 So, hi everybody, we're going to start very briefly. So, first of all, welcome to the R&D session. My name is Daniela Requena Suarez, and today I will moderate this event. But before we completely start with the event, I just want to let you all know, especially people joining us online as well, that all of this session is being recorded to facilitate note-taking after their meeting. And of course, all of this data will be destroyed once this purpose is served. So, with that being said, I think that we can start, and I can explain a little bit of what we do as a GF-O-I R&D component, just following from the presentation that Martin did earlier today. So, let me just switch slides and... Yes, so the purpose of this session is to go through the GF-O-I R&D priorities and to discuss about them. So, just to fall back on what the R&D priorities actually are, they consist of knowledge gaps and research needs that are identified by the R&D community, which basically means us. And also through interactions with country participants and network of experts throughout the world. And of course, these priorities are not static. They need to be evaluated over time and they have to be updated. Perhaps there are some priorities that are coming back or some new priorities. And then there's also other topics that we might consider as an R&D community that it has gone through the R&D phase and it's now more operational. So, the last discussions that we held on these priorities were actually on January of this year in the capacity... Just before the capacity building summit and that was quite useful because we got the input not only of research experts but also from actual country needs. And what do we do? No, which are these priorities? Let me just... Yeah, no, so let me first say what we do once we have identified these priorities. So what the GFI R&D component does is that we advocate to create more activities related to these research needs, such as, for example, expert meetings to address these topics. Also maybe encourage the production of literature reviews. For example, for a priority where there's a lot of R&D work that is ongoing but that still hasn't been consolidated into a review itself. Whoa. There's also through the MGD component and if the MGD component requests it or basically if it finds a need for it, then we can also assist in the production of new guidance materials based on existing studies and also on existing case studies that have worked in the past. They are also each priority and each method that is underscribed into each priority are assessed through calm assessments which is basically saying how operational is it. Is it really like an exclusively R&D topic that still cannot be applied in a national forest monitoring system, for example, or is it something that is already on the brink of being rolled out? Also for capacity building activities, we also encourage the direct support of researchers to country needs through development and use and also advocating the key research needs from the research community, for example, by advocating for the generation of more funding or perhaps for establishing side events and conferences and so on. And finally, but last but not least, also stimulating engagement with the other R&D components like I've said, the capacity building component is one that we work with very closely and of course the data component is one that we foresee that we will work side by side as well and also the methods and guidance component. With that being said, here are the current R&D priorities and these are the presentations that we will see today. So first we have degradation and regrowth mapping. Then we have biomass and emission factors followed by early warning systems which are also called deforestation alerts. There's also the cross cutting topic of uncertainty assessments and finally land use and greenhouse gas estimation. So all of these topics have a list of sub topics and they are all quite, there's a lot to be discussed under each of these topics and of course there are other topics that we also consider in a separate category but they're just not priorities themselves as of now. So today we are actually going to hear from four of these five topics and so we will see R&D progress, also country progress on these topics and for this we have the pleasure of having a really diverse panel in terms of experiences, levels also where they're coming from, their background and so on. So for degradation and regrowth mapping we will be hearing from Gilberto Camara from INPE then from biomass emission factors we will be hear from Natalia Malaga from Wageningen University. For early warning systems we will be hearing from, actually there will be a dedicated site event tomorrow. Therefore we encourage for anybody that is interested in this priority topic to join this site event tomorrow. For uncertainty assessments we will hear from Chad Babcock from the University of Minnesota and finally for land use and greenhouse gases we will be hearing from Viola Heinrich from the University of Exeter and also from J.F. St. Potsdam. So each participant, no, each panelist has been given the very difficult task of trying to summarize the research, the state of research in these very broad topics in five minutes or less. So I know that this is a challenge and I really hope, like I wish you good luck. But after that what we want to do is we also want to open the floor for questions from the audience as well as from people online via the Q&A feature in Zoom. And we want to have a dedicated discussion for each priority for about 10 minutes or so depending on how much the presentation of each panelist has lasted. And then after that we will have a closing discussion in which we can basically discuss the role of the J.F.O.I. R&D community broadly, more broadly. So for that there are some questions, of course everybody already I assume has some questions but there are some questions that we would like to keep in mind. For example, what are your thoughts on the current list that I just showed and the topics that you will hear from today? And how can this perhaps be refined? Of course I haven't shown a lot of detail but I hope that maybe you have some comments based on the presentations today. Is there a theme or topic that you see that is very much missing from the priorities right now? We would also like to know about this. Also steps of how to move forward as a J.F.O.I. R&D community. Has been not so active recently but we're getting started again and there's a lot of ways in which we could be active but of course we want to target ways in which people would actually join and interact. And finally for example, if there is an expert workshop meeting that we are organizing and we definitely will aim to organize expert workshop meetings in the coming years, what should it focus on? One of these priorities and another topic. So these are questions to keep in mind. So with that being said, I would like to invite Gilberto Camara to take a shot for the first presentation. So Gilberto, yes. I can share the slides here. So if you want, you can come here and control the slides. So thank you. I have it here, so let's see. So the floor is yours. Well, thank you, Daniela, and thanks everybody. And also I mentioned thanks to you and Martin for organizing a gender balanced panel. It's very important these days. So I would like to go into this issue of the challenges of measuring the degradation. So let me start by saying, what was the next one? Beg down. Whoops, how do I get down here? Clear? Okay, so now you see that degradation has been a recent major topic of interest and needed to say that we have these recent paper in science talking about the drivers and impacts of forest degradation, where the whole issue is discussed at length and there are other recent topics. The problem here is while it's important to recognize that the degradation is a major factor it should be taken into account. The problem is whether we know how fast to define it and if we cannot provide an accurate definition of it, how can we measure? So we first need to define and then to measure. We know, like intuitively, that degradation happens as a process which may lead to deforestation or clear cutting the end, which may end up by itself, but still requires a huge amount of effort