NASA ARSET Estimating Biomass using GEDI

0:01 Speaker 1: Welcome to this RSET training series, Estimating Biomass and Change with JEDI and the obi IAPI. Today is session one of our training series and it's focused on estimating biomass using JEDI. My name is Erica Portis. I'm a scientist at NASA's Jet Propulsion Laboratory in Pasadena, California, and I focus on the study of terrestrial ecosystems using SAR. I'm also an instructor with the RSET program and I will be hosting this training. Before we begin today's training, I'll say a few quick words about the r SET program.

0:42 Speaker 1: The NASA Applied Remote Sensing Training or our SET program, offers free trainings on the use of remote sensing and model data, along with methods and tools for the global science and applications communities. Our trainings are tailored to different experience ranging from introductory to intermediate to advanced. We provide trainings in several thematic areas including agriculture, climate and resilience, disasters, eco conservation, health and air quality, water resources, and wildland fires. Our SET trainings are either online or in person, and there are two modalities.

1:26 Speaker 1: One is live and in structure. Lad like today's training, or the other modality is asynchronous and self paced. All trainings are free of charge and utilize open source data and software. Some of our trainings are delivered in both English and Spanish like this one, and all of our trainings have a Spanish presentation available a PDF presentation available. So let's start today by providing an overview of this training series estimating biomass and change with JEDI and the OBIEAPI. What is happening beneath the canopy?

2:11 Speaker 1: One fundamental question that many people are interested in is forest structure and what is happening beneath the canopy. Now, from many traditional satellites, we do not know this because optical remote sensing sensors can tell us things about, for example, the greenness of the vegetation, but not about the understory or the structure. Spaceborne lighter can do this, such as JEDI, which measures forest structure across more than twenty billion locations on Earth, and you'll learn to use this data how to do this in this training.

2:50 Speaker 1: So by the end of this training, attendees will be able to identify the fundamental characteristics of JEDI data and carbon monitoring data structure, uncertainty, and system requirements. Apply methods for estimating biomass and biomass change, including the integration of Jedi with Lancet time series and calibration with forest inventory data. Evaluate the accuracy and uncertainty of biomass estimates using validation data. Use the Obian tools and APIs to generate and analyze biomass change products, including visualizing results for user defined areas and comparing carbon gains to different climate scenarios.

3:39 Speaker 1: And finally, evaluate advantages and disadvantages of the footprint level versus gridded level JEDI biomass products to determine which data are appropriate for a given use case. The prerequisite for this training is the r SET Fundamentals of Remote Sense as well as the Spaceborne Lighter for monitoring vegetation, structure and biomass using Jedi and this was a course from October of twenty twenty five in our Set training and below that is a list of suggested trainings. This training series consists of two sessions.

4:22 Speaker 1: The first session today is focused on estimating biomass using Jedi and the second session will be next Thursday, May twenty eighth and it will be focused on estimating biomass change with Jedi and Thebian API. The homework opens up on May twenty eighth and it will be due on June eighteenth. It will be posted on the training web page. There will be a certificate of completion to participant who attend all live sessions and complete the homework assignment by the do date.

5:01 Speaker 1: So with that, let's begin with Session one Estimating Biomass using Jedi. We have a guest instructor for this training, doctor Sean Healey, who's a research ecologist with the US Forest Service and US Department of Agriculture. So by the end of today's session, participants will be able to identify the fundamental characteristics of Jedi data and carbon monitoring, including data structure, uncertainty, and system requirements. Explain that Jedi structure and how uncertainty is characterized. Demonstrate how Jedi data are fused with Lancet time series and calibrate it with forced inventory data to estimate biomass change, and finally, apply Jedi estimation methods to visualize and generate biomass products for given user defined areas of interest.

6:02 Speaker 1: How to ask questions Locate the three points on the bottom right a sub menu, click on it. A sub many will open and select Q and A so window will open and write your question in that window and we will address your questions at the end of the webinar. Feel free to submit your questions throughout this webinar. We'll do our best to address as many questions as possible during the live Q and A session. Any remaining questions will be answered in a Q and A document which will be posted on the training website approximately one week after the training.

6:41 Speaker 1: So let's get started. This next part is going to be presented by doctor Sean Healey. We're very fortunate to have him, so welcome doctor Healey.

6:56 Speaker 2: Hello, I'm Sean Heally. I am speaking for doctor Jijiang Yang. We both work for the USDA four Service, the Inventory and Monitoring Program. I am very happy to be with you today to talk to you about JEDI. JEDI is an acronym. It stands for Global Ecosystems Dynamics Investigation. It is a NASA mission. It's a light our instrument that is on the International Space Station. I'll talk a little bit more about that, but before I do, I want to clarify. So I will be taught starting with JEDI, but I will also be talking about obi Wan, which is another Star Wars reference.

