation, turbid water, clear water, floating aquatic vegetation, emergent aquatic vegetation, non-photosynthetic vegetation, and soil.

29:22 Speaker 2: We then can decompose a pixel that is a mixture of all of those different spectra into the different percent abundances of each of those end members. The assumptions that go into this analysis is that we assume that the pixel is a linear mixture of end member constituents. so that we can solve for the relative fractional abundances of each end members. And the other assumption is that all of the end members that are possibly contained in the pixel have been included in the analysis. The result is a rafter that gives you the percent abundance of each of those different and members, and those are what we include in that random forest classification model.

30:07 Speaker 2: We then get to a spectral angle rule image raster by performing spectral angle mapping. Spectral angle mapping is a technique that we use on imaging spectroscopy, where we can imagine that every spectrum can be represented as a unique vector in n-dimensional space where the number of dimensions is equal to the number of bands that you have in your image. The spectral angle can then be determined for every pixel relative to a reference spectra. So we have those end members and we call those our reference spectra.

30:50 Speaker 2: And then we can calculate the distance between every pixel relative to those reference spectra. These values are then assigned in radians to all pixels. And so that is our spectral angle rule image raster that tells us what are the radians from these reference spectra in n-dimensional space if we consider all of these pixels to be unique vectors. You can then actually use this to classify your data because you can come up with a threshold, right? So you can say, that the smaller the angle is between two spectra, the more likely they are to belong to the same class.

31:30 Speaker 2: And so in this example here, we have our reference spectrum T, and we have two unknown spectra, A and B. And if the classifier was trained with spectrum T, then spectrum A is going to be more likely to assign to class T than spectrum B. because the angle between A and T is smaller than the angle between T and B. So we can create a rule image that contains the angle difference between that reference spectra and the pixel spectra, and that is what we use in our machine learning algorithm. The last technique that we use is another common imaging spectroscopy technique called continuum removal.

32:16 Speaker 2: And continuum removal allows us to compare individual absorption features using a common baseline. And what this allows us to do is normalize the reflectance spectra. And so in this example here, we have the reflectance spectra from Coqueline-Clark that shows a white pine reflectance spectrum from the near infrared to the shortwave infrared. And we can see here that there is this dip in reflectance or an absorption feature at 1.73 microns. There's another one at 2.1 and another one at 2.3. So we can fit a convex hull over the top of the spectrum using these localized maximum over a given region of the spectrum.

33:04 Speaker 2: And then we can remove that continuum by dividing the original spectrum by the continuum or the hull. The results of all of these inputs and the machine learning algorithm is that we are now able to create two decades of invasive aquatic plant maps that are classified at the species, the genus, and the life form level. We can see here an example of one of those time series where we have maps that have been created from that machine learning random forest classifier with all of the many inputs we discussed.

33:42 Speaker 2: that show submerged aquatic vegetation classified to the life form level and emergent aquatic vegetation classified to the life form level, as well as water hyacinth in purple that is classified at the species level, pennywort, which is that native floating aquatic vegetation classified to the species level, and water primrose classified to the genus level, as well as water, dry vegetation, riparian soils, and shadows. And then we can start to track how, over time, the distribution of these different classes changes with these different annual snapshots.

34:21 Speaker 2: Overall, across the years, using a reserved set of the field data for independent validation, we calculated the accuracy of these maps, and it ranges between 89 to 92% depending on the year over the time series. And these maps have been really important for enabling many science questions and supporting invasive species management and decision-making in the Delta. So I'm going to go through a couple of examples of some of those applications using these maps derived from airborne imaging spectroscopy.

35:00 Speaker 2: So we have been able to conduct invasion biology and ecosystem engineering studies. In 2009, Maria Santos and colleagues looked at the submerged aquatic vegetation maps, these are the life form level maps, and then started to look at the annual persistence and spread of the different years. And you can start to characterize where there is growth from year to year, as well as where these life forms tend to persist across the years. And that allows us to start to understand what the mechanisms of invasion are.

35:39 Speaker 2: And we found that while there is a general radial spread of this submerged aquatic vegetation, meaning if it's established or it persists in one place, it is likely to spread from that location of persistence. But there are also saltation events, meaning that the propagules are spread quite far and can establish in new places where there was no persistence. And that is likely a result of the transit of these propagules through either high flow environments or by boats. In 2018, Khanna et al. used the very long time series of maps in order to start documenting the spread of water primrose, that Ludwigia, into native marsh habitat.