because it's tough to do. Now, we have had some successes in an operational system called DETER in Brazil, some of you know, which an early warning system which both flags are lots of new deforestation and are lots of degradation and forest fires. And we have had some success by, for example, using shade and members to identify degradation areas. This is one example, but this is done by an expert with visual interpretation. Like I said, it's much easier to publish on remote sensing of environment than to deliver a system that meets the 93% quality that Mr. Claudio has shown to you. So when we have a 93, 95% bar, it's different from publishing 10 papers on remote sensing of environment. So the problem of defining deforestation, from a conceptual point of view, you actually know that you can think of this as a time series. This is a paper many of you may know by Robin Shadson called When is a Forest a Forest? And she says, okay, I have a transition or the natural dynamics thinking of forest. I have transitions that lead up completely deforestation. I have disturbance. And when does the disturbance turn back into degradation? Okay, the first point here I will argue for, it's not possible to define degradation without resourcing to time series. So we need both the spatial and the temporal information to define degradation. Now this is an interesting example where time series is used as a paper by Willa Colosson and Carlos Souza, it's not 2016, it's actually 2020. Now the problem here is the opposite. This is very common on papers. People publish a paper, but then they forget quote unquote to share the code and share the data. So when you try actually to move it to the operational scale, then believe me, we tried, it doesn't work. It doesn't work in the sense that it does not meet the 90% quality threshold, which means that deforestation is not only a problem that you have to define, you have to measure and you have to reproduce the measures. So if you publish a paper, if you do R and D, please share, share what you do. Otherwise I cannot even evaluate if what you're doing is useful. So in that sense, we started 10 years ago and now it's becoming traction, it's becoming operational. The system for satellite image time series analysis of Earth observation data cubes. It's called SITS, it runs on R and runs in anything that has a stack interface like Amazon web services, Microsoft, it can run on CPO. And by the way, if you like it, don't thank me. Thank the guy who's sitting there called Noah Gorellick because he's the architect of Google Earth Engine. So when I wanted to devise it, I just copied the interface from Google Earth Engine in a different way. So it looks nice because Noah did it first. And of course, we are experimenting with large scale measures, a lot of time series. The answer, the simple answer to it is, yes, it's worked better than individual images. Yes, you can detect very much detail on deforestation. And yes, we don't have a measure in equivocal using space and time to measure degradation. That's where we are and thank you very much. I hope I kept to do my five minutes. Thank you, that was perfectly on time. So I would like to open the floor for any questions or comments to Gilberto. I see many hands, I think that you can go. Yeah, okay, thank you. Nice presentation, Gilberto, I think the photo is degradation, but I think I need, I will need more, but the time is not enough. Yeah, in your slide, I mean, second or third slide, you like, what is, pictures of the boundary of the forest degradation. But sometimes it's like in our country, when we have the root in the forest, in the undisturbed forest, I think not only in the root, that is forest degradation, but maybe it's the side of left and right, but like the assumption, like maybe 500 meters or one kilo in the side and right on left is also degraded, it's also degraded. But how about your case in Brazil, thank you. Well, first of all, the definition of degradation may not be the same of Indonesia than it is in Brazil. It's very likely not to be. The question here to you or to anyone is, are you able to provide a measurable definition of degradation so that you can interpret your data and of course, show it to everyone else, because it's like COVID. There is, during the COVID pandemic, we had the measure definition of vaccine quality. It has to reach certain standards established by WMO before it is used in human beings. And so it was and saved millions of lives. So the question here is the same. If you think it's degradation, define it so you can use it for your own purposes and then it's consistent over at least Indonesia. It may not be consider over the tropical belt, but at least it's consistent Indonesia. In Brazil, to be honest, we have not reached such a point because we have not reached a point where everybody agrees what degradation is. But the other thing is, don't expect this to be right in the first time. You expect it to do a lot of trial and error before you converge to a definition that you can reproduce using remote sense. It's not like a one-way street. It's a many-way street and you go back and forth until you reach somewhere where you're safe. Thank you. Any other questions? I think I see a hand, yeah. There, Fred. Thank you, Roberto. I'm Fred from WRI. It's a bit of a hypothetical question, but if you would have, for 20 years, one meter resolution diet data, wouldn't you have degradation? You only have a tree and a tree loss? No. I mean, I need a definition. First of all, if I have a... That's a loss. No, wait, wait, wait. No, no, come on. That's a definition of... No, let me propose a hypothetical question to you. WRI has millions of resources and just 100, 200 million from Jeff Bezos. Developers software, which is operational, which is open source, not unlike Global Forest Watch with its closed source. So I cannot believe Global Forest Watch because I haven't seen the code. Therefore, develop, make everybody test it, and then we agree. My problem here is, if I'm talking about operational system to measure degradation, first of all, the data must be open. Second of all, the software must be open source. Third, the methodology must be open. Otherwise, we're not discussing the same thing. The publish a PDF report accepts anything. Remote sensing of environment accepts anything that has 80% accuracy. I want the degradation measure that's better than 95%. Then we discuss. And I don't think the problem is high-resolution data. I think the problem is, what are you looking for? And can you reproduce once you fight it? Yeah, my hypothesis, that's not true. My hypothesis is the opposite. It's like... When WRI publishes the code that you use for Global Forest Watch, I'll start discussing. Until that day, you don't deserve, I only respect people who publish their code and who publish their data. I only sit in the table to discuss people who are open because I am open. That's fine. You can have that crusade, but that was not my question. But that's fine. It's a crusade that succeeded. Okay. If I had your money, I would do much better. And I think also there's a question here. Yes. Just wondering, which are the common definitions of degradation or... No, I don't have one. And I'm saying the problem here is, you would need to have a definition that would take an account. Their existence of a pristine forest at a certain point in time, the existence of an event that changed that condition to a measurable extent, in the existence of another event where that condition was mitigated in somehow or another, or in other words, there was a recovery of the forest. Otherwise, it's going to lead up to either the current situation or worse. So it only makes sense of thinking of degradation by thinking in time. By thinking there was something before, degradation is an event. It's not a final stage. It's something that happens to an individual part of a forest. So events you have to think time. So time is the crucial bit. Now, because degradation affects an area where there's trees and some of the trees need to be removed, it cannot only be on time. It has to be on space and time. So you have to have a space-time individual, like one of us, which is moving on time, we lose some hair. Like, think of it of hair, okay? You had a lot of hair. And then you're losing some hair. But eventually