Mapping Salt Marsh And Seagrass With Earth Observations

Speaker 1: Welcome back to our RSAT training series, Earth Observations of Blue Carbon Ecosystems. Today is part two of our training series, Mapping Saltmarsh and Seagrass with Earth Observations. Today, in part two, we will highlight saltmarsh and seagrass ecosystems and see demonstrations of mapping saltmarsh and seagrass with Earth observations. There will be one homework assignment for this training, and it is available now on the training webpage. It'll be due two weeks later on December 19th. A certificate of completion will be awarded to those who attend all live sessions and complete the homework assignment before the given due date.

Speaker 1: Once again, my name is Brock Blevins, training coordinator for the RSET program, and I'll be your host. We are also joined today by Dr. Anthony Campbell, an assistant research assistant at NASA's Goddard Space Flight Center. Dr. Kelly Lewis, a scientist in the Water and Ecosystems Group at NASA's Jet Propulsion Laboratory in Pasadena, California, and Cheryl Doughty, an assistant research scientist at NASA's Goddard Space Flight Center. By the end of Part 2, participants will be able to map the extent of salt marsh and seagrass ecosystems using satellite observations, calculate the carbon stocks of mapped salt marsh and seagrass ecosystems, and explore synthesis methods to Estimate blue carbon across ecosystems.

Speaker 1: Once again, if you have any questions during today's training, you can put them in the questions box within WebEx at any time. At the end of this training, we'll try to address as many questions as we can, and we'll post these questions and written answers to the training webpage within a week after the training. And now I'll send this off to Anthony. Thank you.

Speaker 2: For that introduction, I'm going to be talking now about what is a salt marsh and salt marshes in general for mapping them as a blue carbon, important blue carbon ecosystem. Salt marshes are coastal wetlands. They compose the blue carbon trio that we've discussed previously of mangroves, salt marsh, and seagrass. They are found globally, but highest concentration are in temperate regions of the world, including the highest concentration of them in North America. They have some ecosystem commonalities across salt marshes.

Speaker 2: They are high salinity environments. So a alicetic vegetation or vegetation that is adapted to and tolerant of salinity is most commonly found within these salt marshes. And they're tidally inundated. So whether it's daily or less frequently, they're going to be inundated by tidal waters and that's going to bring nutrients and flush the systems, but it's also going to create an anoxic environment for time periods that these systems are adapted to. And they're also low energy environments. This means that there's not a whole lot of wave energy impacting them for the most part.

Speaker 2: Some storms, during storms and other things, obviously waves do impact salt marshes and cause erosion. But these are areas where sediments are going to settle out onto a tidal flat and slowly accrete over time, potentially vegetation will colonize that area and it will become a vegetated salt marsh environment. So salt marshes also, as this training suggests, they're one of the blue carbon ecosystems. So they're carbon dense ecosystems and most of this carbon is held within their soil organic carbon, which is accumulated for thousands of years and can be a very dense and deep carbon stock.

Speaker 2: There's also carbon in the above ground biomass and below ground biomass of salt marshes. These are important for assessing their health and long-term viability of an area or region. There's also other important ecosystem services as salt marshes. They provide storm mitigation and flood mitigation. They also provide adaptation to sea level rise. These systems are very good at collecting sediment and accreting and keeping pace with things like sea level rise to a certain degree. They're also important as habitat for a variety of fish and bird species, and they have a certain aesthetic quality that is useful for humans and beneficial to humans or we enjoy, you know, whether we're cracking through them or using them to hunt for bird species or something like that.

Speaker 2: They have a variety of uses for the people that live in close proximity to them. And now we're going to do a TLS fly-through collected by Advocate Stovall in South Africa. And this is just showing the variability of the vegetation across the marsh. You can see kind of little divots and changes in microtopography of it. There's also some shifts in grasslands and salicornia species that are evident. The kind of divots are where the scans occurred, so disregard them. And now we're transitioning into some juncus species, so some reeds and sedges, and then some bushes and shrubs.

Speaker 2: And then finally, a Phragmites australis patch, the common reed patch, and then we zoom off into the tidal mudflats. So site being a very small transect around 300 meters, you have a lot of variability in the vegetation types and structures across, despite it all being kind of considered salt marsh environments. So when we are doing these global maps of salt marshes or regional maps, we have to kind of consider that high variability. And these are some examples of successful mapping projects and data sets that exist for salt marshes.

Speaker 2: One thing to note here on this slide that Some of these data products are for tidal marshes. Some include salt marsh. For this presentation, we're going to kind of use tidal marshes and salt marshes interchangeably. But the reason for this and the reason for our discrepancy between these is tidal marshes are kind of a more all-encompassing term. Salt marshes, it's going to be very hard to differentiate salt marsh from freshwater marshes with just Earth observation data, since this is a salinity difference.

Speaker 2: Without in-situ salinity data or even vegetation type data, it becomes very difficult. So some research has just focused on mapping all of these kind of together as tidal marshes, and some have focused on differentiating them as best as possible, such as the third data set there, the global 30-meter map of wetlands. That includes the salt marsh category. The two data sets above it at 10 and 30-meter maps of tidal marshes do not include a salt marsh, a separate salt marsh category. So a comparison of those would be interesting and something for people to do in the future for sure as we get closer to maybe being able to map salt marshes as a standalone product or we improve that mapping.

Speaker 2: We also have the global 30-meter data set of change in emissions. This is a carbon emission data set that I created a couple of years ago that looked at the four epochs from in 2000 to 2019, five-year epochs and determines how much carbon emissions happen from salt marsh change across that time period. We also have a new global soil organic carbon map. The manuscript is as a preprint, but the data is currently available. And this is a global prediction of soil organic carbon at 0 to 30 centimeters and 30 to 100 centimeters depth.

Speaker 2: It does have some areas of tidal marsh that are excluded if the predictive power of the model was not high enough for that region. So it does show kind of where we need to get more in-situ data to improve our maps of soil organic carbon, which again is the highest concentration of carbon in these ecosystems. And then finally, at the last bullet point, we have the code that was utilized in the second bullet point mapping approach. This is a Google Earth Engine approach for mapping tidal flats, tidal marsh, and mangroves with an interval mean kind of covariate creation.

Speaker 2: So if that's something interesting, an advanced method of mapping these systems, we're going to talk about kind of a slightly simpler approach that you could use for regional mapping that also kind of gives us a better understanding of what's successful and what needs in sense mapping salt.

Speaker 1: Marshes there are.

Speaker 2: So this starts with the sensors that we use. Obviously, all the data sets we just talked about, they're using Sentinel-2 or Landsat archive data. So that's multispectral data. Data that can be incredibly successful at mapping salt marsh extent and change over time. And that's demonstrated in all those previous products. So obviously the spatial resolution of Sentinel-2 and Landsat is 10 to 30 meters. And that resolution has been demonstrated to be appropriate. But you also get some additional precision when you have higher spatial resolution data products with marshes.

Speaker 2: And we'll look at an example of that. That can be beneficial. These systems can change relatively rapidly, and there's also small landscape features like creeks or even just shifts in vegetation, as we saw with that CLS flyover, fly through. Temporal resolution is also incredibly important, and we're talking about the revisit time here. So if you have a single scene, you're at the mercy of the tides and the seasonal conditions to map your salt marsh environment. But if you are using multiple images, you can understand the phenology of these systems As they are temperate ecosystems, they do have a pronounced phenology that you could map, but you also have to be considering that they're being tidally inundated, and that's going to influence the phenology and the spectral signatures of these ecosystems, depending on the inundation of them.

Speaker 2: And we'll look at that in detail. And then finally, spectral resolution As we talked about, multispectral images are highly appropriate for these. We'll kind of look at what additional spectral resolution might mean for mapping salt marsh environments. And finally, elevation can be an incredibly important parameter. Obviously, this is coming from either synthetic aperture radar datasets like the SRTM, or TandemX, or LiDAR data sets in many countries might be available. And this can serve as a nice, soft clip to your salt marsh extent.

Speaker 2: So above a certain elevation, you're not expecting any more tidal inundation, even in extreme events. So it's very unlikely that salt marsh could be inhabiting that elevation range. So that can be important. And we're going to, as an example site, we're going to talk about nice estuaries. in South Africa, in Estuarine Bay, with a temperate climate. We conducted field work here as part of the Bioscapes field campaign here and other sites in South Africa last year in 2023. This is a microtidal estuary.

Speaker 2: And it also has multiple tidal cycles per day, so two low tides and two high tides. It's semi-diurnal. When we look at the images here, you can see some of the variability across the estuary and other sites The first image kind of shows a typical intertidal salt marsh plot that we've looked at. The image next to it that's labeled Banjpan Lagoon shows a mosaic of seagrass and salt marsh species co-occurring together. And it's one of those complexities when you map these as discrete categories in a blue carbon accounting, the reality is they do overlap.