36:24 Speaker 2: And they showed for the first time that this creeping emergent species, which had primarily been aquatic up until this date, sometime between 2008 and when they started flying the data again in 2015, had substantially expanded into the marsh and had actually started to behave in an amphibious nature, meaning that it was moving into terrestrial and marsh environments and resulting in the mortality of this marsh. thus changing the ecosystem or being a true ecosystem engineer in terms of the structure of these wetlands.

37:00 Speaker 2: We can then use, if we look at this example on the right, we can then use those same metrics of water primrose persistence as well as calculate marsh spatial complexity to start to understand what exactly are the drivers of marsh loss by water primrose. And here we can calculate persistence, and we have a hypothesis that the probability of invasion success increases with the residence time. But we can also think about the effect of the wetland morphological structure or spatial complexity itself.

37:39 Speaker 2: And in an example where you have very high spatial complexity of a marsh as well as very high persisting water primrose, you have many more opportunities for that water primrose to turn amphibious and move upland and into the marsh, resulting in marsh die-off and replacement by water primrose. And this was documented across the Sacramento-San Joaquin Delta. We can also look at how here on the left, the submerged aquatic vegetation is influencing water quality. In this study by Hester et al. in 2016, they took all of these in-situ water quality stations that have been monitoring water turbidity or the clarity of the water for over 30 years, and they calculated a buffer around that station and calculated the percent cover of submerged aquatic vegetation.

38:40 Speaker 2: They then calculated the trend in water clarity and found that water clarity is increasing over the past 30 years in this system. They then adjusted that trend for the submerged aquatic vegetation cover and were able to allocate the portion of that turbidity trend that was attributable to the increase in cover of submerged aquatic vegetation. And what they found is that in very low submerged aquatic vegetation cover environments, most of the trend that is attributable to SAV is very low. It is less than 30%.

39:18 Speaker 2: However, in the higher submerged aquatic vegetation environments, it is likely that in those instances, submerged aquatic vegetation is contributing to the clearing of the waters by trapping sediment and slowing flow.

39:37 Speaker 2: We can then use these maps of submerged aquatic vegetation to start evaluating management efficacy. There have been a number of management interventions in the system to try to meet those co-equal goals of conserving the ecosystem while delivering water for people. One of the things that they did is in 2015, there was a very big drought in California, and they had to install this emergency barrier. It is a rock wall that went across the main river in that stopped the intrusion of salinity from the San Francisco Bay, making it down to the water pumps where water is being sent for drinking and for agriculture.

40:22 Speaker 2: And this salinity, this barrier stopped the salinity, the highly saline water from entering the system and propagating to the pumps. But one of the things that Kimmerer et al. demonstrated in 2019 is it also seems to have changed the hydrodynamics of the system and resulted in a very large increase in the density and spread of submerged aquatic vegetation, particularly in this large lake-like site called Frankstrakt, which has then served as a major source of propagules for the further spread of submerged aquatic vegetation across the system.

41:01 Speaker 2: Just last year in 2023, if we look at this example on the right, Khanna et al. were able to use these maps as well as use three-dimensional hydrodynamic models to ask what the effect of the fluoridone herbicide treatments are on submerged aquatic vegetation. And what they found is that these herbicide treatments that the state is applying every year to try to keep the submerged vegetation under control are not effective in places where water currents are high. So what we see here on this graph is on the x-axis, we see speed that is taken from that three-dimensional hydrodynamic model.

41:45 Speaker 2: And on the y-axis, we see the probability of submerged aquatic vegetation presence. And we can see three different treatment examples here. We can see the black line is when you have no fluorodone present. And so that's just imagine that we're not using any herbicide application at all. And in general, you see that submerged aquatic vegetation presence declines with current speed. When you apply fluorodone, one of the things you see is that at very low current speeds, this is what the pink line and the green line are showing, the mean concentration of fluorodone and a very high dose of fluorodone.

42:27 Speaker 2: we see that it does indeed decrease the probability of submerged aquatic vegetation presence at low current speeds. However, at high current speeds, it seems to be not making any difference relative to the control site or that site with no treatment whatsoever. Additionally, they were able to use the time series of data to find that if you applied herbicides for multiple years in a row at the same place, it doesn't have any kind of long-lasting effect in reducing submerged aquatic vegetation cover, which means that it might be time to start rethinking how these herbicides are applied in the system.

