From Data to Decisions: Geospatial AI for Sustainable Land Management and Climate Resilience - Action Dome, UNCCD COP17

They have got their finger on the pulse when it comes to the future. You need to sit back down and take part in this particular discussion because we are talking about, oh, wonderful, about AI. Okay? And if we're talking about jobs and if we're talking about the importance of how the environment and climate and processes and solutions are going to fit into the future, I'm sorry, you need to be aware of the role of AI. So it would be a shame for you to miss this. So this next session, ladies and gentlemen, is geospatial AI for sustainable land management and climate resilience. Thank you very much indeed. Thank you.

Hello. Okay. Good afternoon, everyone.

That's correct. Okay.

My name is Sukanya Randhawa. I work as lead AI for good at Heidelberg Institute of Geo Information Technology. And yeah, so today, let me start with a talk, yeah. So welcome to beautiful Ulaanbaatar. And if you stand outside this venue, you stand on land that has sustained pastoral life for thousands of years. But today landscapes, not just here in Mongolia, but most of the arid and semi-arid regions of the world are facing unprecedented strain. Today I'll talk about hope, data, and action. I will be talking about how we could use geospatial AI to not only understand the landscape, but to transform that into practical opportunities for land restoration, climate resilience, and sustainable development. Basically, how we could use sophisticated artificial intelligence as a practical toolkit for local communities, practitioners, and stakeholders. So today the focus is mostly on the two IMAGs of Mongolia, the TOV and Ulaanbaatar. We chose these regions as they experience intense convergence of rapid urban sprawl, high grazing pressure, and extreme climate impact. So to reduce this pressure, we look at three complementary opportunities, namely responsible solar, off-road water harvesting, and roadside blue-green infrastructure. This project was funded by the UN G20 Land Initiative, and was led by Heidelberg Institute of Geo Information Technology, also called as HIGIT, in collaboration with Aurville Consulting. HIGIT is based out of Germany and Aurville Consulting is based out of India. So we had quite an international team. And very importantly, we collaborated with the local stakeholders here, the National Federation of Pasture User Group of Herders, led by Dr. Bulgama Desambu, who's also present here. And this was very important because she helped us understand the local Mongolian landscape, how to interpret the local challenges and provide us access to data, which was very important for the results that we provide for the maps that we created so that they reflected the real world conditions. There's a lot of noise at the background. Can we like kind of maybe just ask everyone to Could I request some silence from the audience? It's just a bit distracting. Thank you. Yeah, thanks. So this is a, you know, story of how global technology meets local experience. So let's look at the workflow for this project. This is like an overview and we look at the main components that were important in this project. So first, it's basically connecting the data to action workflow, the dots. And first we look at the top left, which is the innovative land cover mapping. Now this was critical as this was a core foundation for most of our analysis and to ensure its accuracy was a critical first step for us. And we use sophisticated AI algorithms to achieve this. And I'll be talking about more in detail about this in the next upcoming slides. Then we also look at the environmental and social pressures on the landscape. And we do this by generating a land degradation map that looks at the overall degradation and the human pressure, which of course looks at the influence of human pressures on the landscape. Then we combine all of this information into a more clearly defined actionable zones. So we are moving from data to action in what we call as a management priority zones. And all of this information, the management priority zones, along with the innovative land cover mapping is fed into the geospatial tool, LILA, which is a short form for LifeLens. And the tool is a geospatial tool that does a suitability assessment using the multiple data layers that I just described and provides results for restoration opportunities in three key dimensions, that is the solar, water, and roadside blue-green infrastructure. So this brings us to the critical mission of our project, which aligns very well with, of course, the goal of the SCOP, Restoring land is not just about restoring ecosystems, but about restoring resilience, livelihoods, and hope. So just to give a broader context to this work, land is, of course, you know, lies at the intersection of climate, of water security, biodiversity, and resilience of local communities. So we know that the pressure on Mongolia has been quite acute. So if you look at the rising temperatures, it's been three times more than the global average. The precipitation patterns have changed quite significantly, and more than 70%, according to the report, 77% of land, Mongolian range lands, have shifted from healthy to altered condition. Yeah, it's a grim situation, but we look at it as an opportunity, an opportunity for geospatial AI to take this situation and transform it into, you know, opportunities to restore land and target restoration in this format, right? So we looked at examples, restoration examples around the world, and we were inspired for a few of them, as you can see over here. We looked at complementary opportunities. In yellow, you can see the solar projects that have been used. But what is special about these projects, most of them, is that you know, they do not provide just one benefit. They work on complementary benefits, so providing multiple benefits at the same time. So we also have in blue, you know, the water projects across different parts of the world. And then in green, we had the roadside blue-green infrastructure. So we got inspired from these projects, but these are just possibilities. And we know that a restoration Planning project cannot be simply transferred from one part of the world to another, right? You have to understand the landscape that you're working with and the people who depend on it. So, we looked at the Mongolian, coming back to the Mongolian grasslands, we realized that it's a diverse set of ecosystems, just as shown by different pictures over here. There's a gradient of ecosystems, right? It ranges from the mountain grasslands, from the dry steppes to the actually productive steppes, then to the barren and the sand. And in order to, you know, suggest restoration practices, it was important to capture these finer ecological gradients or details at the beginning itself. Only then one could actually make effective decisions for the different places. So, very importantly, the pastoral life livelihoods, it depends very strongly on these grasslands and they rely on healthy grazing systems. They rely on reliable water access and seasonal mobility. So we kept in mind these different dimensions while designing specifically the restoration