Speaker 2: And you have to account for that and understand what that might mean for the carbon account. In the bottom here, we have a patch of Phragmites, just to kind of demonstrate that despite this being less than 100 meters from the image above it, there's quite a bit of difference in vegetation structure. And we are mapping both of those as salt marsh environments. So in order to do that, we need to be considering the wide range of spectral and structural properties that we might be mapping together. Okay, so sensor properties, the spatial resolution here that we're discussing.

Speaker 2: So we see the spatial resolution is another important parameter as we discussed. And in this example here, zooming out to the southern portion of the Nisena Bay, we see if the 100 meter data set aggregating up from Sentinel-2 data does not look particularly, like it's particularly captured the characteristics of the marsh in this area. And that could be, do a variety of things in their mapping approach, but perhaps the tidal inundation has led to a more significant impact on mapped extent there.

Speaker 2: But the tidal marsh, the 10-meter dataset, the global tidal marsh dataset, does a pretty good job of capturing the spatial extent of the salt marsh environment here. But it does look quite a bit different than the 3-meter dataset that we did as part of our bioscapes work. Some of this discrepancy could be because we included a seagrass component that's not visualized here, but that does have some overlap with the 10-meter tidal marsh data set. And additionally, with that 3-meter data set, you might be mapping things a little differently and a little more precision in your mapping approach.

Speaker 2: Sensor properties. So talking about spectral resolution here, we have in this figure, we have a spectral curve of Phragmites Australis to the common reed, and then in those purple lines with the diamonds on the end, those are the extent of the different Sentinel-2 bands. And you can see that, for the most part, the Sentinel-2 bands, not surprisingly, capture a lot of the variability of the spectral curve. But there are certainly regions of the spectral curve that are not being captured and collected by Sentinel-2, and these regions could be and are important in some cases for differentiating spectral signatures of different salt marsh species and are areas where we could improve our mapping with this additional spectral resolution in the future.

Speaker 2: But for the most part, even so, Sentinel-2 and Landsat are highly appropriate for mapping salt marsh environments. And part of this is because of their collection of bands within the near-infrared and the red edge region, which are two regions that have you know, importance for mapping vegetation and wetland vegetation. The revisit time and the temporal resolution of our sensors is also incredibly important. Obviously, Sentinel-2 and LAMPFAT are collecting, have, you know, different revisit times, but relatively frequently, twice to a couple times a month.

Speaker 2: And that can depend on the cloud cover and other things. But here we have an example of why we care about that revisit time. And that's the tidal stage. So on the left, we have a low tidal stage. And you can see most of the mudflats and intertidal areas are exposed. And you could map these relatively easily as one common category. There's not half of them as inundated and half of them exposed, which can make it complicated mapping seagrass, in this case, or intertidal salt marshes. The higher tidal stage image demonstrates that despite having a higher tidal stage in some areas of the intertidal region being inundated, this could be important for mapping the differentiation between salt marsh areas and intertidal seagrasses, because intertidal seagrasses, for the most part, are lower on the tidal frame.

Speaker 2: So they'll be inundated first, for the most part, and you'll be able to map a little bit better the extent of the salt marsh. So that could be important. And using these images and our understanding of our study site, we can kind of improve our mapping. of the salt marsh environment. And obviously we're not just necessarily constrained to those two images. So now we've looked at the entire Sentinel-2 and Landsat 8 and 9 archive for NISNA and we've compared it to the tidal stage at the tidal gauge there for each of these images.

Speaker 2: And we see that for the most part, very few of these images are below mean low-low water. I mean, that is the average, that is the lower of the two tides that happen daily. So it's not surprising that we don't have that frequently occurring. We have lower tides than that. But these extreme events, either spring tides or other events leading to very low water events, could be useful for mapping in the intertidal seagrasses especially, but also the salt marshes. But we're not just limited to those.

Speaker 2: And these other ranges of tidal impact could also be useful for mapping. But it's also important that we know and are aware of when the inundation happens and what images are particularly highly inundated, like those above two meters of tidal inundation might be important here. But you'd want to do that with understanding your study site in particular. And here's just a little zooming into a particular site. Here's a phenocam image from Georgia that shows Spartan Alterniflora. It shows a patch of Spartan Alterniflora.

Speaker 2: At low tide, you see no inundation evidence. There's no water visible. But then at high tide, it almost completely obscures the vegetation beneath it. And you would map this, almost all classifications would map this as open water. So it is important that you at least understand that this could occur in your salt marsh mapping. And you might completely underpredict the extent of your salt marsh. This would be very hard to map. In the distance, you have some exposed spartan alternate flora. Mapping that and the inundated spartanal term floor would likely be very hard with most mapping approaches, but it could be possible if you're aware of it.

Speaker 2: Okay, now we're moving on to our demos. Okay, so here's our first example, code example. We have a classification of Nizhna estuary, including we have water as a class, and that's in blue. We have seagrass in blue. black here. We have tidal marsh in green. We have tidal flats in kind of the tan color. And we have upland areas in the periwinkle purple color. So that's our general classification that will be consistent throughout the examples here. And this classification uses a high tide image. So we can see here most of our intertidal areas are inundated.

Speaker 2: And that's pretty clear. It's a little bit higher tidal stage than used in our sample. And when this gets classified, you can see it still does a fairly good job of classifying areas that we think are seagrass, but maybe it's a little bit more extensive than we would want because the seagrass looks a lot like other inundated areas or other shallow inundated areas. So it's a little harder to differentiate those. And overall, this is around a 90% accuracy. But if we go to a lower, so our code, our simple code, and we've commented this.

Speaker 2: So if you need, if you need an example, first, we define all of our elements. These are here and you look at some other R set trainings for actually creating this kind of training data set. And then we have, we define our image. This is from prior knowledge. We know this is a high tidal stage image, so we define it here. We define the bands that we're interested in. This is all the 10 and 20 meter Sentinel-2 bands. We select these bands. We then define some visualization parameters so we can visualize our RGB pseudo color image.

Speaker 2: So we add that to the map, and this is near infrared blue or green blue. And then we merge together our training data. So again, our water, seagrass, tidal flats, tidal marsh, and upland classes. And then we train a random forest model. We extract the spectral data to our points. And then at 10 meters spatial resolution, And then we merge our, this is our test data set. So we kept our test data set instead of using a random column to distribute it. We wanted to keep it consistent across the examples.

Speaker 2: This is our test data set. Merge that and extract the spectral data. Print the testing and training length of those variables and then train our random forest classifier to get our our trained classifier which we then use to classify the test data set and determine our test accuracy so 90 percent is determined there and then our final set is we define a um we we apply our classifier to the entire image and we define a palette for visualizing this and add that to the map and then you can also export this image and this The EPSG here is a UTM projection for this region, so you'd want to change that if you are changing the region of your study site.

Speaker 2: But this script otherwise should work for just about anywhere if you change the training data and image to your study region. So now we'll quickly go through kind of the results and the impact of including or excluding certain data. So if we zoom in here to our same area of knife estuary, We have a little different look here. And we see this is a low tide image. So you can see a lot of the intertidal region is exposed with very little inundation of the intertidal seagrass and intertidal marsh. This results in, when our classification results, does result in a pretty different and larger salt marsh expense than our previous classifications.

Speaker 2: So some advantages here. We do have a lot of misclassification in the upland area where we have a lot of areas that are fringing uplands that are classified as whether they're on the edge of kind of transition areas like transitions between agriculture and other land use or just in urban areas. You have a lot of misclassification of those upland areas as marsh. So that is some area that we could kind of address with using elevation. And we'll see that in this next classification, which is the multi-date.

Speaker 2: So we merge the two images into a single image collection and then a single image. And then we classify that multi-date image, so the high tide and the low tide image. And this does improve our classification accuracy slightly. And you can see these images are the same ones we used previously. They're both from kind of the austral summer or the southern hemisphere summer for 2020. and 2021, but in, oh, actually there. So they're in two different summers, but they're both in a similar time period of the year, January of 2020 and January of 2021.

Speaker 2: So minimal seasonal differences, but a little bit different, or very different tidal conditions led to slightly better accuracy. And then our Final classification with Sentinel, or this is our second final classification, and this uses Landsat and includes kind of all the bells and whistles. So we use vegetation indices, in this case, NDVI. It slightly improves our classification, and we also use the elevation data set. So this is using Delta DTM, which is a link here. This is a open source elevation data set derived from TandemX and ISAT or improved TandemX using ISAT and JEDI and other information.

Speaker 2: So we have here, we have this, you can see this kind of addresses our problem with having the upland areas being misclassified as salt marsh. We almost have none of that because we have a very successful use of the elevation data set to kind of clip or soft clip our classification. And you can see that the same approach was used here with a low tide image. and a higher tide image. So you can see that variability in the land side images almost just as well as you can see in the sentinel. And that results in a pretty nice classification of our salt marsh extent.