in my case, I'm 20 years now, like this. So I have a degraded hair. So this is a situation where you only, I have, this is my space and time has done this to me. So the problem here is there any good algorithm that takes space and time in an equivocal fashion? Not pixel-wise, but not only space-wise, but space and time. So we can actually come to a measurable definition of degradation. I don't see anyone available. Thank you. And I think we have one final question. Of course, if you can, please make it short. Thank you for the chance and thank you for the presentation. I think it is really challenging in order to monitor this degradation issue. We can refine and we can increase the accuracy assessment using the majorities resolution. But in our country, it may be difficult because in some areas you see disturbance under story types because of coffee management. Can we have solutions for such problems where the farmers integrate under story types but we cannot see the under story type rather than the canopy? It seems to me that you have your solution. Because you said you can see, okay? If you cannot see, you cannot use remote sensing, but if you can see a certain resolution, you can define degradation for your own purposes in your own country. And if you have a team of experts who is able to reproduce that for a significant area, you have a degradation measuring system, such as one we have in Brazil. It's visual, but it works. So if when you can't see, I mean, there's no measure here. I mean, there's no miracle. You can say P-Bensar, but has anyone used P-Bensar in a reproducible way? I mean, scientists will always tell you the answer is next year or next 10 years, but if you're looking now, trust your experts. They know better than the average scientist for your area. Great, thanks. Okay, there's one very short question, which I will allow. It's nice to have the concept. My question is like, you can show how the degradation looks like, but I want to know in your head how you're going to present it in the report, how it looks like. I think, like I said, I take the full openness approach. You know the slide I showed with the polygon? That is on the web. No, each polygon, each identified polygon of degradation is available on the web for free for anyone to download it. That's how we show it. We show this is a polygon of degradation. No, no, it's presented. It's a polygon. You can download the data and you open QGs or ArcGis or whatever. You present it. This is what- Doesn't make a number? No, no, no. It's this concept of degradation. Does not make sense without the spatial information. Degradation is really about something that happens in the space that your textual information remote sensing is able to elicit because your expert has seen it. So there's no sense in presenting the graded area other than by showing the polygon where your expert has identified the degradation. I don't see any way of getting credibility other than that. It might be automatic. It might be visual. I don't care. But it has to be put on a spatial context. Thank you. And with that point, I think that we will move to the next priority. And of course there will be time at the end for wrap up questions so discussions can be continued. So I would like to call Natalia Malaga from Wageningen University. She will present on biomass and emission factor estimation. So let me- Okay. Your perspective. Okay. Thank you. And you will try to do as good as Gilberto. Okay. So thank you for coming. It's a pleasure to be here. I'm gonna address biomass estimation and emissions factors from the R&D perspective in five minutes. Let's see. How do I pass the slides here? Ah, perfect. Okay. Okay. If we're talking about biomass estimates and emissions factors, we have to talk about NFIs because NFIs is how countries are always the main source of information countries used to build their reports and emissions factors nowadays. And so there is a clear difference. We see that most countries in the global north have at least one round and usually more rounds of NFIs implemented away. Whereas in many tropical countries, the NFIs are actually quite recent. And again, in the tropics, some of them have incomplete coverage. I think we heard earlier in the presentations a bunch of the countries actually also struggling to complete their NFIs. And when they do complete one round, most of them struggle to update these national forest inventories. Also, from the other side, from the observation perspective, what is the actual uptake of space based probes for biomass estimates? And this is a study recently published that compiles the information that countries use in terms of reporting for activity data for fires, but also for biomass estimates. And as you can see here on this slide, most of the countries actually use Earth observation probes for activity data reporting. Whereas for biomass estimates and efficient factors, they were actually only four of the ones that were analyzed that were using global biomass probes for reporting purposes, but only indirectly. That means that they were only using it to compare against their ground reference data or also to compare against IPCC default values. But then if we combine these two sources of information, what are the opportunities of also building some biomass estimates and emission factors? So as I mentioned, there are some countries struggling to complete or update their NFIs. And so if we were to use, for instance, space biomass maps as an auxiliary source of information to national forest inventories, we could perhaps look into enhancing the precision of subnational above ground biomass estimates. That means that countries could report at the higher precision that they do right now if they, for instance, have their NFI incomplete, but also could explore more efficient sampling designs for those, for instance, who are struggling to update their NFIs. And so this is a study that we are ongoing study actually where we are looking into using the CCI biomass maps in combination with country NFIs in the humid tropics, but also in dry forests and woodlands in East Africa to see what is the contribution of these biomass maps in these two different biomass settings and this range of biomass settings, but also acknowledging that NFIs in the tropics use diverse source of NFI sampling designs and plot configurations. They come in different shapes and sizes and with this side, we are trying to capture such a complexity as well in the map to plot integration. Okay, if we are also talking about biomass and emission factors, we also have to talk touch bases on uncertainties. I'm not gonna go in too much depth here because I think Chad is gonna cover that front, but there are many countries worried on how to comprehensively account and appropriate the sources of uncertainties in LuluCF reporting. And what do the IPCC guidelines say about it? So it is in general a good practice that countries near overestimate nor underestimate so far as can be judged. Well, that's one good practice guideline, let's say. That is like, for instance, using unbiased estimators to your estimates. Another good practice guideline is to have countries reduce their uncertainties as far as practical. But before reducing your uncertainties, you first have to understand them, estimate them, and then account for them and propagate them. So there is a study as well in Peru that we did when we were integrating these two different sources of information from NFI data and global biomass maps where we accounted for different source of uncertainties that are related with the plot measurements for the map, but also from the integration of these two different sources of data, let's say. But when we talk about biomass estimates and emissions factors, we also talk about change. And this is also a concern that has been addressed before. So the business as usual as countries do right nowadays is using their NFI information to estimate their greenhouse gas emissions and removals from the Luluzi F sector. And there are these two different approaches, the stock change approach, the gain loss approach. The first one, the stock change approach usually is done by countries who are half at least two rounds of NFIs, whereas the other in the other hand is when they only have one round or perhaps go into more tier three