NASA ARSET_ Estimating Biomass Change with GEDI and the OBIWAN API

0:01 Speaker 1: Welcome to this our SET training series, Estimating Biomass and Change with Jedi and the Obi one Api. Today is the second and last session of this webinar. The focus of today's session is estimating biomass change with Jedi and the Obi wan Api. My name is Erica Potist and I'm a scientist at NASA's Jet Propulsion Laboratory in Pasadena, California, and I'm also an instructor with the RSET program. I will be hosting this training today. There are two parts to this training. The first session was last week last Thursday on estimating biomass using Jedi, and today we will cover estimating biomass change, so there'll be a theoretical component and then there'll be a demo using the Obi Wapi.

0:56 Speaker 1: There is a homework associated with this training. It's It can be accessed through the training web page and the due date is June eighteenth. A certificate of completion will be provided to those participants who attend both live sessions and complete the homework assignment by the due date. So let's get started with today's session. We have one guest instructor. He is doctor Sean Healey who also presented in Session one. Doctor Healey is a research ecologist with the United States Forest Service the end Department of Agriculture.

1:36 Speaker 1: The learning objectives for today's session. By the end of the session, participants will be able to identify the key concepts in carbon monitoring, including system requirements, decision making needs, and the concept of additionality. Participants will be able to recognize howbi I estimates biomass change it's including its use of jet lance at time series and underlying data infrastructure. Participants will be able to evaluate the uncertainty and validation of the data, using for service inventory data to assess the accuracy and position of obi WAN change estimates, access the open source IAPI to generate estimates of biomass change and areas of interest, and finally, apply obi WAN tools and APIs to visualize biomass change and compare carbon gains against different climate scenarios.

2:35 Speaker 1: How to ask questions. Please write your questions in the Q and A window, which you can find on the bottom right under the three points. There will be a sub menu that says Q and A. We will address your questions and at the end of this session, and feel free to enter your questions during the presentation we will try to answer all of the questions during the Q and A session. However, any remaining questions will be answered in a Q and A document that we will be assembling, and we will be posting that document on the training website in approximately a week.

3:12 Speaker 1: Now, I would like to welcome our guest instructor, doctor Sean Healey, who also presented session one.

3:24 Speaker 2: Hello, and welcome to the second part of this training about biomass estimation with Obi Wan and JEDI. I'm Sean Healey. I'm speaking for doctor Jichang Yong. We both work for the USDA for Services Forest Inventory and Analysis Program FIA. That's the nation's national forest inventory with plots all over the country. So in the last session we talked about the JEDI mission, it's data products, the way that it estimates biomass for one point in time for one kilometer.

4:05 Speaker 3: Grid cells, and also how that logic.

4:10 Speaker 2: And the sampling theory can be used to estimate changes in biomass for any area that you.

4:18 Speaker 3: Care about that's over a few hundred hectares.

4:22 Speaker 2: Between different years, using lance at imagery and bootstrapping.

4:29 Speaker 3: So we will review some of that.

4:31 Speaker 2: We will also be talking about practical carbon assessment needs. By the end, I will be showing you obi Wan's API. That is how you and everyone else with an Internet connection can access the tools that are built into obi wan. So you know what kind of needs does obi wan address. It's really climate change mitigation in the forest sector. The proposition for pretty much any kind of mechanism to recognize climate change mitigation across agriculture or the energy sector and forests is to identify management activity or practice that reduces the impacts of climate change sustainably.

5:23 Speaker 2: This slide sort of shows the shape of that proposition in the forest sector, so I'll start at the bottom. A fundamental question is how much carbon in the way in the form of biomass is your forest storing now? And you know there are millions of biomass maps out there and they all can give you an answer.

5:44 Speaker 3: We talked about last time.

5:45 Speaker 2: If those maps were not made in a way that allows you to combine.

5:50 Speaker 3: The uncertainties of all those pixels that.

5:52 Speaker 2: Make up your area of interest, they're not as much useful in the context of practical carbon account So you know, an uncertainty framework for these estimates is critical. Last time we also talked about how through bootstrapping. We can answer the middle question how much carbon is your forest adding, so basically between last year and this year, or between twenty twenty and now, what is the change in biomass and the place that you that you manage. And then sort of the top level question, which echoes the main question and all of these frameworks is how does your management contribute to carbon storage trend?

6:31 Speaker 2: So that really implies a question about additionality. So you might be storing extra carbon, but maybe it was going to store extra carbon anyway, So we have to address additionality, which we will in a second. I just want to point out this is sort of by way of context for everything I'm going to be talking about regarding obi WAN, the sort of most common way that people practically do this, at least with romo sensing. Obviously, in a country like the United States that has an inventory that's been working for many years now, we use the inventory to answer these questions.