Speaker 2: It has a lot of agreement. You have some common issues here, like docks and interspersed urban and coastal communities. Sometimes you get massive salt marsh, which you could be doing. adjusting in post-processing or adding some training points in that area to kind of remove it. And then our final approach, and this is the highest actually approach with around 95% accuracy. This is our same approach as the Landsat image, but using those two Sentinel-2 images. So it's a slightly higher accuracy than the Landsat-2 classification or Landsat-2 eight classification, this Sentinel-2 based classification is slightly higher actually, and you see many of the same kind of benefits of using elevation here to have a soft flip and just have a pretty accurate classification of marsh seagrass within Nizhna estuary due to using those two data sets and using those two image collections and image acquisitions and the elevation data set.

Speaker 2: You can kind of see that it does a pretty good job of classifying our study site and this is the one we use in our carbon calculations further along in this section. Those are our example demos and all this code should be available for you to either shift to your study site or run as you understand it. They are mostly all very similar to one another. So I would recommend if you are interested in applying it to apply the multi-temporal elevation script to your area of interest. So those are some, so now you get an idea of what we can do with different data sets and whether it's Landsat or including elevation into our classification, you can see that kind of improvement over time.

Speaker 2: So this is kind of the outputs of our different classification approaches. We can see that our lowest performing was that high tide image, but it wasn't completely non-predictive or anything. It was a pretty good classification even so. And then you have the low tide image slightly improved on that. Our multi-date image combining those two images actually was a slight improvement on just having the low tide data. But then if we include elevation and our multi-date image, that's our clear performance.

Speaker 2: highest performer. And a lot of that is improvement in the upland areas, which we could also achieve with a clip or removing them from our classification. So something to consider. There's different approaches, but elevation did do quite a good job of being that soft clip in this particular example. And then we also had our Landsat 8 classification that included elevation and NDVI. And while this does have a slight drop in accuracy compared to our Sentinel-2 image classification, it has a very similar distribution.

Speaker 2: And it's just kind of that spatial resolution is a little bit worse at mapping our area of change. But it's also two different tidal stages and two different times of the season. So there's other contributing factors here that these aren't a perfect comparison. So now we're going to talk a little bit about carbon calculations. And this is a slide that you saw in our previous session, part one of our training. And this is the tiers of the IPCC calculations for carbon emissions. So tier one would be kind of using global data, whether it's some of the global maps we talked about and kind of just either the IPCC or global estimates from the literature of the carbon stock and what that means for emissions.

Speaker 2: And then tier two would be using more country specific data. And we're going to keep focused on these two, but we're also going to just show some example of some of the advanced methods that maybe could be used in a tier three classification, those global soil organic carbon, the spatial data set, and maybe the spatial predictions of above ground biomass might be useful for these kind of advanced methods. So the area and carbon estimates, so if you want a good approach to mapping uncertainty, there's a link there.

Speaker 2: We kind of just used a 20% uncertainty for these estimates just for illustrative purposes, but I would use an accuracy assessment based uncertainty metric. And you can see that our results for the Landsat 8 and the Sentinel-2, the best performing classification, so those multi-day elevation included classifications, are pretty similar to one another and are well within their uncertainties,

Speaker 2: with a slight discrepancy in the extent of sea grass between the two being kind of the largest area of difference. And then we also have the global tidal marsh layer, which is a little larger extent than these two. So that would be something to consider when you're mapping if you feel like you have some expert knowledge. It might be useful to do your own maps instead of rely on a global layer that is mapping for a very different goal than regional monitoring. And then you have these global estimates of soil organic carbon megagrams of carbon per hectare.

Speaker 2: And you see the main difference between the regional and global estimates is that we have a slightly lower regional estimate of soil organic carbon, or not just soil organic carbon, this is all biomass carbon as well. But there's a slightly lower regional estimate of seagrass carbon than global estimate, and this is to be expected. And there's also a slightly higher estimate of salt marsh carbon in the regional estimate than in the global estimate. Those are just two notes. And when we do our calculations, just a simple equation there using, and their uncertainties are propagated in the propagate package of R to get these estimates of soil organic carbon and uncertainty.

Speaker 2: Relatively simple approach here. But the main difference we see between these is probably that discrepancy in the salt marsh, the tidal marsh, salt marsh carbon account between the global estimate and the regional estimate, with the regional estimate being slightly higher than the global estimate number. You see, for the most part, these aren't radically different from others. The Sentinel and Landsat are very similar. The global tidal marsh is slightly higher than both of those, but still very reasonable in line with in line with the combination of tidal marsh and seagrass for sure, but also just very similar to the other numbers.

Speaker 2: Overlapping ecosystems is another consideration. So obviously we have the global tidal marsh layer and we have some overlap with our seagrass layer. If we were going to harmonize these for a carbon accounts for NYSMA, we would want to address this overlap. So this example is around 55 hectares of global tidal marsh layer and seagrass overlap. And this is obviously something we don't want to double count the carbon in these ecosystems because either they're actually overlapping in situ, like we saw with our image in our study site slide, or they're mischaracterization of one of the two ecosystems.

Speaker 2: So you do want to make a decision, and we're going to talk more about that in the future, but whether it could be based on which of the classifications you feel is more correct for your study site, you could say, okay, the tidal marsh, I think, is more correct. So that's going to be what I'm going to account for. And this seagrass overlap will be an error in my seagrass map, essentially. But it's something to note, and you can also see here, we also did how much overlap between our tidal marsh, which does have some clear errors in it, and the global tidal marsh layer.

Speaker 2: So obviously, you could iterate and improve on that and either try to bring those together or do some field reconnaissance to see which one is the reality for different pixels, different areas of your study site. So now we have some of those advanced methods, and these are some spatial predictions of above-ground biomass and soil. We talked about the soil organic carbon layer, and here you see it visualized. Soil organic carbon from 0 to 30 centimeters and soil organic carbon from 30 to 100 centimeters for a section of Louisiana's Mississippi Delta.

Speaker 2: And then you also have a spatial prediction of above-ground biomass in the data set, the link to the data set that I put out recently that has a spatial prediction of above-ground biomass. And obviously, these don't completely coincide, and you wouldn't expect them to. There's, you know, the above-ground biomass might adapt much quicker than the below-ground oil organic carbon. And you can see some variability in the two. But if you were going to do an advanced carbon calculations, maybe these types of spatial data sets you might utilize to get a better understanding of lost areas and exactly what kind of carbon they might contain.

Speaker 2: Okay, to summarize what we've gone over in this section, we've introduced salt marsh ecosystems, a quick dive on what they look like and where they are. We've discussed the available global resources, whether it's that soil organic carbon layer, or some of the global maps like the global tidal marsh 10 meter data set that are available right now that you could download and look at for your study sites. We also talked about what we need to do and what we need to consider when conducting salt marsh classifications with remote sensing.

Speaker 2: And then finally, we've done a overview of utilizing Google Earth Engine for classification just as an example and kind of understand what are the impacts of that tidal stage, the use of elevation data, how does that impact classification accuracy, and we've reviewed an approach for carbon accounting of seagrasses and salt marshes. Thank you. Hi, everyone.

Speaker 3: My name is Kelly Lewis, and I'm a scientist in the Water and Ecosystems Group at NASA's Jet Propulsion Laboratory in Pasadena, California. Today, I'm really excited to take you on a different type of blue carbon journey, and that's because we're going to be going beneath the water surface and exploring seagrass mapping with Earth observations. While remote sensing applications for this particular blue carbon system may be less mature than blue carbon counterparts in mangroves and salt marshes, there's really a lot of exciting things to look forward to with current remote sensing missions as well as future ones that can help us advance seagrass mapping with the hopes of better accounting for the blue carbon potential in seagrass regions.

Speaker 3: So to kick us off, I wanted to show this really beautiful photo from underwater photographer, Pekka Turi, which beautifully captures the themes we're going to be talking about. So first, what is seagrass? Why is this a really special ecosystem? And a lot of it is due to this root complex you see along the edge. That's really what's responsible for the carbon sequestration blue carbon potential we'll be talking about. We'll also highlight some global data sets with in-situ information that can help set us up for exploring remote sensing data sets.

Speaker 3: And in that section of this training, we're just going to be talking about passive optical instruments. So that is really understanding how light interacts within the water column and with the benthos or seagrass in itself that helps us make estimates of seagrass extent and cover from remote sensing platforms. I'll then quickly flash up some links to demonstrations in Google Earth Engine created first to orient you to different types of data sets that exist and doing some time series analyses as well.