7:37 Speaker 2: It's another acronym, it doesn't really matter what it stands for. But basically, obi wan is an application that is specifically designed with the requirements of carbon monitoring in mind. So we're going to start with the Jedi mission and we're going to be going we're going to be phasing into the obi Wan tool. And I say phasing because obi wan uses observations from the Jedi lidar, it uses statistics built into Jedi, it uses carbon models built into Jedi, so it's really it's a subsidiary tool of the Jedi mission.

8:19 Speaker 2: We're going to start with the theory and measurements behind Jedi, and between this presentation and the next one there is a second version which you will see.

8:30 Speaker 3: We will get into how obi Wan helps us.

8:37 Speaker 2: Answer real questions in the field of carbon monitoring. So starting with Jedi. It is a NASA Earth Ventures instrument. It launched in twenty eighteen and became operational in twenty nineteen. On the picture on the right there there's not actually a on it that says Jedi. I've added that, but that's basically what it looks like. It's the size of a refrigerator. It is a laser.

9:08 Speaker 4: There.

9:09 Speaker 2: Actually there are three lasers on Jedi, and they help us measure the basically the height of trees around the world, or at least in the latitudes that the International Space Station overpasses between fifty one degrees north and fifty one degrees south. There is another rset training that goes over more of the specifics of Jedi how to and focuses on how to get Jedi data. We're going to be talking today really about how Jedi makes inferences from those measurements related to how much biomass is in particular places.

9:49 Speaker 2: And that's on the way to talking about the Obi Wan tool. I should say that it became operational twenty nineteen, but it took one year nap between March twenty three and April twenty four. It has since resumed operations and the PI is rapped Dubai at the University of Maryland. I should also say that NASA very recently selected the Edge mission, which is similar to Jedi, except it's free flying. It will be free flying. It should launch around twenty thirty. NASA selected that through the Earth System's Explorer program.

10:29 Speaker 2: The pis there are at UCSD.

10:35 Speaker 3: And the University of Maryland.

10:37 Speaker 2: So the technologies that you're seeing here will be available into the future.

10:44 Speaker 3: Okay, so what does JEDI do.

10:47 Speaker 2: It really has three lasers, but those lasers are split and dithered in such a way that it really gives us eight tracks. As the International Space Station goes over an area, it's collecting eight tracks worth of.

11:03 Speaker 3: Height data.

11:04 Speaker 2: And as the mission has continued to go on, you can see that the sample is filling in. So it's not an imager. It is collecting footprints along tracks. And the tracks are not contiguous, they're interrupted. But so there are thirty meters footprints along all of those tracks that have been collected since twenty nineteen.

11:29 Speaker 3: And what do these footprints look like?

11:31 Speaker 2: As I said, they're about twenty five meters in diameter. And really what you get the main observable from JEDI is this representation of the vertical forest structure. So as laser energy comes down from the space station, it bounces off of the surface. Some of that energy bounces off the top of the canopy, some of it makes it all the way down to the ground. And you know, because we know the speed of light and how long it took different parts of that waveform to come back.

12:11 Speaker 3: We infer this waveform.

12:14 Speaker 2: The business end of the waveform is really when you quantize that to things that we can put into a biomass model. So relative height our H is basically the height above where we found the ground a particular percentage of the waveform's energy is returning. So our H one hundred is basically the top of the canopy. URH fifty is the height at which half of the stuff that light our energy is returning from. Half of the stuff is below and half of the stuff is above. And so basically we quantize the waveform into things, into these measures that we can put into a biomass model.

12:58 Speaker 2: There are a lot of jet well, there's four levels of Jedi products. It's not it's not a bewildering array of products. The level one two, especially the level one and two are are sort of more basic, uh physics oriented waveform properties. Level four is really where we're going to focus today. That Level four A is how much biomass for every footprint has a biomass estimate. So level four A is talking about how how we model using those relative height measures, how we model biomass. And Level four b is how we take the sample the Jedi gives us of those predicted biomass values, how we fill in the entire area for one kilometer grid cells.

13:55 Speaker 2: So that's really a sample survey sort of a process.

13:59 Speaker 3: And we.

14:01 Speaker 2: Very similar statistics to those used by the forest inventory that I actually work for.

14:07 Speaker 3: So let's just jump right in.

14:12 Speaker 2: How do we get our predictions for biomass at the footprint level. Basically, the Jedi Mission has asked cooperators from around the world for light our data that's collected directly over inventory plots where biomass has been measured on the ground, and throughout this presentation you will see references here it's Duncanson at All twenty twenty two to the source material if you want more details about the processes. But basically, what the Jedi Mission has done is simulate Jedi waveforms from the airborne light are and used the paired Jedi waveform and field data to make a model, a specific model for every continent, and every continent is split between plant functional types deciduous, evergreen, et cetera.