7:09 Speaker 2: But in many places that don't have that, the carbon standards use what people refer to as the spreadsheet method. Basically, there is a running total of which is called activity data, of how much forest is being added or subtracted through management and there's an emission factor, how much what's the mean biomass stored in the average hectare of forest? Most existing carbon standards are built around that. If you use it, you're aware that there are some limitations in terms of statistical difficulty of saying declaring something significantly.

7:48 Speaker 2: There are capability capacity issues in some places. You really have to be able to measure things, measure the area of forest in a consistent way through time. Maybe obi wan is the alternative to that paradigm. Okay, so back to obi Wan this top level question, how does your management contribute to carbridge storage trends? That gets to the question of additionality. So what would happen if you weren't doing the thing that you're doing. My example is in nineteen ninety four, the National Forces in the Pacific Northwest implemented a new forest plan.

8:25 Speaker 2: According to obi Wan, before that forest plan was implemented, the region was losing a small amount of carbon. It's very productive region in the nineteen eighties, and prior to the nineteen ninety four implementation of the plan there was a lot of harvest. After nineteen ninety four there was a surprisingly immediate bounce.

8:45 Speaker 3: Changes were all positive. So one way to look at.

8:49 Speaker 2: Additionality is over time, what hapened, what's happening now compared to what happened before you did whatever it is you are committing to do. The second way to address additionality is to compare the changes that you're seeing now in the place that you care about. Let's say it's a national park with areas outside of the national park, So that's a counterfactual sort of baseline. There are complications that should be considered. Generally, younger forests store carbon faster than older forests, So it's really like there are a lot of requirements I think for doing the counterfactual additionality assessment.

9:30 Speaker 2: Well, and I'm not going to talk as much about that, although you will see at the end that obi WAN does allow you to do this kind of additionality assessment. Pressing into the guts of obi WAN, please recall from the first session that its inferences are really based upon bootstrapped time series of biomass maps. So the example I showed before, Let's say that you care about the change between nineteen ninety and twenty twenty in a particular place.

10:05 Speaker 3: There are several.

10:06 Speaker 2: Bootstrapped time series and all of These bootstraps recall our realization of the uncertainties in both levels of model, both the Jedi level of model and the Lancet level of model. The first bootstrap suggests that the area we care about lost fifteen tons per heck there the second bootstrap suggests we lost thirteen. The third suggests we lost twenty. So across a lot of different bootstraps, that's where our inference comes from. So that's how Jedi works, and we don't just have to look at changes.

10:41 Speaker 2: What I just talked about is a second bullet, the difference between for the same place between two points in time. But we can also look for a single point in time. Let's say we just care about the biomass in twenty twenty. Well, the first the first bootstrap says eighty the second says eighty four.

10:58 Speaker 3: Et cetera.

10:59 Speaker 2: So we could get a distribution around a point a single point in time estimate. We've just talked about the differences between two points in time. We can look at two forms of additionality that I just mentioned. The first is change in the rate of gain between two periods. So we could ask nineteen ninety to two thousand, what's the rate of change within the same bootstrap and compare that to what that bootstrap says is the rate of change between twenty ten and twenty twenty and get a difference between those rates.

11:33 Speaker 2: And that's what we look at across bootstraps. So we get a distribution of that change of change, if you will, the sort of additionality over time, and we get and the distribution gives us an uncertainty. We can also compare two places in this way using the different bootstraped time series. So obi Wan really does allow us to look at additionality questions and really gives us a fairly comprehensive way to look at monitoring carbon change within areas that.

12:07 Speaker 3: You care about.

12:08 Speaker 2: So this issue calibration is really important to obi Wan Optical imagery like Lansat and Sentinel to do not see through the canopy. So basically, as forests grow for the first part of their lives, you know, the pixels get greener as more forest grows, and you know there is pretty good signal about how much biomass there is, but once the canopy closes, it doesn't get significantly greener as more and more biomass or volume or basal area is added. As the forest grows, so that you know, people call this a saturation a signal saturation problem, and it manifests itself.

12:51 Speaker 2: When we're making we're using optical imagery to make predictions, we generally underpredict the very high biomass and we over predict the low biomass. And when we think about changes from a high condition to a low condition, we are almost always under predicting those changes because the high is under you know, everything gets squished towards the middle and you go from high to low, well that that change is also going to be low.

13:19 Speaker 3: So calibration is.

13:21 Speaker 2: A concept where if you have independent data in the places where you have predictions, you can across a large a sample of those come up with a calibration function that that simply is a correction, and it is a correction that is applied to minimize bias. It doesn't make your predictions better, but it removes the biases, so you're no longer systematically under predicting the high biomass and over predicting the low biomass. But again it's it's reducing bias, it's not reducing error. But that's that's really critical for this kind of application.