more complex approaches. And there are some opportunities here as well if we use a space biomass maps in combination with NFI data or without it. One would be for instance, the direct rough estimation of biomass on change or in combination with activity data to produce these world to world estimates. But my question, and this is a general comment here, are we there yet to use these biomass maps as a direct estimate? Or do we also have to consider the need for consistent time series pros to estimate change at least? And another way to go is to, for instance, combine time series satellite data with NFI information to try to address above ground biomass change. This is a couple of studies here where addressing biomass accumulation in the first year, during the 21st year after disturbance in the human tropics in Peru and after the 20 years post-forest establishment in dry forests in Tanzania. But the list can go on and this very much depends on the country needs and the opportunities as these global biomass probes or any kind of Earth observation probes and for countries and for their reporting. That's all, I hope. I think I, I mean, all the time. That was only slightly over time but we can start with some questions for Natalia. So I open the floor for any questions. If not, perhaps I can start with a question. Natalia. So like you say, for example, that products are improving over time and like biomass, space-based biomass estimations are improving over time. And what do you think are the main bottlenecks for their uptake for reporting and for seeing how emission, how they can be combined and used as emission factors? I think that it stands a bit also on the country needs if you're talking about reporting purposes and how like countries could also see this as an opportunity to embrace in their reporting purposes. As I mentioned, there are some opportunities there. I saw during this talk in the previous talk actually how there were some, like I think it was Colombia and also Kenya who were struggling to complete their first round. This could be something useful to see. Of course, if we were to use these biomass maps as direct estimation, for instance, I think that we should consider are they, like is this scale enough for reporting purposes of the countries, or do we have at least, for instance, consistent time series information of that to estimate change if that's a need? Those are some things to consider. But I guess that for reporting at large scale could be also interesting just as a purpose of verification and to compare what the countries are doing as well in their reporting. There, I see two hands raised. First, let's start in the back and then we go. Thanks for the presentation. I don't know if you assess those new above ground biomass product derived from satellite in the kind of a level of confidence we can have, especially for those time series of biomass maps are coming up. I wonder, Zanzolik, maybe with the previous presentation is if we stick with the IPCC definition of forest degradation, which is a decrease of standing biomass within the forest definition, what do you think whether those HGP maps can be used to estimate forest degradation? I will try to answer the same question without touching any of the details. I don't know, actually. The question is what would be the resolution needed for that in the reporting scale as well? Perhaps is it like if, I mean, I think are you focused on the pixel level and perhaps are these maps good enough to be reporting at the pixel level or perhaps you're building on more larger, aggregated level that perhaps in that sense that could be useful because I mean, they're a bit more reliable at that stage perhaps than only a pixel level, but perhaps it would be interesting to see how they could be used for degradation reporting purposes as direct estimates. But I don't know if Gilberto has something to add. Thanks. I just wanted to point out that until now we were not able to see questions from online participants, but yes, now I can see them. So finally, yeah, I can see them now. And then most of them were related to the first presentation, so I think we can just push them to the end of the final discussion, but yeah, I'm sorry about that, but now we will be monitoring the online questions. There was another comment here in the front. Yes, so I think there's a micro, and I think we were having some problems with the microphone before I read, but now it's working. It looks like it is. Thank you, it's a very good topic and also it gets more and more popular and the men get concerns for the first biomass as they mentioned, but I think I also concern from the first ecological view, I think people also think about what kinds of biomass we are talking about from a remote sensing perspective, like we've gone above ground or the hole we include below ground. We care about only for the trees, or we also care about the shrubs and grass and grasses. And not even for the trees part, we only care about the living trees, or we also like standing data trees, this is from different components. And also I think one of the main concerns is, you also mentioned that the remote sensing technology keeps you improving, like we get more and more lighter contribution, maybe in the future we've got to provide some contribution, but how we can get to the time series consistent because we need to concern to care about like a flux, or now we needed to multi-temporal biomass product like this kind of stuff. Yeah, I think those are very valid concerns, I can only say, like the time series consistency is evidently something really important if we want to address change, right? So we really want to report change, not whatever like the difference of whatever the measurements have to do with, or I don't know, circumstances of the measurement. But yeah, and I think that the first concern that you address has to do with a bit also the definition that the country defines of how they do their reporting, is there only in forested size? How do they separate the pools also? That's also not only, also hard to, it's not also something easy to do on the ground, let's say, so it's also important to keep the consistency in both of the definitions and then the harmonization of both data as well. But yeah, I think that right now it measures everything. You can only desegregate what you call for instance, well, you can also estimate below ground biomass from above ground biomass, but you could also, based on, for instance, segregating what are sharp lands from your forest based on whatever forest definition that the country uses as well. And perhaps using a mask, a forest mask or something like that could be useful. Or this harmonization challenges. Yes, indeed. And I think now we're going to wrap up the topic of biomass and we're gonna move for uncertainty assessments and uncertainty estimations. So I would like to invite Chad Babcock. Let me share your slides. Hello everyone. So it's great that I'm following up Natalia because I'm gonna be talking about uncertainty related to estimating carbon using remote sensing products such as above ground biomass maps and things like that. And I'm gonna try to do in five minutes, but I'm definitely not gonna make it. So we'll see where we stop. So just a quick, to kind of get us on the same page here, traditional forest inventory. So usually we follow a design-based sort of estimation protocol for that. So which means we set up a field, probabilistic sample field plot locations. We measure the amount of carbon in each of the plots and then we estimate carbon for the project area using those plots and an estimator appropriate for the sampling design that we chose. And then we can estimate error. And advantages to this type of estimation approach is that estimators, if we craft them correctly, will be unbiased. And then the error associated with these, so the standard errors we get out of these types of estimators we're used to being able to interpret these types of things based on the sampling design and we're very comfortable in this arena. But the major disadvantage here is that it's super expensive to collect field data. So that's why we would like to be able to bring remote sensing data in to hopefully figure out a way to try to reduce our sample sizes, to reduce the costs associated with taking with