43:08 Speaker 2: So, we have these data, and as I said before, they have been acquired at between 1.7 and 3-meter resolution. But a lot of people would argue that maybe we want to find even finer resolution data in order to help us find where those invasions are first starting. Because if you can find where the invasion is first starting, you have a better chance of potentially controlling its spread. So one of the questions that we asked is, can we improve the spatial resolution for mapping using UAS or drone remote sensing?

43:44 Speaker 2: This is a study by Bolch et al. from 2020. where they flew an imaging spectrometer mounted on a drone at the same time as those airplane, those piloted airplane overflights were occurring. And then they used the same classification process, the same workflow from the airborne data applied to the UAV data and mapped the aquatic invasive plants. So on the left, you can see the result of that mapping over the two areas that the drone flew. at 1.7 meter resolution with an overall accuracy that year of 90%.

44:22 Speaker 2: The headwall nano-hyperspec, which was the sensor that was flown on the drone, was able to, using a similar workflow, was able to achieve an accuracy of 94%, and we can see that it had a much finer resolution of 0.05 meters. And what this allowed us to do, if we compare these two maps, is we can see that they have comparable accuracy, but we see much finer spatial detail about the distribution of these different species and life forms of concern. So while using UAVs are great because you get very high spatial resolution coverage, there are operational considerations that we need to think about.

45:11 Speaker 2: So the primary capability here that drones offer is that we can detect these rare and sparse classes that are occurring in very small patches that we might miss with a one-meter or a three-meter pixel. However, if we look at this table down below that was presented in BULT-IT-ALL, we can see here the sensor, which is HiMAP. That was the HiVista commercial imaging spectrometer flown that year. The next line shows the drone, the nano hyperspec sensor mounted on the drone from the two areas that were covered in the study and then extrapolated to the entire area of the delta.

45:52 Speaker 2: And so it takes approximately 16 hours to fly 74,000 hectares with the HIMAP piloted aircraft. And it took about one hour to fly 10 hectares with the drone. If you were going to fly the entire Delta using the drone, it would take approximately 7,000 hours. That is quite a lot of labor to fly this entire region using the drone. Furthermore, if we think about the data volume, the imaging spectroscopy at 1.7 meters from the high map sensor produced about 700 gigs of data. The nano-hyperspec, because those pixels were so much smaller on the drone, produced almost a comparable amount of data, 600 gigabytes, for just two very small regions.

46:47 Speaker 2: And if we were to fly that entire region with the drone, we'd be looking at two petabytes of data, which is quite a lot for somebody to handle when they're doing the mapping. We can then extrapolate what the approximate deployment costs were for both the airborne data based on the contracts with the commercial vendor, the estimated costs of the students collecting these data, versus what it would cost to collect these data over 7,000 hours just from the labor expenses and travel expenses alone. We've talked about the high spatial and high spectral resolution snapshots that we can collect from both UAS as well as airborne data.

47:34 Speaker 2: But one of the things that orbital data provides us is that ability for frequent predictable revisits. So can we use orbital data to improve the temporal frequency of these maps?

47:48 Speaker 2: So when we think about orbital data, one of the things that we need to think about are some sensor resolution trade-offs. Here we are looking at three common sensors, or three sensors that we evaluated to use in this study. On the left, we have the airborne hyperspectral data, or the imaging spectroscopy, and this is shown at 2.5 meters. And you can see here that we have that very highly resolved spectral signal of vegetation that is characteristic of the imaging spectroscopy. In the middle, we have a very common multispectral satellite that is used for mapping called Sentinel-2.

48:26 Speaker 2: And there are two of them. There's Sentinel-2A and Sentinel-2B that combine to provide a temporal frequency of five days. The data are delivered at 10 to 20 meters, depending on the post-processing that you do for your data. And we can see here that it has a, because this is a multispectral instrument, it has many fewer spectral bands which allow for a much coarser representation of the spectra in the image. And on the right, we have a very popular sensor called Landsat that provides data at 30-meter resolution with a 16-day orbit for Landsat 8 and another 16-day orbit for Landsat 9.

49:07 Speaker 2: And again, we find that there is a lower spectral resolution and spatial resolution with Landsat. So given these considerations, we opted for Sentinel because we could post-process it to a 10-meter resolution, which still gave us sufficient resolution to see most of the waterways and vegetation that we were interested in and provided that high repeat frequency. So if we think about those temporal gaps and Sentinel-2's ability to fill that in, Sentinel-2 has a repeat frequency of five days between the two satellite sensors.