14:00 Speaker 2: This last line here is the only time I'm going to mention this calibration can also be used to convert obi Wan's estimates of biomass and biomass change to other variables. For instance, biomass is quite related to forest volume and also a basal area. So if from the National Forest Inventory, for instance, you had a bunch of independent plots, you could come up with a calibration from biomass to those other variables that you may care about in other settings. So calibration is really important, and I'm just going to talk a little bit about what calibration data are available where So Jedi data is available globally, as we have talked about at least between the latitudes of fifty one degrees north and south.

14:51 Speaker 2: So the first option, once we've made our biomass map with Lansat imagery, we can use Jedi data. The biomass predicted for Jedi points that we're not used in calibrating the lands At map, we can use those to calibrate our predictions and come up with that correction. And in general that correction when we apply it makes the high predictions higher and the low predictions lower. And how much data is available for this well, obi WAN uses about half Currently in twenty twenty six, it uses about half of the available data, which is several thousand.

15:27 Speaker 2: It's at least two thousand Jedi shots per ten kilometer square, So there's a lot of data available for this everywhere the section. The second option is not available globally. It's in places where you have repeated inventory plots that have coordinates that you trust, and we work for the National Forest Inventory, and you know, the.

15:53 Speaker 3: United States is where we are applying this.

15:55 Speaker 2: So basically we have several several thousand and plots in every state, sometimes up to ten thousand or even a little more in some big states where we have measured biomass in multiple points in time, and we can compare those measured changes to predicted changes from obi WAN. So instead of calibrating at the level of an individual point in time prediction, we're calibrating the estimates of change. And in general, when we apply those calibration functions, those really simple calibration functions, we end up making the losses bigger and the gains bigger.

16:35 Speaker 2: Okay, so now I am going to move into a couple of validation activities that we have performed one in Nepal and one in the United States through the great support of NASA's SEVERE program. We worked with Nepal's Forest Research and Training Center, which which is the entity that runs the National Force Inventory for Nepal, and they very kindly provided inventory data for three provinces in that country. And basically what we're looking at here is on the left is the obi WAN estimate of biomass change for a five year period.

17:16 Speaker 2: That's an orange compared to basically the spreadsheet method applied to Nepal's inventory data. This is not exactly aligned with what Nepal reported to unf Triple C, but it is very close. It's as close as we could make it with their with their data. So as you can see across the country, you know, there's pretty good correspondence with the amount of biomass that obi WAN predicts for this period. The obi WAN estimate for the whole country and for the three provinces that we had data for tends to be a little bit more dynamic, and I believe that is because obi WAN represents changing biomass in existing forests that are not being cut down.

18:06 Speaker 2: So many spreadsheet methods, which is the alternative treat the amount of biomass in protected forest as static and will result in sort of less dynamic estimates of change. Obi Wan accounts for that change, or at least predicts a change, a usually growth in forests that don't get cut down. So I think that that's probably why obi wan is showing a little bit more gain over this period. But it is pretty consistent both at the national and the province levels. So what can you do with that kind of information?

18:45 Speaker 2: This is Nepal. There are seventy seven districts in Nepal. One thing that we could think about doing with it is if Nepal were awarded red funding, which it was, and they were engaged in deciding how that red funding was going to be split among its districts, One could use obi wan to sort of systematically look at, well, which district is contributing which amount of the national level gain that the country is being compensated for. In no way would we presume to tell anybody how they should, you know, distribute their money, but this is sort of a systematic objective way to look at a distribution of gains across the country by different jurisdictions.

19:34 Speaker 2: So it could be just used, as you know, decision support tool in that context. Okay, so now I'm jumping into a validation activity that we've done here with FIA data. Again, FIA stands for forced inventory and Analysis. That's the part of the Forest service that Yong and I work for. So we're looking at results for two different states, and on the x saxis it's FIA observed above ground biomass density change. So every dot is a group of plots. The plots pretty much represent about a heck there. They subsample that heck there, but they're all about a heck there.

20:17 Speaker 2: So what we're basically doing is grouping together bunches of five hundred FIA plots and ordinating them on you know, the X axis here, like how much they're gaining or losing between two periods, and on the Y axis, comparing what obi Wan's the answer for that group of five hundred plots is. So, for instance, the blue line is what obi wan with no calibration, that's what obi wan gives us.

20:47 Speaker 3: And I mentioned that signal saturation, and.

20:50 Speaker 2: Really what we see is that obi wan systematically underestimates the losses and it underestimates the gains. So you know, that's a main reason why we use calibration. When we use the kind of calibration that was available in Nepaul and everywhere outside the US, using independent Jedi shots for at least in Alabama and in Maine, we see that really pushes the estimates of loss well below the dotted line, which is the one to one line. So we're really, at least in Alabama, for these different groups of plots pretty severely overestimating losses when we use that global calibration.

21:33 Speaker 2: However, once we get to the gain territory in Alabama anyway, that calibration really seems to work pretty well. It brings it up close to the one to one line. Calibrating with FIA repeat measurement data, actually it does pretty well. I mean we're seeing across all of these groups of FIA plots that the performance is not bad. This interval is the is basically the precision that the standard error that obi wan gives us for these changes.