15:10 Speaker 2: So, and important for our purposes, where we're headed is estimating biomass and inference. Where we're headed is we are keeping the uncertainty that's associated with these models, and we won't get into exactly how that's stored, but there is an uncertainty stored with every model, and we use that in this next slide. Actually, so when we go to L four B JEDI L four B, that's the gridded biomass product, where we are inferring the mean biomass for every one kilometer grid cell. What we're doing here, I mean it looks like here we've got a sample, and that's exactly what JEDI gives us.

16:01 Speaker 2: So every one kilometer cell, not every one kilometer cells. Some one kilometer cells are unlucky, either because of the overpassed pattern of the space station or because they're perpetually cloudy. Some one kilometer grid cells don't have much of a sample, and those grid cells get a different algorithm, which I'll talk about in the next side. But for those cells, which is most cells that do have a good JEDI sample, really what we do is we use an estimator, a statistical estimator that estimates the mean biomass from all the predictions that are in that grid cell, and we also produce an estimate of uncertainty a standard error around the estimated mean.

16:53 Speaker 2: That is a function both of the model uncertainty. If you remember from the previous we saved a lot of not a lot. We saved the relevant model uncertainty information, so that gets propagated through this process and also a sampling uncertainty. As you can see, we're not measuring everything in that one kilometer grid cell, so this estimator Patterson at all uses propagates both sources of uncertainty to get a standard error for four our estimate of mean biomass in one kilometer grid cells around the world.

17:34 Speaker 2: So that's the L four B product. I mentioned that there are some L four B one kilometer grid cells that don't have, for whatever reason, enough.

17:48 Speaker 3: Enough Jedi measurements in there to.

17:51 Speaker 2: Do that process, so this is what we do in those cases.

17:56 Speaker 3: The green squares.

17:58 Speaker 2: Are field plots around the world and that are used by the Jedi team to make our biomass predictions. So we've already talked about the field to Jedi model. So there's one model, but then there's then we predict. We use Jedi's predictions for all the waveforms in a particular area. We're using currently we're using ten kilometer squares ten by ten kilometer squares, and we intersect those predictions of biomass with lance At time series information, So not just a single lance At image, but all of the available clear imagery for a lance AT pixel gets fit into this time series function, and we use properties of that function to make our prediction of biomass with a model that we calibrate with Jedi's predictions, so very local model.

18:57 Speaker 2: We're using Jedi predictions to predict a Lancet model.

19:04 Speaker 3: Okay, So, and we've talked about where that cheese.

19:07 Speaker 2: This is important for obi Wan because this really is the basis for how.

19:12 Speaker 3: Obi Wan works through time.

19:14 Speaker 2: And we'll come to that in just a moment, but before we do, i'd like to show you.

19:21 Speaker 3: What the L four B product looks like.

19:25 Speaker 2: You will be seeing a little bit later in an app that I will be walking through and that you will also have available to you. You will see that there are a lot of one kilometer grid cells in version two, which is what the app is seeing, that do not have an estimate. Version three, which has been submitted to the DAC and will be actually I think it is available currently, fills in all of those gaps with.

19:56 Speaker 3: The process that I just talked about.

19:58 Speaker 2: But basically every place in the world has a prediction at the one kilometer scale of biomass from JEDI, and there's also a corresponding standard error of that estimate map. So this process doesn't have to only apply to one kilometer grid cells.

20:16 Speaker 3: It can apply to entire countries.

20:20 Speaker 2: In this case, what we have done is pointed that process at sixty four thousand hectare hexagons that cover the United States, and in this paper we compared those estimates with estimates from FIA, that's the Forest Inventory, the part of the Forest Service that I work for, and really there's a pretty good match spatially, you know, spacially and looking at the scatterplot. So we'll talk a little bit more about validation with FIA, especially in the next presentation, but in general it lines up. And the important thing is that it doesn't have to be a one kilometer grid cell.

21:06 Speaker 2: It can be you know, any shape, and it can be multiple multiple patches too.

21:15 Speaker 3: So now we're.

21:15 Speaker 2: Getting into obi WAN, which is a tool for estimating biomass change, and change is an operative word in this slide because so far all we've talked about with Jedi is single point in time biomass. But we will see in the next version of this presentation, the next session of this presentation that practical carbon accounting.

21:46 Speaker 3: Really requires a view over time.

21:50 Speaker 2: And so obi Wan is using of the measurements, the statistics and the models that we've already talked about for Jedi, but pointing them directly for the purpose of estimating change. Okay, so you will remember this that this is sort of the alternate algorithm for predicting biomass for a single point in time for the one kilometer grid cells where there is not enough Jedi, not enough Jedi footprints directly in the one kilometer cell. So what we're doing there's two levels of model, one from the field to Jedi and then one fitted very locally at ten kilometer by ten kilometer area Jedi predictions to lance At imagery.

22:45 Speaker 2: And the nice thing is that lance At is consistently processed through time. And so what we do then is we can apply that lance At model not only to the present but to lance At imagery. And again it's lance At time series imagery going back in time to the mid eighties if we want to.