Speaker 3: And lastly, we'll round out on some key takeaways. So with that, let's just start off with what is Seagrass? And I think just the history of it is really fascinating. So Seagrass is really involved around 100 million years ago from actually freshwater ancestors, and they're actually the only fully submerged marine flowering plant. They're more akin to maybe terrestrial plants and trees that you might see, which actually have stems and leaves and roots, where if you were thinking about algae, they don't exactly have the root structure that you're seeing on this drawing on the right, and this is of salacia.

Speaker 3: And these grow in generally salty brackish intertidal, subtidal waters, around one to three meters waters of depth generally, but they can grow down to 58-ish meters for the deepest growing species. And they are found on all continents except for Antarctica and over 150 country coastlines that we know of. And they're known as ecosystems engineers because of these special root iso systems where they're actually creating habitat and shaping the environment around them that results in actually reducing coastal erosion because they stabilize sediments, it improves water quality in the region, and it is really that cool root system that is responsible for sequestering long-term carbon in the ocean here.

Speaker 3: Next, we have, like, if we were to look really closely at this diagram, you can see that there's just tons of life that lives on the plate level. In the top right corner, there's and other things living microscopically on and within the plant, but they can also create these expansive meadows where we see this awesome turtle feasting on the seagrass blades themselves. And they're really known as biodiversity hotspots. And if we were to focus on the carbon sequestration component of this, despite covering 0.1 to 0.2% of the total ocean floor, they're responsible at a 10 to 18% of total carbon burial in ocean.

Speaker 3: So that's roughly one acre of seagrass and sequester 740 pounds of carbon. There are estimates that that's 30 times, 35 times faster than carbon sequestration for tropical rainforests. And most of that carbon is sequestered in the sediments under that. So that contributes to that long-term carbon storage and the blue carbon potential. However, there are differences in that burial rate by species of seagrass, sediment types, and environmental conditions. This is from Johannesson 2022, which even looks.

Speaker 3: Really simply at tropical versus temperate regions and different types of seagrass species. They're largely narrow leaves, have different water optimal conditions, and those all create different conditions for storing seagrass, for storing carbon.

Speaker 2: Burial in seagrasses.

Speaker 3: So this type of understanding is really special in the sense that there are these green catalogs that we have now that's starting to track these globally. So there's the global distribution of seagrasses data set, which they started out in green and short 2003, and now it's in its seventh iteration in 2021 that's held by the United Nations World Conservation Monitoring Center. And these are really points and polygons of where seagrasses have been found. But then there's more information on metal structure, biomass, and production in this Paper from Stridham and L2023 would actually look at literature between 1975 and 2020 and catalog things like percent cover, shoot density, leaf density to get a metal structure, biomass information such as above ground, below ground total, production information from the shoots to leaves above, below, and total production rates.

Speaker 3: So these types of global data sets at the institute level help orient us to how we want to go about exploring remote sensing or understanding blue carbon mapping in the underwater environment. And I'm going to primarily focus on passive optical remote sensing. So this is really talking about if you have a platform flying by, like you see in this diagram, and they have detectors that collect photons, and we're going to be turning mainly the visible to near infrared, color that we can see with our eyes, then to the parts of the infrared.

Speaker 3: And what's super cool about this type of technique is that what we're trying to account for are really the journey of the path of light or a photon. And it has many interactions from the sun down in the atmosphere, to the water, and then what gets bounced back up at a sensor. So that's why you see this really colorful graphic with representing these photons have interactions with the atmosphere that gets picked up by the platform. There's the water surface itself. The thing we're really interested in is what's happening below.

Speaker 3: And that can all, that signal can be mediated by many things, the color of water, the water quality, the depth of our species of interest. So these are all things that get mediated before you get a detection at a platform. So we made this note here of really the best times where we can use remote sensing is when the seagrass is most exposed, so things at the lowest tide conditions are really ideal for understanding that mapping seagrass extent. Regardless, this is an optical sensor because it's collecting information about the atmosphere.

Speaker 3: You can't see through clouds, and there's also the water column impact if there's a large water quality sentiment, plume event, and light mass sets, you know, that we see from seagrass. So these are all important things to account for. There are several current and upcoming missions that we can look forward to that leverage this type of technology. So those are, you might be familiar with things like the Landsat series. I mean, especially 739 here, 90s to present, this is a multispectral instrument, 30 meter spatial resolutions, roughly the infield of a baseball stadium, with a 16-day revisit for each sensor.

Speaker 3: Next, you have Sentinel-2, primarily A and B. Recently, C was launched. That goes back to 2015. Also multispectral. And you'll see for CRS applications, you'll see this spatial resolution used in that 10- to 20-year window. And that has a five-day revisit per.

Speaker 1: Instrument.

Speaker 3: Lastly, we have the Surface of Biology and Geology instrument. So this is called for in the U.S. National Academy's survey to create a hyperspectral instrument that can look at these types of features as well, or mapping that surface biology and geology. That's aimed to be launched no earlier than 2022, but a hyperspectral instrument we can look forward to. One of the reasons they bring up the hyperspectral is because it's this spectral information and that passive optical information that can help give us more rich information about the species that might be there, the state of the seagrass health.

Speaker 3: This is a paper from Thorog et al. 2007, and you can see just alone that aquatic versus terrestrial species have different spectral signatures when you look at the hyperspectral and the visible to the infrared. So that's 400 nanometers to 10,000 nanometers. You see this pronounced peak in the 500, so this is like the green.

Speaker 2: Reflectance.

Speaker 3: The way to think about reflectance is that's like the light can be absorbed by a particle or medium. That light can be reflected back at you. So it's being reflected back at you within the green. You see these green leaves. And then they have different types of these red edge reflectances that differ based on terrestrial plants like you see in these dashed lines versus aletheia, which you see in the bold line here. Next, when this Paper looked at different species. You can see the really subtle differences that get picked up in that hyperspectral signal, most pronounced in this 500 to like 680 window, which is really cool if you have this type of information.

Speaker 3: But then you can also see very big shifts in the spectral shape versus healthy leaves, like in bold mine versus brown leaves, senescing or stress leaves versus black leaves. These are all cool things that we can start to pick up with remote sensing. Another thing to really consider is spatial resolution. If you have access to a drone, you can get down to a centimeter scale and start to differentiate between sand and seagrass. But as we go up, space-borne instrumentation, world views, planets, and the meter scale, then we go to things like Sentinel-2 and Landsat 8, which are more available in things like Google Earth Engine.

Speaker 3: becomes a little fuzzier and more pixelated. So this is, I

Speaker 3: think I'd put this in here more just to kind of underscore the importance of, we need to really understand the atmospheric and water column impacts so we can best associate this pixel with, is it a seagrass percent? Is it representative of the dominant seagrass coverage in this region? Or if this might be representative of a adjacent benthic substrate that might not have the blue carbon potentiometer that we're necessarily interested in. Now to link remote sensing with proxies for carbon sequestration, I really like this figure from Simpson et al that I do a bit of a quick modification on just to think about what we can do with remote sensing and making those links.

Speaker 3: So true to this Simpson figure is we can look at things like above ground biomass and spatial arrangements. So thinking about things like You can use NDVI or terrestrial structural entities to map greenness in those lower tide conditions or where the canopy is more exposed. You can look at things like spatial arrangements or understanding how fragmented or connected that carbon stock might be. My minor modification is actually looking at not just species composition, but shallow benthic composition.

Speaker 3: And that kind of goes back to that hyperspectral information, which shows that if we can get a better understanding of other types of species in that region, we can better understand what is the overall landscape ecology and the ability to really classify these images to best represent seagrass extent. Lastly, the amazing thing about having something in space that goes over you at a regular frequency, you can start to make time series, which again informs us on seasonal and annual changes in that seagrass cover and percentage.

Speaker 3: I just wanted to point to some few examples that exist, and this is more of just pathways to get our brain thinking about different things we can do with remote sensing. So this is work from, say, at all 2020. They were looking at Zostra along the European and Atlantic coasts, and they were using Sentinel-2 to map NDVI really the lowest tide setting. And then they utilized in situ information and field data to inform additional maps of seagrass percent cover and seagrass leaf biomass to get at things like above ground biomass information.

Speaker 3: Next, this kind of speaks to that composition point I was talking about. And this is an Elkhorn, California saucer species. And this is the power of hyperspectral or spectroscopic information. And you see in the left column, you have differences in the spectral shape of seagrass and sediment. It can be very, very fine when you average them in that panel C. But when you look at differences, you do see subtle differences that can help us pick out what might be seagrass, what might be cinnamon, but there's still so much work to go on in utilizing this type of information with future sensors.

Speaker 3: Lastly, I wanted to highlight this work out of St. Joseph's Bay, Florida, U.S., and this one actually has a number of species. So you see with the here, Galatia hallouli, syringodium, filia. And what they do is look at Landsat 5 through 8, and they're really utilizing that remote sensing observations to make maps of seagrass presence and extent. And then they combine that with things like leaf area index, biomass, below ground carbon stock to look at how that carbon stock is changing over time by leveraging that long-term time series information.