22:08 Speaker 3: There is bias that is not accounted for in this interval.

22:12 Speaker 2: So when you get something out of when you get an interval out of obi wan, it's the precision, it's not the bias. So this kind of validation is crucial. In main we see something pretty similar except above a gain of twenty so places that are gaining a lot of biomass per year, we're not differentiating those cases very well. I would point out that at least with the FIA calibration, actually the global calibration as well. You know, the one to one line is within the confidence interval that you would get out of obi WAN.

22:46 Speaker 2: Okay, So moving on from estimation of change to estimation of additionality over time. That is, we're asking how much faster or slower is this place gaining carbon biomass than it was in a previous period. So you know, the same same two states we're asking. So we have multiple measurements, several measurements in many cases for fi A, so we we know we're able to measure this sort of acceleration or deceleration of gain at individual plots, and so just looking at the corresponding estimates of gain from obi Wan on both so on the on the right side, those are those are populations that are gaining biomass faster than they were before.

23:38 Speaker 2: And on the left, those are populations that we're that are losing biomass faster than they were before, and the fi A calibration, the red series does really pretty well across the whole the whole range, whereas uh well, actually the global calibration does better with this additionality challenge than than it did with the individual two points in time question. So that is sort of the validation that we have pursued in Nepal, in and the United States. I'm going to jump into API Application programming interface.

24:15 Speaker 2: I should have defined that, but basically it is a set of functions that we provide that you can use to access the time the bootstrapped time series that we're storing on Google Earth Engine. Again, Google has has donated substantial amount of storage for.

24:38 Speaker 3: Us to store that.

24:39 Speaker 2: And what the obi wan API does is it just gives you a bunch of functions that you might use to do very common things involved with reporting and verifying carbon changes in the place that you care about.

24:54 Speaker 3: So the things that.

24:54 Speaker 2: You might specify with the with the obi wan ap API is what years that you can about, how you want to visualize those changes, what kind of authentication you want to build into accessing whatever it is you're putting online. You can define it a baseline additionality baseline, whether it be a different point in time or a different area, maybe a surrounding area or a buffer area. And also it allows you to you know, store whatever reports and whatever format you want. So really the API is maximizing your flexibility.

25:28 Speaker 2: So I guess the question is why are we doing this instead of making a map? You know, the map is easy to distribute. But and this is definitely a sort of new paradigm for accessing space based biomass data. But I think there are good reasons for what we're doing. The first is we can't predict the years and the places that you care about, so allowing you to define those things it is obviously targets the stuff.

26:01 Speaker 3: To exactly what you need.

26:04 Speaker 2: Also, across market and treaty based commitment mechanisms, some users need to show things very transparently, perhaps to donors or to other countries, whereas others I'm thinking of private industry in this country, for instance, that maybe participating in carbon offset markets. Much of this is considered proprietary and they need to control access. So, you know, using the API to develop your own application really gives you the power to build in you know, whatever you need. It's easy to standardize across affiliated clients, maybe all participate participants in a particular carbon market, or perhaps different jurisdictions supported by the same donors.

26:51 Speaker 2: So one could picture, you know, developer hired by a particular donor developing dashboards for lots of different clients in a way that is consistent. Also, we're pre computing these bootstrapped time series because if if when you made your query and you had to wait for all the computing involved with bootstrapping to happen, it would you know, it would take hours. It may take hours. So we've pre computed these things, stored them on Google and that makes it much faster. And lastly, like we do it this way because you know, you may not be the only person who cares about you know, your country, so we're reducing a lot of redundant computing.

27:38 Speaker 3: So this is really sort of an overview of the structure.

27:42 Speaker 2: We do the bootstrapping, we store it thanks to Google on Google Cloud. We're providing the API, which we will get a look at in a moment in our hands on activity. And then you and you're the developers that you pay probably develop your own dashboards, accounting systems, reporting systems, building in whatever considerations are important to you.

28:09 Speaker 3: So the key is this API. And it sounds harder than it is.

28:14 Speaker 2: API is really pretty well documented, is really very simple. It's just a bunch of commands with you know, syntax, the divide that defines important variables for these commands, that queries the stored asset on Google Cloud and gives you what you need, and you will see those You'll see some of the calls to the API in the notebook we are about to look at. These resources are some background about the theory of our estimation methods, about some context for carbon markets. Okay, so before we dive into the hands on activity, I just want to tell you a little bit about what we're doing.

28:59 Speaker 3: We're going to be used a collab notebook.

29:02 Speaker 2: That's a Google collab notebook that calls an asset that we made for this training over Alabama State.

29:08 Speaker 3: In the United States.

29:10 Speaker 2: It's ten bootstraps deep, which is not deep and operationally we would probably want one hundred. But it will give you the idea of what's happening here. It will allow you to experiment with different options and different visualizations, and it gives you actual syntax like how the API is you So you could give this to your developer friend, and your developer friend could really practically build a dashboard pretty easily for you for the area that you care about. So what you're seeing is the link to the collab notebook.