49:47 Speaker 2: Here, what we are looking at on the x-axis is time over the course of a year from January to December. And we can see this is submerged aquatic vegetation. This species is called Egeria densa, what its typical phenology is over the course of the year. And on the y-axis, we have year going from 2019 to 2016 from top to bottom. And these red marks show where those annual hyperspectral images were collected. And so because they're just annual snapshots, they're missing some of the key stages in the vegetation growth cycle.

50:25 Speaker 2: If we consider the five-day repeat cycle from Sentinel-2a and Sentinel-2b, And we then look at the data availability across both the season on the x-axis and the year on the y-axis. And we account for the cloud coverage because you can't see through the clouds to see the vegetation. You can see that even though there are gaps, particularly in the winter and spring when there is more cloud cover, there is still much denser images that allow us to better capture the growth cycle of these vegetations.

51:02 Speaker 2: So we used the training data from the imaging spectroscopy with a machine learning model to map the invasive aquatic vegetation from Sentinel-2, and we compared that to the maps with the imaging spectroscopy. Now, because the Sentinel-2 data is multispectral and it doesn't have high-resolution spectral data, We were only able to get to these life forms of submerged aquatic vegetation emergent, but we were able to distinguish the difference between water permeros and water hyacinths, which are those two, the genus and the species of very high concern in this region and in many regions across the planet.

51:47 Speaker 2: With Sentinel-2 data, we can see that's at 10-meter resolution versus the imaging spectroscopy at 1.7-meter resolution. The overall accuracies are relatively comparable between the two maps, 89% for Sentinel-2 and 90% for the imaging spectroscopy, with class-specific accuracies for water hyacinth and water primrose at 78% and 85% for water hyacinth compared to 94% and 89% for water hyacinth in the imaging spectroscopy. And water primrose was 94 and 92 versus 95 and 95. We can see that both of the maps are capturing the same general patterns of the presence of these different species of concern.

52:36 Speaker 2: However, because of the finer resolution of the imaging spectroscopy, we can see more of those smaller patches of water hyacinth and where they are competing with the water primrose and where the water primrose is encroaching into that emergent marsh. So one of the things that we're able to do now that we have a reasonably accurate map from the Sentinel-2 data is because we have the high temporal frequency, we can track the phenology of these species over time. This is work that's led by my former PhD student, Christiana Ade, who used the TimeSat software package in order to calculate phenology metrics.

53:23 Speaker 2: So the first thing that she did is she calculated the enhanced vegetation index for every image date in the time series from 2018 to 2022. And then she fit a Gaussian asymmetric fitting function to extract phenology metrics. And she did quite a lot of work because the Sentinel-2 is an irregular time series, and these data are designed for AVHRR or MODIS. It was originally intended to handle, again, daily data, not five-day data, so she had to do some manipulation. However, one of the things that's very exciting for this community is that TimeSET is a free and open-source software that and they are going to be releasing a new version soon that will specifically be able to handle Sentinel-2 data, which will mean that there will be a lot less data manipulation that other users have to do before they perform these analyses.

54:25 Speaker 2: TimeSat allows you to derive different phenology metrics, including the start of the growing season, the end of the growing season, the length of the growing season, as well as the rate of the increase and the rate of the decrease of these species. So what we see here is because we have the enhanced vegetation index calculated for every pixel and we know the class of every pixel in our image, we can now compare the phenology and these derived phenology metrics for these different classes. Here what Dr. Ade has done is she has shown for three different pixels what the phenology curve using the enhanced vegetation index is for water primrose in green compared to water hyacinth in purple and that native marsh emergent vegetation in orange.

55:18 Speaker 2: And we can see that there are substantial differences not only in the magnitude of the enhanced vegetation index calculation, but also where the start of the season, the end of the season, and the length of the season are. So what Dr. Adet has done down below is just taken three different sites and looked at the distribution through these violin plots of the start of the season and where that occurs in the day of the year. And one of the things that we can see is across these three different sites and across these three different years, water primrose starts greening up much faster than water hyacinths or the native emergent marsh vegetation does.

56:05 Speaker 2: We can also view that spatially, and we can see spatially where the start of the season is for different sites across the delta relative to where the classes are. And if we look at this example here in Rhode Island, we can see the class map on top with the water hyacinth in purple and the water primrose in yellow and that emergent marsh vegetation in this green. And then we can look at when the start of the season is. And the more warm or red and orange the colors are, the later in the season the growth is.