23:09 Speaker 3: So what we end up with is.

23:13 Speaker 2: A time series of predictions through the Lancet record. Lance Modern Lancet started in the mid eighties. The first lance At launched in nineteen seventy two. So what we are doing here in this slide, this is the country of Paraguay. We have fitted our models with JEDI in the current period, and then we are applying that lance AT model back in time. And what we see through time is that there's a lot of reduction of biomass, especially in the western part of fireway through this time period. Okay, so this is how we're getting basically, how we're getting estimates of change.

24:01 Speaker 2: So when we're talking about a company applying for carbon applying to be to sell carbon credits because of forest management, or we're talking about a country complying with its own commitments internationally in terms of mitigating climate change by managing forests in a way that stores more and more carbon, we really need a straightforward way of accounting for the uncertainty of this car of any carbon management credits that we may claim.

24:41 Speaker 3: And so.

24:43 Speaker 2: You might ask, we've got a time series of lance AT predictions, how do we track.

24:49 Speaker 3: The uncertainty of.

24:52 Speaker 2: The predicted change for a particular for an area that we care about, and in this slide, it looks like we care about a square. You know, it could be as I said earlier, it could be any shape, it could be multiple areas from across, from across the region. We do require that it be at least five hundred hectares because the statistics that we're using, some of the assumptions break down for smaller areas than that. But basically the top set is basically what I just showed you. We make two models feel to Jedi Jedi to Lancet.

25:35 Speaker 2: We apply the model through time, and let's say that we care about the change between nineteen ninety and twenty twenty. We get a prediction in ninety of ninety five, and thirty years later it's eighty tons per hectare, indicating there's a loss of fifteen.

25:53 Speaker 3: So what we.

25:54 Speaker 2: Actually do behind the scenes in obi Wan is change both of those models feel the Jedi Jedi to Lancet in ways that across a lot of iterations represent what we know about the uncertainty of those models.

26:12 Speaker 3: And we call that bootstrapping.

26:15 Speaker 2: And I'll talk a little bit about how we sort of jitter those models in a second. But basically we end up making a lot of different time series of predictions, and the first one indicated a loss of fifteen, the second one indicated a loss of thirteen, third a loss of twenty, and across a lot of these iterations or bootstraps, what we will end up with is a distribution of predictions and that's what you see on the right, and the mean of those predictions, that's what we return as an estimated mean, and the standard deviation around that mean is what we.

27:00 Speaker 3: Support as a standard error of our estimate.

27:04 Speaker 2: Just a little bit of detail about how we are jittering these models this you know too.

27:11 Speaker 3: Two models.

27:13 Speaker 2: The first one we have we have because the Jedi mission provides it. We have it's a linear model that goes from the field to field field biomass to Jedi biomass UH, and we have a parameter covariance matrix for for the parameters in that linear model. We can we j We jitter that model on that basis, on the basis of the uncertainty that that is implied by those parameters. And then we also I won't go into the details, but we also re sample the second level of model ah in in a way that gives us a lot of different time series that we need that we query to to return a predicted change and standard around that change.

28:02 Speaker 3: Okay, so here's sort of a.

28:07 Speaker 2: Red flag that is that may be on your mind, especially if you do a lot of remote sensing.

28:13 Speaker 3: So lansat imagery is optical imagery.

28:17 Speaker 2: And as forests grow, let's say they're growing from scratch, you know, from from an open field. In general, the more the forest expands, the greener the imagery gets until the canopy closes. And at that point, the stuff under the canopy continues to grow. The trees get taller, they put on more and more biomass. But there the imagery does not get a lot greener. There's some shadowing the changes. There's still is some signal there. We have shown that if we make our models very locally it helps, but it's still a problem.

28:58 Speaker 2: Basically, that problem manifests itself in our models. Overestimate low values of biomass and underestimate high values of biomass. So that's a bias right there. And then, especially when we're looking at change from either high biomass to low or in the opposite direction, we will systematically be underestimating that change.

29:26 Speaker 3: That's a problem.

29:27 Speaker 2: So what obi Wan does is very explicitly use calibration to address that problem. And we will be talking about two options for calibrating in the next session. The first one is using Jedi waveforms that we didn't use to make the lance At model, and the second is using inventory data that may be available in a particular country. Both of those basically what we do is we look at the predictions that we see in these time series and we develop a really simple calibration function that in essence, most of the time it pushes the high predictions upward and the low predictions downward, because you know, those are generally the what has to happen if we want to remove the biases that integrating that process into the process that I that I just mentioned is really simple and computationally it's very simple, and we'll see a little bit about that with so there's very little additional computing.

30:38 Speaker 2: I just want to make the point that and we'll come back to this in the next session, that there is calibration possible and in fact it is happening.

30:49 Speaker 3: Okay, So before we.