Speaker 3: So lots of different really recipes on ways that, depending on the type of data you have access to and how visible or detectable your seagrass patch is, many different routes you can go down in understanding and exploring this really incredible underwater blue carbon ecosystem. I'm going to have some links associated with this slide, I believe, can be posted in there that can offer a soft Let's just look at what Google Earth Engine data sets exist. So this is actually from the Allen Coral Atlas, which uses airborne and planet sensors to make coral habitat maps.

Speaker 3: A variety of benthic categories are in there. And this might be a good place if you might not have access or might not be familiar with existing seagrass maps in your region. This is a remote sensing Earth observation that's already plugged into Google Earth Engine to provide a really good base for additional analyses. There will also be a separate link in there on actually doing time series analyses with Sentinel-2 utilizing things like NDVI or a modification of normalized difference aquatic vegetation index that's actually shown in a previous RSET training on monitoring aquatic vegetation.

Speaker 3: So those will all be available to you. And lastly, we're going to round out on key takeaways, just wanted to return to how special these blue carbon ecosystems are. And the reality is that they are threatened by global climate change and anthropogenic activities. There's estimations that these metals globally are degrading by 7% a year. And it underscores the importance of utilizing remote sensing tools for detecting and quantifying spatial and temporal dynamics of these metals over time. and that we have examples of that.

Speaker 3: But to really leverage that remote sensing information, it's really important to pair the field data that exists to get a better understanding of what happens to these carbon stocks in relationship to these changes. So I will leave you with that, and I'm really looking forward to hearing questions in the live Q & A session. Thank you very much.

Speaker 4: Thanks, Kelly, and thanks everyone for being here. My name is Cheryl Dowdy, and I'm going to be talking about measuring co-occurrences in blue carbon ecosystems as part of today's training. So the learning objectives for this section are to be familiar with existing global data sets showing the extent of mangroves, salt marshes, and seagrass, understanding how to compare global data sets in an area of interest using Google Earth Engine, learn how to adapt mapping approaches for co-occurring blue carbon systems, and understand the research needs for future blue carbon accounting.

Speaker 4: In this section, we'll be covering how to synthesize blue carbon estimates across ecosystems where they co-occur. We'll touch on important topics like scope and scale and merging maps and identifying overlap in coastal areas that are interesting to us. We'll cover blue carbon reporting, monitoring, and accounting in co-occurring systems and draw similarities with today's previous trainings by walking through an example where mangrove and salt marshes meet. And we'll leave off on how synthesizing blue carbon estimates and combining methods is needed for the future of blue carbon research.

Speaker 4: It's clear from the past presenters that we can gain an amazing understanding of blue carbon ecosystems with Earth observation at various scopes and scales across the globe in mangroves, salt marsh, and seagrass ecosystems. And our understanding of the blue carbon ecosystems can be slightly uneven across the globe. Because our knowledge is based on a combination of maps made in different ways to map different environments and sometimes to answer different global questions. So you've heard a lot of these existing global data sets from the other presenters in this training.

Speaker 4: And there are new and improved maps continuously being made. So it's best to follow these resources and stay updated with what's coming out there. It's a rapidly changing field. But all of these data sets, they contribute to our global understanding of where blue carbon ecosystems exist. And so together, a global picture has emerged in recent decades of blue carbon ecosystems with the help of Earth observation. So we get to know where they exist across the world's coastlines, where they dominate in certain areas, coexist in others.

Speaker 4: And these ecosystems are continuing to change across the world because of climate change. and other more direct human impacts like deforestation and reclamation, so filling, so that we can reclaim that land. And so, ultimately, all these changes need to be monitored using remote sensing. It's the best available tool we have to track them globally. But where blue carbon ecosystems co-occur can be especially difficult to monitor. These areas are changing rapidly and overlapping habitats often require special mapping considerations.

Speaker 4: And so what about where these ecosystems meet? How can we measure blue carbon in ecosystems undergoing rapid changes and shifts in habitat distributions or their health due to climate change and other human impacts? So there are really advantages and disadvantages of the existing global maps we've heard about and merging existing maps and identifying overlap can be a really great place to start testing the existing data set in your region to perform your study and answer your questions and also to identify needs and spatial data and data resolution.

Speaker 4: So where might there be gaps in our current understanding? And then ultimately, this will help you kind of adapt your mapping approaches to coexisting blue carbon ecosystems where you might be interested in them. Scope and scale determine a lot of what we can measure reliably with remote sensing. It's important to consider the scope of the domain. How big of an area are we looking at? That often determines what data to use in your analysis because of those trade-offs in spatial, temporal, and spectral resolution that we've been hearing about today.

Speaker 4: So for example, the world hasn't been mapped globally at five centimeters because that's a lot of data and a lot of computational power needed. So it's our job to essentially consider these trade-offs so that we can get the best data to perform the analyses at the right scales to answer our questions. And so here are the trade offs in spatial and spectral resolutions. They're clear among four commonly used commercial and public data sets. These are Worldview, Planet, Sentinel, and Landsat. Now some of these data, they're free to access, some are available to certain education and research institutions, and others are more difficult to access.

Speaker 4: So you need to see what's available to you. Now these images also have different temporal resolutions, which matters if we are using one image or maybe say a composite image summarizing a longer period of time to investigate our blue carbon ecosystem. So again, the timing of these images in the coastal zone is especially important given the effects of tides on our ability to map these vegetated ecosystems. Now these considerations are important for applying mapping approaches in blue carbon ecosystems across the globe.

Speaker 4: choosing the imagery at the appropriate resolution to map the differences in your co-occurring blue carbon ecosystem, collecting appropriate training data that reflects all the classes of interest in your area, and certain classes, like, for example, seagrasses, which we've heard about, have some special mapping considerations because they are subtitle So therefore, you might have to subset or get creative with your mapping approach to combine results for different systems to get a final classification map.

Speaker 4: There are lots of different approaches to go about this. But the end result in this conceptual workflow are predicted classification maps of habitat extents. But this could be modified to predict above-ground biomass height or carbon in our blue carbon ecosystems by changing the training input data, if we have that data or can find it, and making some other tweaks in our classification algorithms. So essentially, these are the knobs you can tweak to adjust and adapt these supervised learning approaches to your coexisting blue carbon ecosystems.

Speaker 4: So now I'd like to illustrate how we can measure co-occurrences of blue carbon ecosystems with these considerations in mind. And I'll be focusing where mangroves meet salt marsh along the northeast coast of Florida, where I've had the opportunity to study them since conducting my master's research over 10 years ago. And so in Florida, there is an ecotone formed where mangroves and salt marsh habitats overlap. An ecotone is a transitional area of vegetation between two different plant communities.

Speaker 4: Plants at the edge of their geographic boundaries can be especially sensitive to change. And ultimately, what this leads to is a mismatch of what's really happening on the ground and what we can reliably map globally with moderate resolution imagery. And in Northeast Florida, this is best shown by small mangrove trees and patches surrounded by marsh, undetected in recent global salt marsh and mangrove data sets shown here in blue and orange. Now we'll do a brief demo of some exploratory tools for viewing co-occurring blue carbon systems in a region of interest.

Speaker 4: So the first thing I like to do in my scripted analyses are to set the scene by importing any helpful underlying data and then defining our region of interest. Then we'll import our blue carbon layers. We'll identify the limits of those layers and we'll extract some blue carbon system extents. And you'll see at the end of the script, there's some optional exercises for you. Now let's bring up the code. Okay, so in our Earth Engine script, The first thing that we're doing are importing some useful color palettes from a package that someone else created.

Speaker 4: It's really great to use. I highly recommend it. This will help you visualize some of the underlying data layers that you'll see in a little bit. And so the first thing we'll do really is set the scene by importing the region of interest that I'll be looking at. First, by looking at the shapefile layer for Florida, then bringing in some watersheds to help us look at these ecosystems and SRTM elevation. And so these layers are really helpful in helping us determine where our blue carbon ecosystems exist at these really large scales.

Speaker 4: The next step in our code is really just defining our region of interest. So the way that I've done this is by Drawing geometries.

Speaker 4: So you can go in here and modify your areas of interest using these drawing tools right on the map, which are very helpful. And I've done that for the East coast of Florida that I'm going to be focusing on for this example today. So in this code, what I've done is taking the geometry and convert it to a feature collection And just adding it to the map in the background so we don't have to look at this red box. So here's where we see our region of interest, otherwise known as an ROI. Again, setting the scene.

Speaker 4: And then we can add our blue carbon data layers. So I focused today's exercise looking at blue carbon data layers that are already loaded into the Google Earth Engine assets. And so for the intertidal layer, we are looking at the global intertidal classification that's at 30 meters from Murray et al. 2018. For the mangrove data layer, we're looking at distributions from 1997 to about 2000. This is from Geary et al. For seagrasses, we are using the Allen Coral Atlas. a five meter geomorphic zonation and benthic habitat classification.