29:49 Speaker 2: Before you start playing around. I will give you some further directions in just a moment.

29:55 Speaker 3: But if you could click.

29:56 Speaker 2: On this link, Google may ask you, if you've never or used a collab notebook, if you want it to compile the notebook for you. Please say yes and then stand by for more instructions. Okay, so if you clicked on that link, what you probably see is this, I want to point to your direct Your attention to this cannot save changes. What I'm going to ask you to do is go to file and save a copy to drive.

30:35 Speaker 3: That way what you're working on.

30:37 Speaker 2: First of all, you will be able to save it, and a lot of other people on the same training won't be, you know, disturbing what you're doing. That will be unsatisfactory for everyone. So once you save a copy to drive, what you will see is copy of this. This is a Python notebook using Google's collab and and this is the extension for that kind of file. So so really what you're going to be seeing is a lot of Python code that's calling the a p I

31:15 Speaker 2: and visualizing a lot of the estimates that the API enables. So, uh, you know, the first thing I guess I would point to is there is some you know.

31:29 Speaker 3: Annotation here.

31:31 Speaker 2: But but basically, if all you want to do is see what it does, you can just go down through and press these these triangles and basically that runs the code that is hidden you. But you can see the code here this first this first box is just setting up some variables that are calling.

31:54 Speaker 3: The obi wan end point the API.

31:58 Speaker 2: Okay, so a lot of this is really sort of generic, but it will help. I mean, I really feel like with this code you're probably fifty percent of the way towards making you know your own dashboard for the area that you care about.

32:12 Speaker 3: So all of this is sort of generic.

32:14 Speaker 2: And right now this is pointing to this Alabama study area that we've made for this training. But soon enough we will have at least an asset, we will have bootstraps for the entire United States. And when you do that, when we do that, you will only have to change this, you know, this address. So once we have a national asset and an international global asset for all of the bootstraps available.

32:53 Speaker 3: This code will really this is one line of code to change the actual address of that asset.

33:01 Speaker 2: All of the rest of this code, this Python code, will be applicable. So and right now you know this the Python code is calling this visual this viewer, and you know we can turn this the asset off and on.

33:20 Speaker 3: You can do whatever you want.

33:22 Speaker 2: I mean, you can visualize it with whatever imagery or perhaps company records behind it. One thing I would point out right now, what we're displaying is the mean of the biomass predictions across ten bootstraps. I mentioned that ten bootstraps is not a big number, and here's a ten kilometer tile where the the bootstraps obviously simulated errors across all ten that happened to give us very high biomass values.

33:57 Speaker 3: The color ramp here is is biomass.

34:00 Speaker 2: So there's nothing squareish under underneath there. You can turn that on and off up in the upper right hand corner. But basically, you know, what we're seeing is one way to visualize the asset. Okay, so again this could be any visualization you want, but in this case, what we're showing here is you pick the area that you care about. Again, this is in the middle of Alabama. I sort of arbitrarily drew a polygon in your own in your own development, you would probably import a KML.

34:49 Speaker 3: Or a shape file for the place that you that you care about.

34:53 Speaker 2: Okay, so we selected an area of interest, and now I'm going to let's say we care about twenty nineteen. So we for that area of interest, which is thirty seven thousand hectares, the estimate is one hundred and thirty six tons per hectare. That's the units here for biomass. That's without calibration. Let's try with calibration, see if that changes it from one hundred and thirty six.

35:32 Speaker 3: It did it.

35:32 Speaker 2: You know, it's a fairly high value, and it made it higher in general. So what we see the standard deviation for that estimate. Again, that's the precision, that's not necessarily the overall accuracy. Okay, So you can you can toggle this on and off to see the effects of calibration.

35:55 Speaker 3: Okay.

35:55 Speaker 2: So here's here's another thing that you might want to do for the area that you care about. Let's look at.

36:04 Speaker 3: The trajectory.

36:08 Speaker 2: Of this area and basically what you will see is I mean, I just drew that fairly arbitrarily, but we saw a gain up until around twenty thirteen, and then a loss. It is a fairly actively It looked like a fairly actively managed area of again thirty seven thousand hectares, okay. So, and I find that looking at the estimates for different points in time really sort of gives a lot of insight into the overall trends for the area that you care about. So, you know, estimating a change with a standard standard error, let's just press this button here for I think it's from the beginning of the time series twenty nine or nineteen ninety nine to twenty twenty three, the estimated change is a gain of ten.

37:12 Speaker 2: If we only care about the end points the you know, the gain was about ten. If we had to find that as nineteen ninety nine to twenty thirteen, it would have been a big bigger gain, and from twenty thirteen to twenty twenty three.

37:25 Speaker 3: It would have been obviously a loss.