30:51 Speaker 2: Get to sort of a hands on activity that we in which we will be exploring.

30:57 Speaker 3: Jedi data.

31:00 Speaker 2: With a Googleer's Engine app. I just want to point out these resources. There is also, of course, the earlier R set trainings in this series which talked about light R in general and light R specifically from Jedi. Okay, so the hands on activity that we will be pursuing Before we get into it, we'll talk a little bit about what it is. We will be exploring the distribution of Jedi shots. You know what this when you look at Jedi data, how.

31:38 Speaker 3: Dense is it.

31:40 Speaker 2: And how does that density change from place to place. We'll be looking at what Jedi actually measures at each of its footprints.

31:51 Speaker 3: You will see very clearly that they.

31:54 Speaker 2: Line up in paths that echo the overpass pattern of this National Space Station. Lastly, we will be experimenting with the estimator that I mentioned earlier that propagates the uncertainty of the sampling, the model air and the sampling error.

32:20 Speaker 3: So I just want to point out that this.

32:25 Speaker 2: Is possible because Jedi data is stored on Google Earth Engine. Here is the thing that you would search for in the catalog, and here is the link to the Jedi app that we will be exploring. Now, Okay, so here we are. You have registered for Google Earth Engine. If you weren't already registered, you've clicked on the link, and this is a Google Earth Engine application that we built. Really it was Jijiang Yang who built this for this training. And what we're doing here is looking at a number of things.

33:11 Speaker 2: What I don't I think that you can see my cursor here. What we're looking at now is.

33:18 Speaker 3: Jedi L four B.

33:20 Speaker 2: Those are the statistical estimates at the one kilometer grid scale,

33:27 Speaker 2: so you know, just zooming in, you will see that there are places where there's no estimate. Those are places where, at least at the time of this version, which was version two of Jedi, there weren't enough Jedi footprints in there to make an estimate. So if you want to see imagery underneath what we're showing, you can just press the satellite button here, and I'm just zooming into a particular place. So these are the one kilometer estimates. I'm going to press this inspect pixels button and just clicking on a you know, a pixel at random.

34:14 Speaker 2: Pretty high biomass three hundred and fifty seven tons per hectare. We've got a standard air of ninety five tons per hectare.

34:25 Speaker 3: So this is actually fairly uncertain.

34:27 Speaker 2: There's a lot of variability in this because the standard error is maybe thirty percent of or actually it's yeah, well it's about thirty percent.

34:38 Speaker 3: That's a pretty high error.

34:41 Speaker 2: In addition to seeing so you can explore really this is sort of a tool for you to explore Jedi data.

34:48 Speaker 3: It's global.

34:50 Speaker 2: I'm focusing I think it's this is Alabama. But as you zoom in, in addition to these one kilometer estimates, you can.

34:59 Speaker 3: Choose the Jedi product that you would like to see. L two A that's a footprint level.

35:08 Speaker 2: So I'm just zooming in so that you can see the density of the thirty meter footprints. It's really pretty dense.

35:20 Speaker 2: So as we as we zoom in, you can see this looks brown.

35:26 Speaker 3: The forest here looks brown.

35:28 Speaker 2: I have a feeling that this is a sort of wintertime image.

35:31 Speaker 3: When I click.

35:33 Speaker 2: On one of those pixels. This L two A is the product that has those relative height values. So the top of the canopy is about thirty one meters tall. For the for the footprint that I clicked on fifty that is sort of the midpoint of the canopy. Stuff that intersects that intercepts Jedi is about nine meters, so there's a pretty far distance between the midpoint of all the all the photosynthetic material in the top. There is also this L two B product that's also a footprint level product. It is the predicted predicted can it be cover for a particular waveform?

36:21 Speaker 2: And this one is showing about point nine can it be covers? That really tells me this is not brown because it is.

36:30 Speaker 3: Barren.

36:31 Speaker 2: Is brown because this is a wintertime high resolution image. If you want to see what's under there, you can this slider bar opacity can make this well our Jedi data invisible for a moment. I'm just going to switch over to another instance.

36:51 Speaker 3: This is on the coast of California.

36:56 Speaker 2: I'm going here because the redwoods are here, they're very they have very you know, tall heights obviously, so I'm just going to zoom into, you know, a place here and instead of looking at the one kilometer estimates, I'm going to look at the the Jedi footprints. You can see that there's you know, there are i think twenty billion high quality Jedi shots around the world currently. I'm going to display them against forest data. So these are I mean, the redwoods are famous for being tall. Obviously, I'm going to press inspect pixel that's here, I get a cross hair.

37:48 Speaker 2: I'm going to zoom in, making sure that I hit the middle of this pixel. So we're predicting almost five hundred tons of biomass per heck.

37:59 Speaker 3: There, that's a lot of biomass.