doing an inventory. So in, for instance, you know, LIDAR height metrics we all know are highly related to forest height and therefore biomass. And then we also have optical remote sensing sources that are good at looking at forest cover and all these sorts of things are related to forest biomass and forest carbon. So the trick becomes, how do we actually incorporate this stuff into our estimation approach? And there's two general, very general areas or approaches that we can use to be able to do that and still get out an uncertainty estimate that we're comfortable with from a sort of design-based approach. So we model assisted estimation where we actually use where we can use a regression model to relate the remote sensing data to the field data and then be able to go through and get and craft an estimator that falls within the design-based paradigm of statistical inference. And then we can also go into a model-based approach which becomes quite a bit more complex because we have to handle some pretty complicated error structures pretty explicitly but there are ways to do it. And now I'm gonna jump into an example here for in the state of Oregon on the west coast of the U.S. So we're gonna try to estimate forest biomass for this state and then also the counties within the state. So those polygons that you're seeing there and then we're gonna try to incorporate remote sensing data to be able to reduce uncertainty here. So what you're looking at here is a map of above ground biomass created using a whole bunch of remote sensing data and a machine learning algorithm. And then yeah, so there's biomass and so this will be our remote sensing product that we're gonna be bringing into the National Forest Inventory data we have. So this is the US Forest Service FIA data we're gonna be using as our response variable in our estimators here. Now we're gonna look at four estimators. So we're gonna do a direct estimator where we just try to estimate directly from the map. Then we're gonna look at a design-based estimator which only uses the field data and then we're gonna look at a model-assisted estimator which attempts to use both in a design-based framework and then we'll look at a spatial model-based estimator that tries to do it in a model-based framework. All right, so can we see anything? Okay, so we've got a direct estimate is the green dot there and then we got a design-based estimate which is the orange color and then we've got the purple which is our model-assisted and then the pink is our spatial model-based approach and we've got it for the years 2001 to 2016 and then those vertical bars are my error bars associated with the estimates. The big thing that jumps out here is that the direct estimate is substantially higher than all the estimators that actually include field data which is to me it's evidence that there is a systematic over-prediction of biomass within this particular product and then we can see with the model-assisted and the spatial model-based approach it brings it down back to where essentially the estimate from the field data alone comes in and so we can argue here that we're being able to kind of take that bias, we're able to correct for that bias and then still generate an estimate and if you squint you can actually see that the model-assisted and the spatial model-based confidence intervals are a little bit narrower than the design-based estimate which gives us an indication that this remote sensing product is actually approving our precision but not substantially and then we can zoom into a county level here so here our sample size is around 80 actual field plots in there but I'm just gonna bounce ahead to the one that's a little bit more substantial. So this one in this particular county we have an average of about 30 plots for each year and we can see that our model-assisted and our spatial model approach is actually narrowing those confidence intervals quite a bit but the spatial model was winning out here. So there's a spatial random effect inside of that spatial model there that's trying to borrow strength essentially from plots in neighboring counties to help inform the estimate in the local county and then we can go extremely well. So now we're looking at sample sizes at four or five or even we have three years here where we actually don't have any field data in that county and we see that that spatial model is starting to win out here but the model-assisted approach is also we can see there's improvements in a lot of the years but what's interesting is that in the 2005 and 2007 we actually didn't have any field data in the county but the spatial model because we're in a model-based approach we can actually estimate uncertainty in that model-based framework and that's how we are able to still generate an estimate there. All right, last slide. So we're out again, oh great, that's a great news. So I changed colors on you guys. So the green one, the green points are the design-based estimator, the orange points are the model-assisted estimator and then I have it labeled geostatistical model but this is that spatial model-based approach we were looking at in purple. So each dot here is the uncertainty interval width for a particular county year estimate and then the trend lines there, those are just a moving average to kind of see how things change on average and then we're plotted against the number of samples within the county in that particular year and we can see them in sample sizes are over 40 in this particular case study we see that the model-assisted and that spatial model-based approach seem to be doing pretty similarly but when it's sample size is less than 40 and it's super blurry up there we actually, we see a pretty big benefit from moving into the spatial model-based approach as opposed to the model-assisted estimator and then I've got three conclusions here but I'll only talk about the third one. So both approaches here seem to be able to lower forest carbon, can potentially lower forest carbon inventory costs by reducing the number of plots that we need in an area to get a reliable estimate and yeah, and I think I'll leave it there and just start taking questions. I need to monitor them. Yes, thanks Chad. So any hands up also from the panelists is accepted and also from the online Q&A I will check this time. I see one hand over there and another one over there in the... Thank you. Thanks Chad, so can you tell us about so the design-based approach is accepted as a protocol. You tell us about what it would take for a body to believe or trust these other approaches you're talking about. I guess for me to be able to believe these approaches I would be interested in looking at essentially the coverage probabilities associated with the uncertainty intervals. Those can be pretty tricky to be able to get though for small areas because you need to know the truth in that particular forest which is involved to actually going out and measuring every single tree and trying to come up with an estimate. But there are actually like... There are STEM maps that are pretty significant in size where we can do some sort of simulation study to be able to show that these particular confidence intervals are adhering to the properties that we would like to see which is why we like the design-based principle approaches so much. And we actually do... We are looking at a STEM map in Harvard Forest in Massachusetts where we're trying to run this particular type of simulation and I didn't have slides on it here but they appear to be working. So... Thank you for the interesting presentation. My question is... You mentioned about the 20 plots for the last one and what about the size of the plot? Do you consider the size? And the follow-up question is have you also analyzed the relation of the size? For example, in case of circular plots what happens if 10 meter, 15 meter, 20 meter? Did you do this kind of analysis? Yeah, that's an excellent question. We didn't do it here. So the forest inventory analysis program they have a very well-defined plot configuration and so that was the data that we were working with for here. But in theory it should be able to work with a lot of different plot designs. I would imagine that changing the plot size itself would change the relationship between the remote sensing product and the actual field