56:45 Speaker 2: And the darker the green, the earlier in the season the growth is. And we can see that water primrose seems to have a distinct advantage in greening up sooner than both the native emergent marsh as well as this other invasive species, water hyacinth, which may explain why it is out-competing or appears to be out-competing water hyacinth in this system. This also has very strong implications for the management of these species. This is from a paper by Wolkovich and Cleland in 2011, which shows that we have these different phenologies where we have early going to late in terms of relative abundance and a number of different hypotheses related to how invasive species are behaving in the environment.

57:39 Speaker 2: And there are a number of different hypotheses around when they're going to be successful as described in this paper. But one of the key things that they've also done is they've mapped these different phenology effects between these different species to target specific management opportunities. For example, we saw in the previous example that maybe priority effects are in account where we have one species that is greening up earlier than another species. which means that a targeted removal earlier in the growing season may be a factor to have more successful management of these species.

58:21 Speaker 2: So

58:25 Speaker 2: that is the end of this case study. And now I'm going to hand it over to my colleague, Justin, who is going to provide a summary and next steps.

58:36 Speaker 1: Thank you, Erin. So now I'll go over a summary and some things to look forward to. So invasive aquatic plants modify the environment, change ecosystems, and threaten biodiversity. Invasive aquatic plants have significant impact to human health, likelihood, and economics. Remote sensing can provide large scale, repeatable, and consistent monitoring of vegetation functional types and invasive genera and species. Airborne imaging provides high-quality, high-resolution data for baseline maps and monitoring.

59:12 Speaker 1: Drone data is infeasible to deploy across large scales, but useful for sampling at high-priority sites and monitoring targeted interventions. Satellite data, coupled with machine learning, can miss new invasions and small patches, but fills in temporal gaps for phenology and invasion dynamics. Field surveys, where they're possible, provide a critical link between and across datasets and scales and quantify uncertainty for managers. So, looking ahead to Part 3, Part 3 will cover the monitoring of invasive grassland species with hyperspectral remote sensing.

59:47 Speaker 1: We hope to teach you how to identify basic considerations for the application of hyperspectral data for mapping invasive plants. outline the key benefits of remote sensing for mapping invasive plants compared to field-based techniques, and identify the limitations of remote sensing for mapping invasive plants. Again, as a reminder, there will be one homework assignment which will open on the 28th. You can access that from the training webpage, and you'll submit it via Google Forms. It's due by 9-11-2024.

1:00:20 Speaker 1: And to get that certificate of completion, You'll need to attend all three live webinars, complete the homework assignment by the deadline, and then you'll get your certificate via email approximately two months after the completion of this course. Listed here is our contact information, as well as our website, our social media, and links to our sister programs, Develop and Survear. Of course, I would also like to acknowledge and thank Aaron, as well as all of the people on the RSAT team that have made this training possible.

1:00:58 Speaker 1: Here are some resources that you can check out on your own time, as well as access to some of the data in TimeSat software. And Aaron has provided some references, which are relevant to the presentation that you all just saw. And of course, thank you all for attending. And now we will move to the question and answer session. So please put your questions in the Q & A box.

1:01:30 Speaker 2: Thank you. All right, everyone.

1:01:37 Speaker 1: And hopefully you can see the question and answer document. We'll start at the top. And again, if we don't get to your questions today, we will answer this and it will be posted on the web page. So don't worry if we don't get to your specific questions. So question one is a holdover from our previous session. Is there a way to search by species to see which tools might have data associated with that species rather than searching in each of the platforms for that species? So the definition of invasive species is dependent on the specific ecosystem and context.

1:02:12 Speaker 1: Remember, all invasives are native to somewhere. So that makes it challenging to create a universal remote sensing database for all species. Remote sensing plant species vary significantly based on the ecosystem and the distinctiveness of the target species compared to other plants in the area. The ability to differentiate a particular species relies on its unique spectral characteristics relative to the other plants and whether the variability between different species is greater than the variability within the target species itself.

1:02:41 Speaker 1: Additionally, differences in phenology are often utilized to aid in distinguishing these species. For more information, you can check out the article that we've linked in the Q & A doc, which, again, will be made available to you. How does the water hyacinth cause water loss via evapotranspiration? This is an interesting question. So as part of the plant's biologic process, water is taken into the plant and released into the air. This causes more water loss than if the open water channel were simply exposed to the air without the influence of plants.