37:28 Speaker 2: So that's why looking at it at the mean estimates across time first is pretty is pretty valuable. I mentioned that there's this counterfactual additionality where you where you apply, where.

37:43 Speaker 3: You look at the changes in your area.

37:45 Speaker 2: Let's say it's a park compared to maybe a buffer zone that is not managed under your management plan.

37:52 Speaker 3: Uh, and is not you know, the.

37:55 Speaker 2: The area that you're committing removals from the atmosphere to. So we're going to skip that. It works, but you have to be able to trust your comparisons if you want to make an inference of like this, the differences that you're seeing are actually a good indication of the impact of your management. So we're instead going to skip to looking at changes in the change, changes in biomass change changes in the rate of biomass accumulation.

38:38 Speaker 3: So we press that.

38:41 Speaker 2: And we're looking at the difference between Let's say that this is our baseline nineteen ninety nine to two thousand and five in the place in the target is twenty fifteen to twenty twenty one. So you will see here in the outputs that these are These are indeed the play that the times that we care about.

39:10 Speaker 3: You see the area of this place. This is the change in the annual rate of gain. So we're.

39:20 Speaker 2: Compared to the baseline time period, we are gaining two tons per hectare per year less in the target period, which is twenty fifteen to twenty one. Okay, and when we have across bootstraps, this is the uncertainty that the answer across bootstraps implies.

39:44 Speaker 3: So I guess I would.

39:46 Speaker 2: Say that, you know, this is a demon This Python notebook is a demonstration, but it's also you know, a lot of a lot of Python code anyway, that gives you good examples about how you could make your own application dashboard, or at least it gives a lot of examples of calling the API for things that you care about. So I think that after a brief message, I will be on for answering your questions. I look forward to them. Have fun with the notebook.

40:23 Speaker 1: Thanks, thank you very much doctor Heally for that great presentation and demonstration. Now let's do a quick summary of the most important points discussed during this presentation. Today we learned that obi wan can address climate change mitigation across different sectors by identifying management practice that reduce impacts. It's important to understand that uncertainty for carbon estimates is critical and that concept of additional is how management practices contribute to carbon storage trends.

41:05 Speaker 1: So what would happen if you were not doing what you were doing. Obie uses hierarchical bootstrapping to get a distribution of predicted changes and optical imagery from sensors like lansat do not see through the canopy and are likely to underpredict high biomass while overpredicting low biomass for a single point in time, as well as underpredict changes between high and low biomass conditions. And finally, calibration is a simple correction based on independent data and applied to the predictions to minimize bias.

41:44 Speaker 1: There is one homework associated with this training series and the homework can be accessed through the training website as of today, the due date is June eighteenth, and there will be a certificate of completion to participants who attend both live webinars and complete the homework by the deadline. Participants will receive a certificate via email approximately two months after completion of the course. If you have any questions about the material that has been presented, feel free to reach out to doctor Sean Healey through the contact information in this slide.

42:28 Speaker 1: We've reached the end of today's session and this training series. I'd like to thank doctor Sean Healey for his excellent presentations. I'd like to thank all of the participants for their interest in this topic and all of the great questions that have been coming in. We will now start our Q and A session.

42:51 Speaker 4: Thank you Erica for the SUMMARYE and wrap up. My name is Savannah Cooley and I am and our set trainer with the Ecological Conservation Program and I will be facilitating our Q and A session today. With that, we can go to the Question and Answers document where we have been populating all of the excellent questions that have been coming in. And let's go ahead and dive into question one, which asks JEDI and obi wan seem very useful for estimating biomass change, but high stakes carbon decisions requires strong evidence of additionality, uncertainty, control, and validation.

43:43 Speaker 4: In your view, what is the current practical boundary between using obi wan for scientific monitoring and using it for operational carbon credit or policy verification. So with that, i'll pass to you Sean. You've started your answer here all and I'll let you elaborate.

44:03 Speaker 3: Thanks Savannah.

44:04 Speaker 2: Yeah, this is really sort of a fundamental question, isn't it.

44:09 Speaker 3: You know, when.

44:12 Speaker 2: Authorities issue carbon credits, and when countries use particular methods for accounting for the additionality of their forests their forest management. You know, there's always a decision, Uh, you know, what is reliable enough, what is what meets the standard of of what can be used?

44:34 Speaker 3: What's you know, what's the best available?

44:37 Speaker 2: I think in most many cases, uh, obi wan represents the best available remotely sensed answer. Uh, It's not as good as having a longitudinal field survey, but that's not available in many places. So I mean we see a lot of people, uh, you know, countries, especially projects using this emission factor at tivity data paradigm that you know, that relies on a statistical framework, just like obi WAN does.

45:08 Speaker 3: I believe that obi wan.