measurement of biomass. And I've seen a few papers out there where they try to figure out an optimal size at least looking at, I don't remember, airborne LiDAR data and stuff. But yeah, so that will affect how the remote sensing product relates to the field data but the estimator should still work out in terms of being able to figure out what the uncertainty is for the estimate. We have a few more minutes. So I think we can do two short questions or maybe three, who knows? Yeah, very nice to see and good example of combination of ground data and satellite error or remote sensing data and how we can also do a proper accuracy analysis and error estimation. Your slightly showed the whole state there was kind of a trend and was it to average biomass, I think I saw, right? So the question is from that and that's related to the question that was asked before, I think. So with your estimation and your confidence in intervals or standard error, at what point you start to confident that you get a significant change in average biomass across the state? Yeah, that's a tricky one. So yes, the confidence intervals are, they're valid for the particular year but it's a stretch to kind of compare confidence intervals across years, especially with the model-based approach that we were looking at. Like so the model-based approach very explicitly tries to deal with spatial autocorrelation which is a big problem if we didn't deal with that. You would see confidence intervals that were super narrow because our sample size would be artificially inflated essentially. So when we start talking about things across time we then have to start talking about temporal autocorrelation which will also be a significant factor in here. So the best thing in the world would be, take that spatial model and extend it to a spatial temporal model but then we run into a bunch of huge computational challenges because the matrix that we end up having to invert becomes astronomically large. But yeah, so I think a better way to go about trying to estimate change is to try to actually have plot level measurements of change and then produce a change map between the two remote sensing products and then just run the estimator that way. Then we wouldn't run into that same issue but then you would have to have repeated measurements on some of the plots every year to be able to do that on an annual basis but usually we have repeat inventory. So like at least in the United States on the East Coast they re-measure FIA plots every five years so we can at least get five year estimates out of it that way but that would probably be the better way to go. Let's have first from the back and then here in the front. Yes, thanks Chad. Chip Scott from US Forest Service, Silver Carbon. One of my questions is, it was really interesting to see how as the sample sizes went down the improvement by using model-based approaches or model-assisted approaches really made a difference. Do you see the application for areas of missing data or particularly large blocks of missing data? So considering the model-assisted approach, I mean we would always need to have plots within the area we were trying to make estimates for. I think if we're in a situation where we don't have plot data in a particular area but we do have plot data in the surrounding areas then we would wanna be moving over to the spatial model-based approach. But yeah. One last question. Sorry, this might be a long and tricky one because it sort of combines everything that was said today. But no, I wanna sort of go back to Gilberto's comments about he'll only sit on a table where transparent and open source code is made available for such assessments. And we're well aware in the biomass harmonization community that no matter how much we're trying for the develop these model-based and model-assisted approaches, none of that is open source. So I guess I have two questions at the moment. How close are we to having this transparent and public and open source? And I guess then coming back to Gilberto is from the country's perspective, if you were to adopt remote sensing products, what level of open science and transparency would you expect to take this up for the update of space-based maps or your national assessments? So the model-assisted estimator and the spatial model-based approach I was using here, the methodology for that is open. I mean, there was nothing super special going on with either of those estimators. And the methodology is definitely publicly available. Yeah, I guess that's my response. Do you wanna chime in? You might all have known, perhaps, that we had four years in Brazil with the extreme right-wing government of Mr. Jair Bolsonaro. And Mr. Jair Bolsonaro was not happy with the fact that the EP was producing deforestation data, which he did not control. So he, on the paper, on the newspaper, said that my institution, EP, which I was a director, I was lying, at the behest of NGOs and foreign powers. And then he sacked the director, which was my successor, Professor Galvão. But, as Claudio said, the system did not go out. He did not mention, did not manage, because the reaction on the judicial system, I mean, the head of the Supreme Court says, it's not a decision the president can make, it's an obligation of public power to be transparent, the head of the Brazilian Academy of Science said you have to be transparent. So that tells you how important it is to be open. Otherwise, the right-wing government by Bolsonaro would have shut it down and produced, because he actually wanted, numbers that he would like to see. So I think there's no option. I don't see options to other than that. If you want to survive and you want to have a democratic government, it's as simple as that. Okay, I think we will wrap up the uncertainties and now we're going to move onwards to land use and greenhouse gas estimation. So I would like to call Viola Heinrich. I, first let me just, we're doing really well on time. So, I mean, we're slightly behind, but it's going better than expected, I think. Yeah. I think I might delay, I'm sorry. Hi everyone, yeah, I'm Viola. I have one foot still in Exeter, one foot in Bristol, and my hand is through the door at JFZ as of five days ago. So I'm going to be talking to you about land use and greenhouse gas fluxes, giving primarily an Earth observation perspective. And it's not moving forward. So I think we saw a version of this figure earlier from Martin in his talk this morning. And one of the key sort of issues around, oh, is that there are still large uncertainties and differences between the methods that are currently estimating land use fluxes at global scale in this case. And this was a figure produced for the IPCC, working group three, where we can see these large gaps that are emerging between different estimates. And this is really critical to understand and in some cases reconcile these differences because we're moving forward in the Paris Agreement and the global stock take. And so understanding these differences is really, really important. There's been a lot of work being done to understand the sort of the science global modeling estimates, which are on the top and the national inventories and FAO data in the middle. But actually what's been emerging is new data sets from the earth observation communities on the global scale. And that number or that graph is sort of down here. And we can see that there are large, large differences. And this is what we need to understand further. And that's a work in progress of how we can understand and indeed reconcile those differences. And part of the issue with these earth observation data sets from, especially from the forest environment is they're currently, so this is one example from the WRI, from the Global Forest Watch. This is potentially annual data from terms of the emissions from the forest. Estimates primarily deforestation. And one of the challenges within that, they're obviously, as we heard, there are issues around how we include uncertainty, how we include degradation and also associated recovery. And so this is one of the key advances that we want to move forward in the research and development realms and how we could potentially improve these global data sets on forest greenhouse gas flux maps, such as this one. And we've heard from Gilberto already that there