Speaker 4: With this dataset in particular, we have to also convert it to extract the seagrass habitat layer. And the way that we can modify and play with all of these different sets of data are by first inspecting the data themselves. So the way I always like to do that is taking the exact link to your feature collection and making sure we know what we're looking at. So for this example in the Allen Coral Atlas, you'll wanna read all about it, definitely cite the papers if you use their data, but you also wanna go into the bands and get an idea of what data values you're looking at.

Speaker 4: And so, In investigating this data set, you'll find that there are seagrass habitats mapped and you're going to want to extract just those habitats. So I've done that by focusing on the classes that are just equal to the number 14 and that's how we extract our seagrass data set. from this combined benthic habitat data set. We didn't have to do that with mangroves or intertidal because they were mapped separately. But again, what you want to do is go check out these data sets to really get a grasp of the underlying data that you're using.

Speaker 4: And the last data set I've included is really just for comparison. It's the ESA Sentinel Global Cover at 10 meters. So again, these are all data sets that will help us examine the extent and distributions of our blue carbon ecosystems in this region. And when I add these maps to the layer at this scale, it might be a little bit difficult to see, but you'll want to zoom in and I'll do that in a little bit and really get an idea of where each of these three blue carbon ecosystems exist based on those data layers that we're importing that are already existing in Earth Engine.

Speaker 4: And you have the option to upload data layers of your own to examine. But again, I wanted to stick to ones that I could easily cite.

Speaker 4: So the next step in this code is really to take those underlying data layers and explore them. So the first thing that I want to do is to identify the limits of, say, mangroves and seagrass. The first thing that we need to do in order to get this done in Earth Engine is to actually take our mangrove map that exists in, say, raster format and convert it to vectors or shapefiles. And here in Earth Engine, you'll find that scale and maxpixels are very important parameters to tweak in order to get your code to function.

Speaker 4: So here we're opting for a scale of 60 meters, again based on the resolution of the underlying mangrove data, which is 30 meters. And you might have to play around with this max pixel setting depending on what kind of errors Earth Engine gives you. And so keep in mind that the scale here for mangroves we're looking at is 60. So once we've converted our mangrove image to vectors, we're really just sorting our vectors by latitude and limiting it to the first one to get the maximum latitude of the mangrove vectors.

Speaker 4: And then we'll plot that on the map. And so I've done the same thing with sea grass. And you'll notice here that the scale is a little bit different. And again, you really need to understand the underlying data and have a general sense of your region of interest to know what kinds of things you're looking for. So here in this region, the seagrass areas are a little bit larger, whereas the mangrove areas I'm looking for to the north are a little bit smaller. And so again, we're converting the seagrasses to vectors in order to get that maximum latitude seagrass.

Speaker 4: And so you might have to play with whether you're looking at the maximum or the minimum depending on where you're situated. around the globe if you're northern or southern hemisphere. And if we then zoom out and look at where those are mapped, hopefully they will show up on this live demo. Now I've hit mangrove first and seagrass second, but the seagrass popped up first. So we'll see if this code is going to cooperate and there we go. So we just zoom in a little bit and then we can see our mangrove range limit pop up.

Speaker 4: So if we zoom in to the seagrass range limit, this is down south along the eastern coast of Florida near the St. Lucie Inlet. We'll slowly let our layers populate and we can play around with the underlying imagery to help us really understand and begin to analyze these data layers. You'll see that this is.

Speaker 2: Really a.

Speaker 4: Vector showing you the area where your limit is. It doesn't perfectly match the vector because, again, of those scaling factors we had to use in converting these to vectors. But we're finding our northernmost patch of seagrass shown here in green. Now I'll zoom out and hopefully the mangrove range limit to the north will pop out. And you'll notice that I didn't bother trying to map the range limit for intertidal and you could probably guess that is because the intertidal layer exists throughout this region of interest.

Speaker 4: So if we tried to find the range limit would probably be at the top of our region of interest. And so here, the mangrove range limit has populated. We can zoom in and we see that it's in this tiny intercoastal waterway neighboring the open ocean here. So exploring and mapping at these scales really gives us an idea of the discrepancies in our data sets and where they are limited in terms of geographic scope. So here we have an idea of where our data sets do and don't overlap. Now the last step in this simple exploratory script is really just calculating the area that we're getting for each of those blue carbon ecosystems.

Speaker 4: So the way that we normally do this in Earth Engine is by creating an image that's based on pixel area. The default is in meters, so we need to convert it to whatever unit we want, and today I'll be looking at hectares. And then we're updating that image to just reflect where the mangroves, where the intertidal, and where the sea grass are. And then from those images that have been converted to area, we're reducing to get the sum of the area within that ROI geometry that we've defined. And again, you might have to play around with scale and max pixels to work through any errors you're getting in Earth Engine.

Speaker 4: So when we hit run and we extract all these areas for the three blue carbon ecosystems in this region, we're able to find that along this east coast of Florida, we have about 8,000 hectares of mangroves, over 29,000 hectares of intertidal areas, and about 780 hectares of seagrass that are primarily to the very south. Now from here, you have optional exercises to adapt this code for your own area of interest by playing around with the geometry imports. You can also calculate area along coastal segments by using this little snippet of code I've left in here to create a 10 kilometer grid over which you can extract your results latitudinally.

Speaker 4: And I'm sorry that's actually yeah it's a 10 kilometer grid that can be modified to be smaller if needed. And then the last option you have is to take these area estimates and convert them to carbon using either tier one, tier two approaches. And so now I'll head back to the presentation. Okay, so hopefully from that demo, you've learned that we can change our region of interest and explore these blue carbon layers in our own areas to identify the gaps in our current understanding and data sets and identify the limits of your blue carbon ecosystems and begin to extract blue carbon results from global data layers and maybe identify where you might need to do mapping exercises of your own.

Speaker 4: So exploring our regions of interest helps us identify gaps in our current understanding of blue carbon ecosystems extends. Then we can adapt our mapping approaches. And so here, after such exploratory analyses, I'm showing work that I've done to better improve our understanding of mangrove and salt marsh ecosystems in these areas. And these are the results of that where I've classified mangrove and salt marsh using imagery ranging from 1.8 meters to 30 meters to map these habitats. And so clearly, there are really big differences in the extents and the boundaries that we can measure at different resolutions of satellite imagery.

Speaker 4: Zooming out is even more clear that similar mapping approaches on different images can yield different results. but each helps improve and update our understanding of the extents of blue carbon systems in this ecotone region. Now let's compare our new area estimates to some of the best global and regional data layers available. So here we can see that we're within range of state estimates from a few years ago in 2017, and we're adding information that's supplemental to the global layers. And so, with this confidence in our extent estimates, we can then begin to look at carbon stocks.

Speaker 4: Now, simply using Tier 1 global estimates of 739 carbon per hectare in mangroves and about 334 carbon per hectare in salt marsh, we get an estimated 1.6 total blue carbon

Speaker 4: teragrams from high resolution imagery and about 1.41 total teragrams of carbon with 10 meter sentinel. Now the total blue carbon here we've learned is mostly from salt marsh because of its larger extent in the landscape. Now with improved mapping, we can reduce the uncertainty in global area estimates of mangroves salt marshes and seagrass to continue to improve these blue carbon estimates and reduce uncertainty at the global scale. And that's really what's needed to go from these tier one approaches to tier three approaches.

Speaker 4: The ongoing needs to combine readily available data that we've learned about today on the extent, height, biomass and carbon stocks and fluxes from the world's blue carbon ecosystems. That's really how we can go improve our blue carbon accounting estimates from these tier one estimates using global values to tier two estimates using country specific data to a tier three approach where we're using really advanced methods and detailed country specific data say in the form of gridded data sets on blue carbon that will really help us capture all of the variability in blue carbon.

Speaker 4: So all of this information is really needed to help get us to these advanced methods for blue carbon accounting. So there's much work to be done using advancements in mapping technology to improve area estimates, first of all. And we can then improve uncertainty in these estimates as we measure more things on the ground so. Where ecosystems are changing what their carbon stocks are above ground and below ground what fluxes of carbon are they. either letting go of or drawing down every day. So altogether, these things are really needed to combine high resolution spatial information on those extents, stocks and soils, above and below ground biomass and fluxes for a complete picture of blue carbon accounting.

Speaker 4: So these components are all necessary for a full understanding. And another part of this is treating blue carbon ecosystems as interconnected networks in a way that will expand our understanding beyond just the saltwater intertidal ecosystems and how blue carbon moves between them. And so, in summary of this section, with improvements to observation technology, like the improved spatial, temporal, and spectral resolutions of our Earth observations, as well as the advancements to analytical tools like we've shown you here today in the many approaches to using and classifying remotely sensed data, Together, these can help us parameterize the various components of blue carbon, including extents, biomass, height, carbon stocks, but also cycling or carbon fluxes.