1:03:14 Speaker 1: I've included a link to a NASA resource about evapotranspiration, but ultimately you can think of this as a function of the surface area and sort of a forced evaporation where the plants take in the water and force it into the air as part of their biological process. What methodologies or indices would you recommend for effectively assessing the impact of these aquatic invasions? Specifically, How can we accurately measure and monitor the extent of this environmental disaster given the limitations of traditional in-situ approaches?

1:03:51 Speaker 1: As I believe Erin covered excellently in the presentation, this is an example of where remote sensing really provides that leg up that we wouldn't have with in-situ approaches as these invasive species, particularly in aquatic environments, lead to a lack of access. Using remote sensing gives us a way of looking down at these things without having physical access to the site. Question four. How can environmental factors such as temperature and precipitation influence the predictive accuracy of remote sensing tools?

1:04:30 Speaker 1: Also, when integrating machine learning algorithms with remote sensing tools such as Sentinel-2, what are the limitations for monitoring invasive species? So temperature and precipitation are commonly recorded with remote sensing and can be accounted for in predictive models. The water column can significantly affect the remote sensing of submerged aquatic vegetation due to several factors that alter the spectral signals detected by sensors. The key effects being light attenuation, wavelength-specific absorption and scattering effects due to variations in water quality, water surface reflectance, that's sun glint, water depth, and nonlinear spectral mixing.

1:05:09 Speaker 1: As you recall, the assumption in the spectral on mixing was that the in-members are a linear mix. And I see that there's a missing link that we will get in there. Specific to machine learning, machine learning is useful in monitoring invasive species. See slides 31 and 32 from the presentation, which again will be on the web page. for more information on classification with the random forest method. And limitations will always be specific to your environment, your species of interest, the data set that you're using, and the model you choose to use.

1:05:47 Speaker 1: Question five. Will we get the reference spectra from invasive species worldwide or on a regional basis? Is there any library that is open access? So one of the most comprehensive spectral libraries in vegetation is the Open Access Ecosys database. So we encourage everyone to contribute to the database to get more aquatic plants represented and the USGS also has an open access spectral library, but with fewer species. Question six. If you set the class's number to the amount of species known, how does the classifier do?

1:06:26 Speaker 1: When using a supervised classification algorithm and setting the number of classes to the amount of known species, the classifier's performance varies depending on several factors, including the spectral separability of the species, that's the distance between their spectral signatures, the quality and resolution of the spectral data, and the algorithm's parameters. If the spectral signatures are similar, the classifier might struggle to differentiate between species, which leads to misclassification.

1:06:54 Speaker 1: By setting the number of classes to the exact number of known species, you're assuming that each species has a unique spectral signature. If that assumption holds true, most classification algorithms can effectively classify the species. However, in cases where the species have overlapping or very similar spectral signatures, the classifier might not be able to distinguish them accurately. If the reference spectra are well represented and capture the variability with any species, the classifier's performance will improve.

1:07:24 Speaker 1: On the other hand, poor or unrepresentative spectra can lead to incorrect classification. The spatial and spectral resolution of the data also plays a role. High-resolution data that captures fine spectral differences can enhance the ability to distinguish species. Conversely, low-resolution data might obscure those differences, leading to decreased accuracy. In the case study, high-resolution data was utilized to achieve certain results. If we were to use Sentinel and Landsat data for the same study, what level of accuracy could we expect?

1:08:00 Speaker 1: How would the resolution and data quality of Sentinel and Landsat compare in terms of achieving similar outcomes? I believe this was directly addressed, but just for the sake of having this as a resource, in the case study shown, Sentinel-2 was successfully used with slightly reduced accuracy and fewer forcer classes. However, because of the 10-meter pixel size move, we missed several small patches of invasive plants. We determined that Landsat with 30 meter pixels was not capable of resolving patches in our system and resulted in both low accuracy and very poor maps.

1:08:37 Speaker 1: Question eight, is there any concern of a different species not initially determined that may skew results of detecting in members or any other spectroscopy conditions? Yeah, so that's certainly a concern. And because plant invasions and degraded landscapes are highly dynamic, and you may continue to have variations in community composition and turnover over the space of in the Delta, they found that it was critical to continue to conduct field surveys to support validation of the work. Remote sensing is a complement to field surveys.