38:04 Speaker 2: You can find some that are over a thousand tons. Keep in mind that these footprint level predictions are not necessarily representative of the whole hectare twenty five meters that might just fall luckily or unluckily by chance on one tall tree, and a lot of that thirty meters might be you know, just one column of wood. So for that small area, the density is over one thousand tons per hectare. Over the whole one kilometer or even larger area, you know, it's it's going to be some footprints that fall in between trees.

38:46 Speaker 2: This is the last thing I want to show you here. For an area that you care about, this panel over here polygon rectangle delete analyze this access is the statistical estimator that I talked about earlier that propagates the uncertainty both from the model and from the sample. And so basically I'm just drawing a pixel, and I could draw several pixels, not pixels, polygons, I could draw several of them. When I have the place that I care about, I can press the analyzed button, and over here this panel tells us that the area is six hundred and twenty seven hectares.

39:38 Speaker 2: It's estimating three hundred and fifty five tons per hectare. That's megagrams per hectare. It gives us the standard error this one. It's a pretty.

39:50 Speaker 3: Low standard error.

39:51 Speaker 2: The forest here, I was careful to not include a bunch of different.

39:59 Speaker 3: Forest.

40:00 Speaker 2: If it was really various, the standard error would be higher. There's some other information metion here. There are fifty three clusters, so each one of these lines we treat as a cluster sample. And across those fifty three clusters there's fourteen hundred shots. So the last thing I want to talk about, because this leads into the next session, is that we have built into this app a little button that says, obi Wan report we care about for whatever reason, we care about this place, and we built into this app something that gives some background and records at least what today our estimate for that area was.

40:49 Speaker 2: This could be built into any kind of data system you want. It could look however you want. We'll talk about that in the next session. I just want to point out that even in this fairly simple app, we have options for documenting the out the outputs.

41:07 Speaker 3: That that our estimators are using.

41:10 Speaker 2: This app will be live for the foreseeable future for the purpose of just letting you explore around the world the JEDI data and the capacity for estimating mean biomass that is that is built into obi wan.

41:25 Speaker 3: Thank you.

41:28 Speaker 2: So now we will hand this over to Erica who will lead the Q and a portion of this presentation, and as I said earlier, a lot of what we have hinted at in terms of carbon monitoring, we will explicitly cover in the second session.

41:48 Speaker 1: Thank you very much, Sean. And next we will do a brief summary of the key concepts that were presented during this session. So let's quickly review the key concepts for this session. JEDI is a sensor on the International Space Station. It is an active sensor that measures the distance by timing how long a laser pulse takes to return back to the sensor, and this provides detailed information about elevation canopy heights and three the three dimensional structure of the vegetation. JEDI is a full waveform lightar.

42:27 Speaker 1: Unlike discrete return systems, this sensor records the complete energy profile of each pulse, which captures the entire vertical distribution of vegetation within Jedi's twenty five meter diameter footprint and this as a result, this provides a much richer structural information,

42:51 Speaker 1: so each footprint provides the complete vertical structure of the canopy. And given the as I mentioned, JEDI is on the International Space Station, it doesn't have a regular orbit, so it covers latitudes between plus and minus fifty one degrees and produces both footprint level data at twenty five meters and gridded products at one kilometer resolutions. The relative heights are a property of the waveform that can be plugged into a biomass model, and the Level four A is a biomass density predicted for individual footprints.

43:33 Speaker 1: Level four B are statistical estimates of mean biomass for a one kilometer grid cell using JEDI sample of Level four A biomass predictions. NOWBI one is a forest canopy analysis application that uses JEDI lighter observations together with forced inventory data to estimate biomass and monitor changes in forest carbon over time. JEDI uses sampling theory to estimate mean biomass and uncertainty. Looking ahead to Session two, participants will be able to identify key concepts in carbon monitoring, including system requirements, decision making needs, and the concept of additionality.

44:20 Speaker 1: They'll be able to recognize or you'll be able to recognize how obi Wan estimates biomass change, including its use of JEDI lass time series and under the underlying data infrastructure. Participants will be able to evaluate uncertainty and validation using Forest service inventory data to assess the accuracy and precision of obi wan change estimates. Access open source ob wan API to generate estimates of biomass change and areas of interest, and apply obi Wan tools and APIs to visualize biomass change and compare carbon gains against different climate scenarios.

45:03 Speaker 1: Homework and certificates. There is one homework assignment with this training. It will open on May twenty eighth, that's the next at the end of the next training or the next session next Thursday, and you can access the homework from the training web page. The Answers must be submitted via Google forms and the homework is due by June eighteenth. The certificate of completion will be given to those participants who attend both live webinars and attendance is recorded automatically. Participants complete the homework assignment by the deadline and please expect to receive the certificate via email approximately two months after completion of the training.

45:56 Speaker 1: If you have any questions about the material that was presented to they, please feel free to contact doctor Sean Heally through the email that you see here. We've reached the end of today's session, Thank you very much to doctor Heally, and we are now ready to start the Q and A session.