Speaker 4: Ultimately, overcoming technical and scientific barriers to measuring blue carbon while addressing the social, governmental, and financial considerations will together be our only chance to rely on blue carbon ecosystems as a natural solution to climate change to benefit our lives and maintain the health of our coastal environment. So thank you all for your attention and interest in this section on co-occurring blue carbon ecosystems. If you find yourself researching coastal ecotomes in the future, please feel free to reach out and get connected with me.

Speaker 4: Best of luck in mapping blue carbon, everyone. Thank you.

Speaker 1: A special thank you to all of our presenters today. Let's review the key points from both parts of our training. One, blue carbon is carbon that is captured, removed, and stored by ocean systems, both.

Speaker 2: In the biotic and abiotic components.

Speaker 1: Mangroves, seagrass, and salt marshes share the following traits, presence of high carbon stocks, evidence of long-term carbon storage, and the capacity for people to manage and effectively measure greenhouse gas emissions, and removals resulting from changes to those ecosystems. Blue carbon has climate mitigation value, but mangroves, sea grasses, and salt marshes have many other functions and co-benefits in the form of ecosystem services. Urgent action is needed to conserve, protect, and restore blue carbon ecosystems, and we need to develop effective policies that can support this.

Speaker 1: Global datasets are openly available that show mangrove extent, canopy height, and biomass. We introduced basic criteria for assessing the suitability of existing datasets for your purposes and how to use Google Earth Engine to generate your own mangrove data extent data. We introduced basic criteria for assessing the suitability of existing datasets for your purposes and how to use Google Earth Engine to generate your own mangrove extent data. We also explored how to estimate mangrove canopy height, biomass, and carbon stocks in your area of interest and know what data is necessary for more precise mangrove ecosystem.

Speaker 2: Carbon stock estimates.

Speaker 1: We saw examples of using spatial data help us parameterize the various components of blue carbon, including extents, biomass, height, carbon stocks, but also cycling or carbon fluxes. Finally, we demonstrated how to map the extent of salt marsh and seagrass ecosystems using satellite observations. We covered approaches and considerations to calculate carbon stocks of mapped salt marsh and seagrass ecosystems. And we explored synthesis methods to estimate blue carbon across ecosystems. A quick reminder, this training has one homework assignment and it is available now on the training webpage and answers should be submitted via Google Forms by December 19th.

Speaker 1: Certificates of completion will be issued to participants who attended both live trainings and submit the homework before the 19th. Certificates will be issued by email about two months after the training. We want to again thank the great work being done that contributes to this field of research and application of Earth observations, as well as the development of this training. This slide shows the contact information for myself and today's speakers, as well as links to the RSET website, YouTube channel, and Twitter or X, and our sister programs, Develop and Survear.

Speaker 1: Thank you very much. And now we'll move over to our question and answer session. OK, so I guess we'll just go ahead and start with the first question. And for our guest speakers here today, if one of these questions were addressed by you, please feel free to unmute and answer. I'll be placing links that you provided in these answers into the chat as well. And yeah, let's just take this away. Oh, and I just want to let everybody know, please feel free to contact the guest speakers if you have any questions going forward after the session.

Speaker 1: Question one, are there libraries of spectral signatures of salt marsh species?

Speaker 2: Yeah, hello, this is Anthony. I think I answered this one. So yeah, the USG Spectral Library does have some salt marsh species in there. It might not be comprehensive, so you might have to do kind of a data search for other locations where you can find additional sources, but they do exist out there. Great.

Speaker 1: Thank you very much. Question two had something to do with having some errors in the code in GEE. I'm trying to run a code of four six-step mapping over the entire area of Cote d'Ivoire, but GEE keeps returning the following message, memory limit exceeded. Any suggestions on how to resolve this?

Speaker 2: Yes, I mean, just chunking your code, this doesn't sound like it's unique to this training, but you could have been running into it with us too, but doing sections of the country could probably address that issue. Yeah.

Speaker 1: Yeah, it sounds like it, you know, something with a memory limit, you're probably trying to feed it too much information at once.

Speaker 2: Okay.

Speaker 1: Okay. Question number three, can you share some more information about the methodology validation for historical images? Maybe a link to some research articles?

Speaker 2: Hello, everyone.

Speaker 4: This is Cheryl Dowdy. So I answered this one, and I provided this source that I like to share. It's by Olofsson et al. from 2014. And they have some really great best practices for doing an accuracy or a validation assessment of a land cover change analysis. So it's really all dependent on the imagery you have and the data you have and how far that extends back in time, but definitely check out that resource to get started if you're interested in doing a good validation assessment.

Speaker 1: Thank you for the additional information down there. Is it possible for salt marsh seagrass and mangrove to overlap on.

Speaker 2: A single map?

Speaker 4: Yes, so this question came in a little bit before the last section on co-occurring blue carbon ecosystems, so hopefully we covered this. Yeah, and that one example from Florida, we kind of walked through just three global data layers where they're co-occurring, but depending on where you are in the world, as we saw with the different distributions of these ecosystems, they may or may not be where you are, but definitely check out, and if they're not mapped yet, have fun mapping them.

Speaker 1: Thank you. Question number five, is there any index to identify salt marshes?

Speaker 2: Yeah, so I think I answered this one. So, there are, you know, some particular, you know, basically any index can be used and useful in salt marshes. And some of the very common ones like NDVI, NDWI. There's a lot of papers going into, you know, what ends in what.

Speaker 2: most useful for estimating different things, but it's going to depend on what your research question is and where your study site's located. So, it's best to just test and see what you find.

Speaker 1: Okay, yeah, MDVI and NDWI, good place to start. This next question, it referred to a slide. I'm not sure exactly what slide it was. But the question is, how can we estimate carbon in soil that you explained in the last slide? And this also came in a little bit earlier, so it's possible that the presentation covered some of that later on.

Speaker 2: Yeah, so I assumed this was the slide where we just showed the spatial soil organic carbon maps, and that was data from this Maxwell et al. paper that someone in the chat pointed out had been published. When we recorded it, it was preliminary. So that's available now, and the data set has been available, so you can download that and look at how it works for your study site. I think in their model, they use a variety of satellite and other parameters to estimate soil organic carbon. So, yeah, hopefully that's useful.

Speaker 1: Thank you for pointing them in the right direction. Question seven, for salt marsh and seagrass mapping, how can temporal variability For example, tidal cycles and vegetation seasonality be accounted for in classification models to minimize errors?

Speaker 2: Yeah, so this is a great question, and I point you to some sources and algorithms that have been developed to do just this. The first source there, the decode algorithm, is on GitHub, and I think it's implemented in MATLAB. So if you're comfortable with that, it's a time series algorithm that takes into account both the seasonality of these ecosystems and the tidal cycles. And then the other approach is for filtering out tidally inundated data and also quantifying how much data is inundated or how many scenes are inundated.

Speaker 2: So those are both appropriate approaches for doing just what you've asked. Thanks.

Speaker 1: Thank you. What specific challenges arise when distinguishing salt marsh and seagrass in areas of overlapping vegetation or high turbidity? And how can remote sensing tools address this?

Speaker 2: Yeah, so another great question. So one of the problems here, and I think we kind of talked about this in the presentations, but is that, you know, we're mapping with discrete boundaries, and these discrete boundaries can be important for quantifying things like carbon. But if you can do a mapping approach that allows for overlap, either fuzzy boundaries or maybe a probability of the ecosystem occurrence, those are great ways to kind of account for this overlap. And then there's also, you know, for the high turbidity question, just more frequent revisit time.

Speaker 2: We kind of touched on this. So you'd have the potential to have a clear image a clear water quality or an image where the water quality is clear and high enough that you might be able to map inundated, submerged seagrasses is kind of your best approach for that.

Speaker 1: So increasing the probability that you're going to get a clear image, yes. Okay, great. Are there any examples of seagrass mapping by sourcing images from PlanetScope's Dove satellites?

Speaker 2: Yeah, I think there's lots of work in just doing exactly this. I have two examples here, one from the seashells using the Nixie, so the open source tropical PlanetScope data. And then the Chesapeake Bay has done some great mapping where they're moving towards using PlanetScope, and they've historically done it with aerial images for decades. And so they're intercomparing.

Speaker 1: Those two approaches. Quick follow-up on that. PlanetScope data, is all that open source or just some of it?

Speaker 2: Not open source. There's different approaches to acquiring PlanetScope data. You can request it through ESA. If you're funded by NASA, you can request it through NASA. And then I think the Nixie data are mosaics in the tropical zone that are free and open source.

Speaker 1: Okay, thank you. Question number 10, how can uncertainty in carbon stock estimates from global data sets be reduced to better align with localized management practices and reporting requirements under frameworks like the Paris Agreement? I love it. We're talking policy now. I'm not sure who answered this one, but yeah.