1:09:11 Speaker 1: It allows us to fill gaps that both surveys can't fill, but remote sensing is not a complete replacement for field observation in these dynamics. Question nine. It seems the presentations and the webinar deal with water invasive species. What about the inland invasive species? Are there any studies? Would appreciate if you could share some details on the same. I have great news. Part one was about the general overview of invasive species monitoring and remote sensing. But part three, which is upcoming, deals specifically with grassland invasives.

1:09:48 Speaker 1: So we've got you covered.

1:09:49 Speaker 2: On that one.

1:09:52 Speaker 1: With which GIS images and scales have you carried out these investigations? These studies were conducted using imaging spectroscopy images, so that's all of the listed sensors there, and multispectral satellite imagery, Sentinel-2. The delta is approximately 75,000 hectares, and the software used varies from study to study, but key software includes Envy and IDL. Python and R. And speaking personally, I am excited that we're going to be doing even more with some coding techniques in RSET. So look forward to those.

1:10:34 Speaker 1: Question 11. Is it possible to calibrate remote sensing-based water quality monitoring data using in-situ water quality data with the help of a relative index from empirical mathematical equations? If so, how can the relative index be validated for each water quality parameter? Yeah, so we have a whole list of courses using remote sensing for water quality. So when this is available, click on those links, go and check that out. Question 12. Do you guys have a table for which platform is most suitable for specific species mapping, monitoring, and for which scales?

1:11:14 Speaker 1: Unfortunately, I'm not aware of any database that fits that description. It would be great if it did exist, but you have to imagine that you would have to consider an incredible number of species, hundreds of platforms and sensors that each have their own technical specifications, and the geographic and environmental differences, which may make one sensor more appropriate for your case. Furthermore, that wouldn't be able to consider any customizable sensor and platform combinations, such as the UAVs that were used in the water hyacinth study that was discussed today.

1:11:46 Speaker 1: So that would be a tall order, but if somebody could put that together, that would be a huge boom. Question 13. When you're calculating costs, are you including any time for ground truthing efforts? How much of your tests using the different imaging require on-the-ground surveys for QAQC per The costs shown in the case study in that table are just for the image acquisition, and they don't include surveys for validation. In these studies and working with managers in the Delta system, field surveys are a critical component to quantifying the uncertainty of the maps and help to recalibrate machine learning models under highly variable conditions.

1:12:26 Speaker 1: Again, you can see the answer to question number eight about the dynamics environment that you have in invasive degraded ecosystems. Question 14. What are the transferable aspects or concepts approaches from this case study that would apply in other contexts? That is, what can we generalize for use elsewhere? So the trade-off between platforms and resolution considerations are something that every case study needs to evaluate for their own context. We've articulated the trade-offs, and now you can evaluate these trade-offs for your own study purposes.

1:13:06 Speaker 1: The general frameworks of analyses and the multiple tools we have presented are widely available in open source software such as Python and R, and may be useful for your studies. Additionally, we hope the management studies conducted with maps give users an idea of the types of analyses that they could conduct once they have their own maps developed.

1:13:28 Speaker 2: And let's see.

1:13:30 Speaker 1: Skip a couple so that we can answer these in time. Is TimeSat freely available? Yes, we have a link to it, so go check that out. Question 17. Do the vegetation indices change in different environments? Does it need to be calibrated for each region? For example, the information from the Bay Delta showing the primrose and hyacinth, will that same imaging taken in North Carolina show the same plant species from the satellite images? We have a webinar entirely dedicated to spectral indices, so you can check that out as well.

1:14:07 Speaker 1: Short answer is yes. Would this method be able to detect duckweed separately? So using imaging spectroscopy, we've been able to detect duckweed in the California Delta case study. The ability to detect duckweed will depend on the size of the patch of duckweed. the density and the ecosystem context. See answers to questions number two and six for more details there. Question 20, is there a rule of thumb for the maximum reasonable area of coverage for UAS high priority sites above which it is better to use manned flights?

1:14:50 Speaker 1: So when deciding between UAS and manned flights for covering high priority sites, the choice depends on several factors. including the area to be covered, the mission's objectives, the terrain, and the specific capabilities of the UAS. There isn't a strict rule of thumb universally applied, as is the case with many of these, but there are general guidelines that can help inform the decision. So drones are ideal over small areas for high resolution data collection and detailed surveys. They're flexible and effective for these areas, and UAS could be preferred for high resolution low altitude data.