46:25 Speaker 5: Right, Thanks again, thanks very really thanks.

46:28 Speaker 6: Eric, tag team and one to the spreaders and tag team in here with with Erica because she has to step out, So I'll be leading ast in the quanization here and thanks Bred for putting in there on the on the spring, we can go steck to the to the first question and and then Sean, I'll be I read the question and then feelfree to provide the answer to it. So the first question is are we and some of this questions getting really early in the during the webinars, So I'm assuming that maybe maybe you.

47:05 Speaker 5: Already answered some of them uh D during your clop. But but the first one says, are we measuring bio must change or are we using a sophisticated model based approximation?

47:20 Speaker 3: H Thanks for the question.

47:23 Speaker 2: The answer is, I guess if you give me a choice between those two things, it's a sophisticated model based approximation. I mean, it's almost impossible to measure directly measure you know, a lot of trees over a large area, so there's there probably.

47:40 Speaker 3: Is no direct measurement of biomass change, but there is. Uh what.

47:46 Speaker 2: What we're using is a model based mode of inference where we are accounting for uncertainty that is built into our models. We're also accounting for uncertainty of the introduced by the fact that Jedi is only sampling the landscape and not not collecting an image over the entire population. So the answer, the answer that I put there is that it's you know, it's a model based estimation paradigm, and the uncertainties that we produce are our best understanding of how the errors in our models propagate to the final answer of biomass change.

48:32 Speaker 5: Great, thanks Chub. Question number two, it'sa and space based biomass estimates be accurate and reliable enough to support high stakes carbone decision making. And if not, what monitor approaches would be more appropriate in cases? What comprehensive if ground based monitoring is not peaceable, can space based biomass estimation and service practical do not be complimentary approach? And to what I explain the approach is practically upblet long question there, but jump in.

49:11 Speaker 2: Yeah, two great questions right out of the gate. You know, I think that the gold standard is having a field based inventory, uh that goes back in time and is that is consistently and frequently measured. That is pretty expensive, especially if you know you don't have an inventory for anything else. It's it's hard to imagine, you know, carbon credits paying for that kind of that kind of inventory to be collected all the time. So uh, in in lieu of that, in many places, we don't have any forest inventory.

49:48 Speaker 2: And I believe that what we're talking about is the best estimate that we can get from remote sensing in many parts of the world, and I think you will UH in the next presentation. I believe that in many places that is not good enough. I mean, it's not good it's you know, there are there are definitely biases that creep in. So I although I would say it's the best we have and maybe the best available answer in a lot of cases. There we know that there are biases and that that will be covered in the next the next presentation.

50:27 Speaker 2: I would say, though, and you also see this in the next presentation, that when we in the United States we're using remeasured data from the Forest Inventory and Analysis Unit of the Forest Service to calibrate these estimates, that it's getting much closer to the truth. And I think, you know, in this country, we are entering a space where this can be used for decision support. You know that we we provide probabilistic estimates of uncertainty, which all always includes the possibility that you know, there's you know, there's some bias in there or some errors in there.

51:08 Speaker 2: But I do believe in this country, especially that in the United States, UH, that there is room for using what we're talking about in in operational carbon accounting.

51:22 Speaker 5: Excellent. Big question number three, UH, should temporary force carbon storage be treated as like we willant to permanent fossil free or emission reductions.

51:37 Speaker 2: Well, uh, you know, I I think that the answer depends on the time frame of mitigation that we're talking about.

51:46 Speaker 5: Uh.

51:46 Speaker 3: A lot of.

51:47 Speaker 2: People consider the next few decades as you know, the the real uh you know tip place where tipping points will or will not be passed.

51:56 Speaker 3: Uh.

51:57 Speaker 2: And uh, you know, these upcoming decks decades probably will be you know, see the peak of carbon emissions, hopefully.

52:05 Speaker 3: And you know a lot of people say that.

52:11 Speaker 2: Mitigation in the near future is more important than mitigation four hundred years from now. I don't have a position on that, but I do recognize what the questionnaire is asking. Like, mitigation is not in forest carbon storage, it's not permanent. You know, disturbances are part of life in forest ecosystems. So that question is more of a general forest climate change mitigation question. So I guess the answer to that is it depends on your time frame.

52:51 Speaker 5: Okay, Question number four, is ourh fifty half the elevation of our age one hundred or half the biomass of the column?

53:02 Speaker 3: So the answer is it's neither.

53:06 Speaker 2: As the light from the laser passes from above and eventually bounces off of the ground, some of that is intercepted by the very top of the canopy, some of it's intercepted by the branches and leaves at different levels of the canopy. And RH fifty is really a measure of sort of the midpoint of the density of the stuff that the light passes through and bounces back from on its way to the ground. So if you think of a profile, and I showed a profile of the waveform, that's the point at which half of the energy is returning has already returned back.