45:13 Speaker 2: Accounts for more factors than that it accounts for changing changing emission factors within the same class through time. The short answer is, uh, you know, the whether obi wan or you know, the activity data emission factor paradigm or anything else is acceptable is up to you know, particular creditors. My personal opinion about obi wan as some of the some of the validation activity sort of suggested. I think it's good outside of the United States. The calibration with independent JEDI data is very useful and it removes a lot of the biases that we see, but it it's not it also introduces biases and in some cases we're over predicting losses, which you know, that's that's the primary problem.

46:10 Speaker 3: So I'm not sure that it is appropriate for.

46:17 Speaker 2: You know, UH funding a critical decisions and monitoring outside of the United States yet, although I do think that it probably represents the best remotely sensed answer anybody is going to give you. So I think it's appropriate for science, but maybe not for for the offset markets in the United States, where we have systematic collection and availability of change calibration data. I think our validation activity suggested that. Yeah, I think it is approaching that level where you you might base funding decisions on that, crediting decisions, you know.

47:02 Speaker 2: And again it's it's not as good as having field plots in your forest, you know, going back in time. But in a lot of cases that that kind of field effort UH is just not feasible given the amounts of money that are involved. So that I think that's a fundamental question, which is why I spend a lot of time on it, and I've.

47:24 Speaker 3: Thought about it a lot too.

47:26 Speaker 2: You know that the statistical framework that that we talked about is really you know, everything that we're doing is geared towards making this more acceptable to countries when they when they choose their method for reporting and for carbon standards, when they choose their methods for for monitoring the credits that they issue. So that's that's a long answer to a good question.

47:55 Speaker 4: I think, thank you so much Sean for the thoroughness of your answer and really going into some of the nuances and the key considerations. And I think, yeah, what you emphasized around the importance of the knowing the uncertainties and ultimately that it will be the creditors that be able to decide that existing information, you know, and also the high costs of field based methods of doing these kinds of analyzes that obi WAN still despite having uncertainties, also represents the best information that we know of available and from a remotely sense standpoint.

48:43 Speaker 4: So thank you so much for that thorough answer, and we will move to question two. Can the integration of Jedi derived biomass structure lansat SLASH sentinel landcover history and SIF which is solar induced fluorescence based photosynthetic activity improve estimates of watershed scale carbon capture and biomass change dynamics under climate and disturbance pressures.

49:11 Speaker 2: Well, yeah, that's a good question. I In general, the answer that I wrote addresses, you know, what imagery we use in obi WAN. As I read the question more thoroughly, I'm not sure that's the question that's being asked, but I'll just to elaborate on that question. Yeah, the better the imagery is correlated with biomass, the better our answers are going to be. We use lance at because it's available systematically back backwards through time.

49:44 Speaker 3: In terms of understanding.

49:48 Speaker 2: Climate pressures, climate processes and stuff like that, obi WAN is not a it's not a process model, right, We're not.

49:57 Speaker 3: We are only monitoring, uh, past trends.

50:02 Speaker 2: We're not trying to extract processes that we might use to extrapolate under different climates and things like that.

50:10 Speaker 1: Uh.

50:11 Speaker 2: In terms of disturbance pressures, yes, certainly we're uh. I mean in a retrospective way. We're allowing the quantification of biomass change in a particular area that may be subject to or or protected from developed disturbance processes related to development, and it facilitates comparison of the additionality there. So yeah, I think that's yes, Yes for disturbance processes, no for.

50:43 Speaker 3: For climate processes.

50:46 Speaker 4: Thank you. Question three asks can the obi wan API be used for urban prediction? Uh?

50:56 Speaker 2: The short answer is I wouldn't. Buildings often full Jedi waveforms. They can look like trees, and so I frankly just don't trust our predictions over urban areas, just because that that sort of fundamental error will propagate its way, and we haven't tested it, but I don't think it's probably it's probably not the most appropriate thing to track urban forests in particular.

51:27 Speaker 4: Right, even if even if an application I imagine we're to have a very accurate map of building footprints that they could mask out, there's still the consideration of the Jedi geolocation error, right which would which would.

51:44 Speaker 3: That's right?

51:44 Speaker 2: Yeah, yeah, And just that's that's up to about ten one sigma is ten meters, so you know, sixty seven percent of the footprints might be ten meters away, and you know some might be considerably farther away than the coordinates say they.

52:04 Speaker 3: Are, yeh.

52:07 Speaker 4: Question four asks how do uncertainty approximation approaches differ between m d N based probabilistic modeling and obi WAN based biomass change estimation, and can these approaches be integrated to improve watershed scale carbon and ecosystem decision support.

52:27 Speaker 3: So I don't even know what MDN stands for.

52:31 Speaker 4: I don't either. If actually, if someone who asks this question wants to clarify that part in the chat, pause a moment or circle back to this question.

52:49 Speaker 2: I mean to answer the question about watersheds. I think obi wan is appropriate for watershed carbon decision support.

52:59 Speaker 3: You know that we're where as.