1:15:27 Speaker 1: Manned aircraft, on the other hand, are more efficient for covering large areas quickly with greater flight endurance and range. Manned flights are, however, suitable for broad area surveys where a slightly lower resolution is acceptable. So you'll need to consider flight endurance and range. Drones are limited by battery life and range, generally covering smaller areas in a single flight. whereas manned aircraft are capable of covering larger areas in one mission due to longer flight times and higher altitudes.

1:16:01 Speaker 1: Question 21. Have there been any biocontrols explored to tackle the issue of invasive aquatic plants? Yeah. In the Delta, they tried to introduce weevils to help control the water hyacinth, but with mixed success. There's a blog post linked here that tells the story of the experience. Why EVI for phenology? There are many indices that can be used for phenology studies, but in this case study, we evaluated several indices and performed a sensitivity analysis to determine how variable our results would be to our selection of the indices.

1:16:39 Speaker 1: It's recommended that you select an index most appropriate to your study and conditions. Again, refer back to question 17 for that link to our dedicated training on spectral indices. And I will skip to the ones we have answered already, and then we'll come back if we still.

1:16:59 Speaker 2: Have more time.

1:17:02 Speaker 1: Are HMAP and ASA free for download?

1:17:07 Speaker 2: No.

1:17:08 Speaker 1: AVRIS is open access and free for download. HMAP and ASA are commercial and have to be contracted to be flown over your study region. So of course you have some cost considerations.

1:17:21 Speaker 2: There as well.

1:17:25 Speaker 1: Okay, jumping back up, speak sort of extemporaneously here. Could I access airborne magnetic gravity and spectrometry data for GIS mapping and interpretation purposes? I don't know of magnetic, but I know there was a NASA study, really fascinating technique, using a pair of satellites that did map the Earth's gravitational anomalies. So you can look into that. I'm not remembering the specific name. I believe it might have been GRACE. But yeah, so some of that data is available. The rest, not too sure about.

1:18:15 Speaker 1: Seems like you're getting into some interesting work. How can I validate the real-time monitoring data generated from mobile camera integrated with a remote sensing-based ML model to ensure it accurately reflects invasive species presence? So I am not aware of remote sensing using a smartphone camera, but since you specifically mentioned ML, I do know that there are models that use computer vision to identify species that are relatively robust. So if you're using a smartphone with one of these computer vision libraries, the one that comes to mind is Seek.

1:19:11 Speaker 1: by iNaturalist. It can assist you in identifying a species to figure out if that's something of concern or not for your area. And let me see if there was anything I skipped further up. I think we're good to continue on further down this list.

1:19:36 Speaker 2: Is there any.

1:19:36 Speaker 1: Association or club I can join? Question 23. Is there any association or club I can join to stay connected with people having interests related to these sessions? There are a lot of user groups. So you're going to probably want to start looking around your area for people who are already meeting. I know in my local area, there's a user group specifically for coding for GIS. So that is a great place to start. Also, online communities and the people who left their contact information in the chat. Obviously, if they're attending this webinar, they have some interest in remote sensing and invasive species.

1:20:25 Speaker 1: So all of those great places to start. Question 26, how can I identify invasive species in a kelp forest and seagrass meadow? Which sensor is recommended to use for optimal results? As with the answers to many of these questions, this is going to be context specific and going to be specific to your study and your area of interest. So I can't give a blanket recommendation for which sensor would be optimal. but you're going to be looking for something that fits your needs in terms of the spectral, spatial, and temporal resolution.

1:21:15 Speaker 1: I don't know if Aaron, nope, it looks like Aaron is answering this live. So yeah, please refer to that answer while I move on. Yep, our training on monitoring aquatic vegetation. Question 27 seems to be similar to question number four, so we will just refer you back to that answer.

1:21:45 Speaker 2: And those trainings.

1:21:55 Speaker 1: Let's see. With which images have the investigations obtained this high spatial resolution? Yeah, high spatial resolution is going to be low altitude, so you're looking at things like UAVs, UAS.

1:22:17 Speaker 1: Are there any practical exercises that could be shared for reproduction using, for example, Jupyter notebooks?

1:22:25 Speaker 2: We....

1:22:28 Speaker 1: Figurbrock, yeah. So not specific to these studies, but in the future, please look forward to a more advanced remote sensing of invasives coming up in the next year. We're considering doing a more hands-on approach to the same topic so that we can give people a feel for the tools.