53:49 Speaker 2: So yeah, it's really a measure of the waveform. And you know, depending on the distribution of canopy material, that could be half half of the height of the very top, or it might be taller than that or shorter than that, just depending on how that material is distributed.

54:08 Speaker 4: Vertically interesting, Okay, question number five Is it possible to asswer the lighter data using the duncans on at all vapor or is it open source?

54:23 Speaker 2: In general, all NASA sponsored research is open source data. However, the Jedi Mission is collaborating with about one hundred or maybe it's more by now providers from around the world who have put a lot of work into surveying measuring trees and collecting light our data over those trees. So that's proprietary information in many cases shared you know, through goodwill. But part of that goodwill is also about shaped with agreements protecting against sharing. So as much as the mission would like to share that data, it has to respect the providers of that data and it does not share it.

55:16 Speaker 5: Hey. Number six, how do you see the potential of incorporating information from Centinel or higher resolution satellites in addition to the last model?

55:26 Speaker 2: Well, Sentinel too is that's sort of the sensor I addressed in my answer here, that's a fairly easy one. NASA provides harmonized lance at Sentinel data set.

55:42 Speaker 3: I think that globally.

55:45 Speaker 2: It may only go back to twenty eighteen, but basically they make the two sensors sort of interoperable and that would be easy to integrate.

55:54 Speaker 3: We have not done that because.

55:58 Speaker 2: The stuff that I was talking about today, you know, we wanted to go back to two thousand and Sentinel wasn't available at that time. But all of these methods could be adapted. We haven't done it, but they could be adapted to other well, to wall sources of imagery.

56:18 Speaker 3: All right.

56:21 Speaker 5: Number seven here, how do I use Landsat imagery with Jeri data. Which software can I used to do this? Or can I use different coal barriers from landsids such as m dB I e b I. Yes, and additionally kind of a similar other questions are US addresses it's possible to use sending A one or two instead of lanserts for biomass estimation.

56:46 Speaker 3: Yeah, I'll address I just addressed that one. But you know what software would you use?

56:52 Speaker 2: I mean, you could use a lot of different kinds of software, but it would be easy on Google Earth Engine. I think that I mentioned that Google is storing the Jedi data set as an asset. It's very easy to use and it's US also stores lands At and Sentinel and other sources, so fusion with other data sources.

57:14 Speaker 3: One easy answer would be Google Earth Engine.

57:19 Speaker 5: Sober okay. Number eight. Would the bootstrapping time series you mentioned be similar to performing multiple simulations like Carlo simulation?

57:32 Speaker 2: The answer is yes. There's a stochastic element in both of those. I would say that the bootstrapping that we are doing, we're bootstrapping models either Field to Jedi or Jedi to lance At or both. And that's a narrower term and accounts for just the uncertainty involved with those models. Monte Carlo simulation would allow you to incorporate variability from other sources also, so I see Monte Carlo simulations as sort of a broader class of stochastic simulations.

58:16 Speaker 5: Okay, let's do the next couple of questions and and then, just for the benefit of the audience, remember that eventually all the answers will come up in the in the Q and a document uh several days in in a couple of days a week max. Okay. Question nine is estimating biomasses in Jedi applicable and efficient in drylands or grasslands.

58:48 Speaker 2: I I have not personally worked a lot with grasslands. I believe that Jedi's waveforms are.

58:59 Speaker 3: I'm not sure they're great for grasslands, to be honest with you.

59:04 Speaker 2: And and but in dry land forest, I would think that the measurements are are you know, perfectly reasonable to use. We have not tested obi wan so obi Wan brings in the element of modeling Jedi to landsat I'm not sure how well we do in dry land forest. We have not yet tested or validated that in dry land forces specifically. But as I say here, I think we do need to do that. So it's kind of an incomplete answer, but that's that's what I know so far.

59:45 Speaker 5: Yeah, I'd be interested to see how it behaves. All right, Question number ten, I think you mentioned that the assumptions are going to little biomass estimations breakdown for areas less than five hund there? Did I hear that correctly? If so, can you speak a little more about that.

1:00:06 Speaker 2: Yes, So, we we're working with model based inference, and there is a component of model based inference that involves spatial auto correlation of the errors in our models. And uh, that is relevant at at very fine scales that we and we do not account for that uncertainty, so you know, at finer scale, at very fine scales, Uh, you know, we simply you know, given the assumptions of our of our method, we don't trust those estimates at very fine scales. So that's why we recommend at least five hundred hectares M makes sense?

1:00:53 Speaker 5: Okay, Yeah, we're gonna We're gonna stop here. And but again, as I mentioned with the and a document will eventually be available for our participants, uh in the in the training web page. Thanks again, Sean and your team for for helping with this and for all that valuable information. And yes we'll see everyone in Session two next week and next Thursday. To have a great day all, and we'll see you on Thursday.