are steps and potentially obviously issues around how we define degradation but there are, have been research that has really improved and really understanding the drivers of what is driving degradation and deforestation. And in line with that, we can use remote sensing and earth observation data to some extent. If we have a simple definition of what degradation and deforestation is, we can look at the associated carbon recovery or the removal that is occurring in forests that are recovering from deforestation or secondary forests. And moving forward within that, when we have temporal data sets from earth observation, such as the LVID data sets, we can actually begin to look at the temporal trends and really understand changes in the temporal patterns of the sink and understand the sink and how it's responding to climate change but also to anthropogenic drivers. But we're not just looking at forests although GFOI has got the word forests in it and we also need to look beyond that. So first of all, to really understanding of what else is going on within our follow, not just the forest, the F component and also look at the other land use, land cover changes that are arising. And on a global scale, there's been some efforts to develop and improve maps of looking at annual changes in land use, land cover. On regional scales as well, there are initiatives emerging, for example, the map IOMAS, one which started in Brazil and it's sort of bounced off to other countries like Bolivia, Peru and in Indonesia as well. And these data sets, although not perfect, can give us a good understanding of the annual temporal changes in different land use categories, and not just forests. And indeed some of these data sets such as map beyond must have been used to improve the global models, the dynamic vegetation models which are used in the global stock takes. So these remote sensing regional estimates can improve the other estimates as well. And I mentioned we need to look beyond forests, not just but also looking at the trees that are growing outside forests. So this was a paper published last week actually where they were using planet data amongst other things to really pick up and identify these trees that are occurring outside forests which within which are obviously of course, a lot of biomass and carbon stored within that as well. And a lot of temporal and spatial patterns are rising and we need to understand those dynamics as well. And I wanted to say something about it. So this is looking more at the land use, land cover component, but beyond that we need to move forward to understanding the greenhouse gas component as well, especially the carbon component. And that's sort of an element of research that's really in its early days. So this is an example from the See Trees Initiative based at JPL where they are looking not just at mapping the emissions and removals in the carbon stocks within the forested landscapes but also moving beyond looking at wetlands, shrublands and grasslands. But I would say that this is a really emerging field that there is a lot of scope and further research to be done in. And sort of going back to the first figure that I showed and looking at these, understanding these and reconciling the differences between national inventories for example and Earth observation data sets and what lessons we can actually learn for measuring reporting and especially verifying and these different emissions, these different estimates. It's really important as we've heard already that each of these data sets and information have their own assumptions in how for example they consider land to be managed, the different Lulu CF categories that are included. We heard in Indonesia there are 23 categories and different countries have different number of categories. And of course we heard about the uncertainties. So these are all elements that we need to consider moving forward within comparing Earth observation data sets within data sets such as inventories. We've also heard a lot about transparency and this is really important moving forward especially within the global stock take as part of the enhanced transparency framework. And that includes a number of things such as being open in how you obviously present your data and making the data as far as possible open access. And that just doesn't go just for the academic world but also as far as possible to other rounds in the policy area as well for example. But what has been shown is that we can use global Earth observation data sets that they can be used as a useful benchmark. For example when comparing to inventories. And when doing this it's really important to ideally use countries definitions of forest and their land use and cover as far as possible and as is defined. And this shows the potential of what global Earth observation data can do. And just moving forward to some of the key priorities beyond what we've done so far. I think as I've mentioned before there is an effort to move beyond forest and looking at the full mapping of a folly. But we also need to incorporate the, look at the associated greenhouse gas fluxes which is an emerging field of science at the moment. There are numerous and growing number of global products and these are really advantageous because of the internal comparability that that provides us. And that's especially important for the global stock take. But on a regional and country scale we need this spatial information and that's really critical for making these comparisons. Because these often do provide greater detail in terms of the land use classifications but they can also be used to harmonize the global products. And finally you sort of to conclude I think it's really important that what is really what we need to continue doing is comparing the national information that's available with the global earth observation data. And finally, yeah, that is really important for the global stock take which is a very critical element of the Paris Agreement. So thank you very much and I'll go back to my seat. Thank you Viola and I would like to start with a question online, our first online question. So Em Mutinem from FAO asks, it would be excellent to hear from you Viola if there are plans to integrate soil organic carbon emissions, for example from mangroves and pitlands. There's an active working group of experts on pitlands within the global pitlands initiative. Hello, thank you for the question. I think this is one element that was raised earlier as well and looking at what earth observation can provide in terms of looking at what's happening below the soil and looking at the soil carbon elements. And this is something again, that shows the potential and the importance of continuing to use a field data to complement the earth observation data and sets that are emerging. So I think we can use these land use, land cover annual or sub annual data sets to look at the land use changes. But when it comes to looking at the greenhouse gas emissions from soil for example, it's critical that we continue to involve the field inventory and communities within that. Thank you. Are there any hands? I see one hand up here, center front. Thank you, Viola. My question will be like, why we have to compare between the national data and also the global? Why we have to combine? Because that is different. The country they have their own reporting and the global could be something more indicative. Why you have to compare? Oh, can you hear me? I am not a Jack-O-Mor-Grassy, I should say upfront, who is the expert in this. But I think moving forward in the global stock tech, there is an effort to understand the sort of the progress in which we're moving towards in reaching the Paris Agreement goals. And if we're comparing the sort of the progress that we're making is made using, looking at national inventories for example, but whether we're on track to sort of reaching the Paris Agreement goals and looking at how we can achieve those goals, that's done using bookkeeping models and other sort of independent scientific estimates. And if these two datasets aren't speaking the same language or we're comparing apples to oranges, whatever comparison you wanna make, and we can't understand if we're on track to meeting the goals of the Paris Agreement, which are based on the sort of scientific, independent