53:00 Speaker 2: I said in the presentation, uh, talking about larger areas over a few hundred hectares, and that certainly is most watershed scales. So I would feel comfortable using obi wan to to evaluate carbon trends in particular watershed, as long as it's not you know, the really fine fine scale of huck levels.

53:27 Speaker 4: Thanks. Question five is a question about the black stripe line in slide eleven.

53:36 Speaker 2: Yeah, those those were the Those were the figures that just have field measured change on the X axis and predicted change on the Y axis. And the closer that that that black line is the one to one line.

53:52 Speaker 3: So the closer.

53:55 Speaker 2: A particular population is estimated to the to the real change, the closer it will be to that black line.

54:02 Speaker 3: So that's just there for reference.

54:07 Speaker 4: Question six asks, I know that over steep slopes, the lighter waveform tends to stretch. This can lead to overestimating canopy height. Do the current versions of relative hype metrics such as rge on one hundred and R ninety eight in the level two A product include any slope correction algorithms or have they successfully mitigated this slope stretching effect.

54:33 Speaker 2: Yes, so, I mean, if you think about the waveforms we looked at last week, if the level of the I mean, it's based upon the elevation of the ground that's found. But if you think about, you know, the circle that is a footprint. If that ground elevation is much higher on one half one side of the circle than the other, that's going to introduce a bunch of noise to that waveform. And that's why the Jedi mission is using a much smaller footprint, so you know, minimizing the potential for that ground level to change considerably.

55:13 Speaker 2: But it is twenty five meters and if you mean you can go out in the woods and walk twenty five meters and go down a slope, so you know there is noise there. And that's why Jedi provides waveforms basically the L two a relative height metrics, using a variety of ground finding algorithms specifically to address well. This issue is one of the main reasons why Jedi provides multiple versions of our age and some of them, I mean, the intention is that some of them are probably more locally representative than other.

56:01 Speaker 2: So I've included a link to all the documentation for those algorithms. So the short answer is we minimize it as much as we can, and we take a range of measures to to to find the ground appropriately, and you can find you can find different choices for that algorithm in the in the Jedi data.

56:33 Speaker 4: Yeah, and maybe I would just add that this is a fundamental issue with just how full waveform light our works, right where that we don't actually have a way to trace individual returns within a footprint in terms of where within that footprint they landed. And so there's no that that's a fundamental characteristic of just how waveform light our works.

57:00 Speaker 3: That's right, Yeah, it's well put.

57:03 Speaker 4: So in question seven, we have the following, could you share more applied cases of satellite data fusion, especially examples where multiple data sets such as Jedi, Lancet, Sentinel, solar inducescens or synthetic aperture radar are integrated to improve biomass, carbon or ecosystem monitoring.

57:26 Speaker 2: Yeah, I put I put a link to Laura duncanson at All's recent paper. I think it's accepted now, but I included the link for the free version. You know, that's that's fusion of ice at two, which is available outside of you know, above the latitudes that the Jedi is limited to, so the boreal mostly the boreal region that is not studied by Jedi.

57:51 Speaker 3: Ice at two does does see it?

57:53 Speaker 2: So that that paper and integrates fusion of Lancet and ice at two. There's also a lot of work with Jedi specifically that has followed fusion with.

58:11 Speaker 3: Terras are like tandem x uh so radar data.

58:17 Speaker 2: So you know it, and you know the Jedi asset is available on Earth Engine, so you know the the possibilities are almost limitless in terms of fusion through a number of different methods.

58:33 Speaker 4: Great, thank you, and and with the time here, this will be our last question for the live Q and A. There are more questions that have yet to be addressed, and we will make sure to have those addressed and published in the Q and a document that will be posted to our training website within about a week from from today. So let's end here with question eight. How is calibration droplated to every lansat pixel when using Jedi slash F I A.

59:06 Speaker 3: Yeah, and I'll start with my last sentence there.

59:08 Speaker 2: Obi wan is a tool for estimating biomass and biomass change over large areas. So the focus is not perfecting every prediction, it is removing the biases across the population. Because because we're focused on a population level parameter, the mean change or the mean biomass, So you know, in general, we use data from across the across the landscape to come up with a simple correction classical linear calibration function that we apply to all of the pixels. It doesn't necessarily make any particular pixel better, but it removes the biases across across the whole population.

59:48 Speaker 2: That's the intention anyway. So yeah, I guess the question is how is it extrapolated. The calibration is fitted us the large population and it's applied to all the pixels knowing that you know our target is the population parameter and not any particular pixels trajectory.

1:00:13 Speaker 4: Great, Well, that that's our time here for today. I want to thank you doctor Sean Healey for such a fantastic presentation and demo of obi Wan and think all participants for your interest and participation in today's training.

1:00:36 Speaker 2: Thank you Savannah, you guys have been great. This is really a well supported training I think so.

1:00:43 Speaker 3: Thank you, Thank you all,