Speaker 4: Yeah, Cheryl here jumping in. I agree. This is a really... Great question and essentially where we're headed with a lot of this blue carbon research that we're doing so. I mean wetlands, especially coastal wetlands. These are really uncertain in our global carbon budgets and that's why a lot of us are continuing to work to improve mapping the extent of these areas across the globe. and adding this other information that really helps us get a holistic picture of, you know, blue carbon, but also how it changes in a system.

Speaker 4: And so it's capturing this variability across different levels that's really going to help us better parameterize our own estimates of carbon, say, at a country level. So if we're instead of using one estimate of blue carbon for your entire country, if you can map these ecosystems and get a better idea of the true number and the kind of range of variability that comes with that number, that puts you in a better place to have a more accurate and less uncertain reporting for the population. for these reporting requirements.

Speaker 1: Great. And hopefully, you know, those in attendance over this webinar series will be employing some of these approaches to do just that themselves. So, thank you. Question 11. In areas with overlapping blue carbon ecosystems, such as salt marsh mangrove ecotones, This kind of goes back to the question about overlapping, but to accurately delineate boundaries that ensure correct carbon counting.

Speaker 4: Yeah, so this is all about mapping with the best available data and technology that we can. So with higher spatial and temporal resolutions, we're going to hone in on these more certain estimates.

Speaker 1: And you say spectral resolutions. Is something like PACE data going to be something that will be useful in these areas?

Speaker 4: Yeah, I can definitely see how that would be helpful in mapping sea grasses and things like that as well. So anywhere where that added spectral information will help us better separate out these land cover classes of ecosystems or even different vegetation types or species within these ecosystems. Yeah, we're all using the improvements to the spatial, temporal, and spectral resolutions to improve how we map a lot of things.

Speaker 1: Yeah, and for anybody who's watching that's not familiar with PACE, PACE is a mission that launched earlier this year, and we started getting data in, and it's amazing the spectral analysis that they can do with the sensors. Okay, so 13, how can Earth observation tools be used to monitor fine scale changes in salt marsh and seagrass dynamics that are sensitive to climate and anthropogenic stressors?

Speaker 4: Yes, so again, it all comes down to the data that's available to us. And so I wrote that the trick to monitoring is having satellite imagery over time. So a lot of us end up using Landsat because it takes us all the way back to the 80s in some cases. Sentinel imagery at 10 meter resolution is a little bit newer, so that probably came out within the 2010s, whereas planet data is even newer than that. So as we increased our spatial resolution, these technologies have become more available over time, so they don't extend as far back in time.

Speaker 4: So again, we're just limited by what's available, but definitely check out Planet Imagery and its student accounts for getting access. It's some of the higher resolution data that's available with a good temporal resolution too. And if you're interested, check out Bullock et al. There's a link there. And Anthony earlier mentioned an algorithm called CODED, I believe. So these are all examples of algorithms that take these really high-density time series information to help us look at land cover changes.

Speaker 4: So check that out.

Speaker 2: Thank you.

Speaker 1: I realize that I skipped question 12, and it was talking about Google Earth Engine mapping, carbon accounting, and model accuracy when applying these tools in regions with sparse in situ data. And it looks like there's another article here by Olson on doing some validation analysis. And I will drop that link into the chat. And by the way, we'll have this up. Hopefully, this transcript cleaned up. and posted for accuracy on our training web page within.

Speaker 2: About a week.

Speaker 1: So, you'll have all these available to you as well. Okay. Well, I know we're over time, so we're going to address a couple more. Anthony, Cheryl, do you have to run, or can we take a few more?

Speaker 2: I can take a few more.

Speaker 1: Okay.

Speaker 3: Yep.

Speaker 4: I'll be here as well.

Speaker 1: Okay. Thank you. Question 14, how is L-band data from the upcoming NISAR going to help out in the estimation of blue carbon stocks? Will it be polarimetry or interferometry that will be the game changer? Also, the S-band data will mostly get bounced out of the top of the canopy, I guess, so what information on biomass can we get from it? It looks like this person is pretty familiar with star data, and we are all looking forward to NISAR first launching, but then also giving us massive amounts of data.

Speaker 2: I took a stab at answering this, just with the caveat that I'm not a SAR expert here, but I do think, you know, NYSAR is going to be hugely important, and I think we're mostly talking about mangroves here, both for mapping mangrove extent. You know, synthetic aperture radar has been hugely useful for that, but also mapping the top of canopy, like the person posing the question suggests, has also been used in a couple instances to map mangrove height globally and then also get our biomass estimates from allometric equations based on that mangrove height that's mapped.

Speaker 2: So, since we kind of know that these are basically the, you know, the mean sea level is basically where these ecosystems start, we can use that to quantify the height of them with the top of canopy measurement. So, it's going to be super useful and I'm excited to see what kind of applications get NISAR gets used for in the future. Thanks.

Speaker 1: Well, we certainly will be doing some trainings on NISAR data. We just recently had one that last month that you could look at to get prepared for NISAR. They talk about it quite a bit.

Speaker 2: Question 15.

Speaker 1: Why don't you check the petrotrophic activity in the salt marsh? Is there a GEE function for that?

Speaker 4: Yeah, this is a really interesting question and it kind of shows all the various applications of remote sensing we can use. And so, for this example, you know, if you wanted to look at heterotrophic activity using some similar approaches to what we've seen today, you'd be pairing your data from the field with the imagery. And usually the first step in doing that is developing some algorithm that relates your metric of that activity to the satellite data. Now, I might have misunderstood your question.

Speaker 4: So conversely, you're measuring the impact of that heterotrophic activity on the landscape. If it causes a change that's detectable with satellites, that might be a way to kind of monitor the impact of this. But no, there's not really a straightforward function in GEE to do that. You would probably have to develop that on your own.

Speaker 1: Thank you. And this next one is here for you as well, Cheryl. Are we calculating a blue carbon capacity zone or blue carbon storage or unit for each ecosystem?

Speaker 4: Right, so in these examples we're estimating those carbon stocks based on the best available information of the ecosystem extent. So whether we're getting those from the globally available data sets or we're mapping them on our own. I'm relatively new to the greenhouse gas and global carbon accounting applications of this work, so I might not understand what you mean by a blue carbon capacity zone. But right now, essentially what we're doing is trying to measure those carbon storage as a single unit in time for each of those single snapshots of ecosystem extent in time.

Speaker 4: So hopefully I answered your question.

Speaker 1: Thank you.

Speaker 2: You know what?

Speaker 1: I saw that you guys were filling in answers to these questions earlier, and For our last two, is there a question that you answered that you thought we should move up the list and address here before we leave today? Either it was very interesting or presented an application you didn't think of or thought it was just some great information to impart upon the participants here today, either Anthony or Cheryl. And if not, I can just keep going down the list.

Speaker 2: I think all the questions are great, but I don't have one in particular. Okay.

Speaker 3: All right.

Speaker 1: Well, I'll do this number 17 here. Are there tools to measure and monitor carbon fixation and fluxes in the ocean ecosystems on a global or regional scale for carbon accounting?

Speaker 4: Yes, I answered this question and I might have misread the keyword ocean ecosystems there, but as I see it, it would be similar to what I've been mentioning in terms of this upscaling approach that we've been seeing today, where we're taking data from the field and we're pairing it with the remotely sensed imagery in order to upscale those, you know, ground or ocean estimates to predict over a larger area. But in terms of measuring and monitoring carbon fixation and fluxes in the ocean, that might be a little bit outside of my wheelhouse, but I know that in coastal wetlands like the salt marshes and mangroves of Florida, we're working to take flux measurements from the field and pair that with satellite imagery to do something similar on a larger scale.

Speaker 4: So there are lots of different ways to try to use the remotely sensed data for your purposes. And if I haven't answered your question, hopefully we can point you to some better sources on ocean ecosystem carbon fluxes.

Speaker 1: Okay, thank you. Thank you everybody for staying a little bit later than our usual time here, but we really appreciate your attention in this webinar series. Is there anything, Anthony or Cheryl, you'd like to say to the group before we close out?

Speaker 2: We appreciate you coming and hope you enjoyed the training. Thanks so much for everything.

Speaker 4: Yeah, thank you, everyone. And please feel free to connect with me, especially if you're interested in those co-occurring coastal ecosystems. And good luck mapping your blue carbon ecosystems, everyone.

Speaker 1: Yeah, I think we had quite an informal community of practice meeting here from around the world. Usually, we have over 100 different countries represented in these trainings. So, and I saw people sharing their information in the chat. So, connect with each other, and let's address this together. Thank you very much, everybody, for attending, and thank you very much to our guest speakers for their time in this training. Have a nice rest of the year, everybody.

The Story