models compared to what the inventories are reporting and those are what's being submitted to the UNFCCC, so we really need to understand that. And the Earth observation datasets can come in and provide a sort of an MRV perspective within that. So I hope I've answered your question. Oh, sorry. Well, some questions are difficult to answer with it. Any other hands, any other questions? Let me check. Non-unline, more questions. If not, we can also move forward for the general discussion. So we've already heard from the five topics and of course, like I said, there was some questions that we would like to keep and discuss before we close this session and for that, let me share my slides again. Yeah, well, yeah, I managed them. So let's see, we have a red person there. Yeah, so we have some interactive things that we wanted to try, but actually it is not working. So let me get to the question. I think everybody can see my slides. Okay, so everybody that has a phone might, oh yeah, thanks. So for everybody that has a phone, I said that we should think about the questions and I would like for all of you to participate, including the people that are joining us online. And then the first question is related to the topic that we might have not discussed today or we might have mentioned very briefly that you think should be included as a priority. And of course, even if it's, for example, a small topic beneath the priorities that we have already discussed, but that you would like for the R&D component as a community to focus more on, please share it to us. Share it to us. So I see that somebody put the code, but then there's also topics related, for example, to regrowth and restoration monitoring which are interlinked, the frustration free commodities or tracking commodity changes. There's a big one about capacity building, of course, topics on soil organic carbon, which we discussed in the land use priorities as well. Multi-purpose national forest monitoring systems that is very relevant, especially when we're discussing the number of plots that we should have. Also, deforestation risk mapping, but I think that the largest ones are soil organic carbon, restoration, regrowth, capacity building and restoration. If anybody wants to comment on why they put one or maybe on one of the ones that have more votes, please raise your hand. This is meant to be interactive in many ways. I see a hand in the back, yes. Thank you, and thank you to all the presenters for the presentations, really interesting. I put restoration there and I was really looking forward to hearing about regrowth, which we didn't hear much about, but I would just like to emphasize that I think it would be great to have restoration as an additional topic. In here, I think restoration and regrowth, it's nice to pair them together, but restoration is also kind of beyond regrowth, restoration, there's a whole spectrum of restoration. I think there's a lot of opportunities for research and development. And at FAO, we are coordinating the UN decade on ecosystem restoration, and so it'd be really great to collaborate with GFOI to just support increased research and development on restoration, we're leading the task force on monitoring best practices, also working closely with the science and finance task forces of the UN decade. So really great to connect on that. No, thank you. And definitely something to discuss as well and to keep in mind. I will move on now to the second question. We only have three. So the second question, as a GFOI R&D community, of course we're starting back up and there are several ways in which we can engage. It doesn't mean that we are planning to engage in all of these ways, but of course we want to see which ones you find the most relevance to actually work towards a common goal. So we have expert meetings and workshops, regular online meetings, newsletters, webinars, online forums, or discussion boards, also collaborative projects and special issues, which, yeah, and special issues in journals, for example, what do you think would be the most relevant? And while we're answering that question, we might want to answer one question from the online on degradation mapping, Gilberto. So I think it's relatively a simple question to ask, to answer, let's see, to ask, yes. What percentage of canopy coverage in a pixel area are you considering for defining forest, non-forest in order to calculate degradation? And what percentage of forest change are you considering to conclude whether there is degradation or not? Well, I think it's very easy to answer, but very easy to ask, but very difficult to answer. Again, it depends on your definition. I would not consider percentage of pixels. I would imagine that the pixel, whatever representation represents your data and you have data on, again, on a timescale. So you may use mixture models, which is nice to we've seen that mixture models can provide very interesting indications of degradation, especially when there is some fires which cut part of the cut through the forest, which has already been degraded by logging, for example. But I would argue the problem here is exactly on the definition of the textural information that you are going to want to capture in your remote sensing data. You would see, well, the texture of a degraded area is different than the texture of a pristine forest. That is obvious for everyone who has looked at the images. How are you going to delimit these polygons, which, where does the degradation stop? And this is really something that it's very hard. And I would encourage those who are interested in develop. I don't think there is someone who's going to come up with the answer. Like in most science, things go incrementally and everyone contributes to the answer. Thanks, Gilberto. And just looking quickly at the results, we have seen that 34 people have answered here and online as well. So it looks like expert meetings and workshops are a clear activity to take, followed by collaborative projects, webinars, online forums or discussion boards, maybe regular online meetings as well. And special issues and side events are the ones at the bottom. But of course, there's different ways to engage and this is information that we will definitely take up in the planning of our R&D activities in the coming years. And since the first expert meeting and workshop was, since that was a topic that was chosen by most of the people, then that's good because the third question is related to that. So just, I would like to ask everybody here, what do you think our next R&D expert workshop should focus on? I guess indeed, yeah. I mean, if they are priorities and at some point they should be addressed in a dedicated meeting, right? But from the answers from the participants here and online, it looks like also touching upon what was already said at the beginning of the final discussion. Regrowth and degradation mapping is a topic that is of interest. And also like there was a dedicated workshop for countries and how they report on degradation. So this is a topic that is a challenge to all of the sub communities within GFI. But of course, the answer is all of them, right? I would hope so. So I think this is the last poll and with that we would like to conclude our event. There are some questions online that are still missing but I think that we can move forward unfortunately. But I would like to thank all the panelists for their great work presenting, keeping somewhat to the time schedule and also for the discussion held today. Thank you everybody. And Martin, I don't know if you want to say anything. We all made it through day one of the GFI plenary. So more days to come, not yet fully. There is still a reception and that's next thing. So everybody invited down in the atrium basically in this big open hall when you walk in, there's some nice posters, there's some people to talk to, there's some snacks, ice cream I've heard. So that's the next stop. And then I think in our later, there is from six to eight, there's drinks and snacks on the big terrace upstairs. So the social side of today, I mean, there's a lot of social interaction already, very thematic, but now also maybe focus more on the social side. That's the time of the day where we start that. So thanks again and we'll see you all down and later on up on the terrace. Thanks.