opportunities. So let's come into the first part, which is the innovative mapping approach, which the idea was to capture the ecological and social realities of the landscape. So what is interesting is we first looked at the global land cover products that were already available, because land cover maps are usually the starting point to most of the restoration work. Now, as you can see, we looked at European Space Agency, which is ESA written over here, and mostly it shows just one color, one class, grassland. There's no finer difference or ecological classes that are captured. So they fail miserably at capturing these finer details. Then we looked at the Google's dynamic land cover map, and we saw that most of the grassland was interestingly falsely classified as cropland, as shown in yellow over there. So that's a huge blunder. And so we realized that in order to do something meaningful in terms of ecological assessment, we had to do something more. We had to build our own maps that were not just accurate in capturing the ecological gradients within the Mongolian grasslands, but it could be scaled and applied to other parts of the world as well. So that was a challenge. And how we did that, so how did we do things differently? What was the secret recipe, so to say? It was bringing context in the learning. So we did not just rely on looking at what a satellite sees in an image, but we provided enrich understanding of the landscape. And that's what I mean by context. So how did we do it? We look at not only just multispectral satellite imagery, But we also look at some Google's Alpha Earth embeddings, which is nothing but compact representations of the environmental context of the land. And this is learned through vast amounts of data. And so we combine both these streams and put, you know, we build a custom deep learning model that took all these inputs and produced as outputs accurate and consistent land cover maps. So this was the model, how it looks like. I know it can be a little intimidating, so I'm not going into the technical details of this model, but we just stick to what is important in terms of decision making perspective. So the few important features that stand out is that the model is multimodal and context aware. By what that means is it's able to intelligently combine these different streams of data. and produce consistent and accurate land cover maps. And this is quite an achievement for these arid or semi-arid grassland type of regions because as you saw, Google and ESA and most of the global products really failed. So, which is where our maps brings in the cutting edge, so to say. And then this model is scalable across different regions. And we paid a lot of attention in terms of validations. As you can see, we used about a million pixels. There was a rigorous validation exercise, and we achieved overall accuracy of 80%, which is quite significant. And we had 10 classes, and we provide a pixel-level confidence score, which could be important in terms of management, you know, for decision-making, to take into account this confidence score. So, but okay, so now we know how the land looks like at today, for example, right, at this point. But what we really need to know is how it changes over time. And that is also another critical information. And coming to that, as what you can see on the right here is a time series of different land cover maps that we generated using our algorithm from 2020 to 2025, so over a period of six years. And what is very interesting is that you can see it's really striking that the greenness changes so significantly over different years, right? The signal is just jumping. And what does this mean? It means that this grassland, Mongolian grassland is a very sensitive ecosystem. It responds very sensitively to the rainfall, right? So if you have a very dry year, most of the land looks very dry and degraded, but that doesn't mean that the ecosystem function has degraded. So it was very important for us to distinguish between climate variability and land degradation, right? And so we realize that NDVI or greenness is not enough to, you know, to be used as a metric to identify or define land degradation. So we found our hero in another metric, which was the rain use efficiency, which is basically the greenest normalized by the rainfall. So with that, we could kill the climate noise, so to say, and that essentially what in simple word it means is the productivity of the land or the response of the land to the rainfall. So if it's an altered land, the response will be different from that of a healthy land. So that's what we capture over the years. So now that we have a signal, we look at how the signal changes over time. So we look at the trend over six years. And what you see over here is this data, we convert it into action, basically three management categories, which is declining, which is basically that the signal was showing declining signals. What basically means is that you need to prioritize restoration in these areas and, you know, reduce the pressure. Then you have a more stable signal shown in white over there. And that basically means that you monitor and maintain the management. And when you have a blue signal, which is showing actually improving conditions of the land, that would mean that you protect something, maybe restoration is already happening in these places. So it's protect and to avoid new stressors in these regions. So we know how the land is changing, but what are the drivers behind this change? According to the definition of UNCCD, land degradation is not just an ecological process, but it has the human pressures have a strong influence on land degradation. So how do we capture this dimension? So in order to capture the human pressure, we have three indicators, which is looking at the average grazing pressure, which is a six year weighted livestock density. Then we look at the persistent grazing over time, like which areas were persistently showing high intensity of grazing. And we look at the long term human influence as well. So we captured these three things and put it together into an index or a normalized index that we call a social pressure index, which is shown in the map over here. The more red it gets, the higher is the human pressure. And so once we come with this SPI score, what we call a social pressure index, we know how the human influences the landscape. Now we have two pieces of the puzzle. We have the ecological, we have mapped the ecological degradation and we have mapped the human pressure on the land. Now we bring the different pieces together where what you see over here is basically where everything comes together. And, uh, we convert basic, the AI, um, you know, the innovative land cover map that we had, the land degradation trend, the social pressure, all of that is combined into management priority zones. Now these are these clearly defined actionable zones, and we give it different categories, which lands are urgent. which require urgent attention, which are at risk, which are more robust and could be monitored, or which are severely degraded or more stable, or which show recovery signals, right? So instead of asking decision makers to go through complex technical layers of data, we translate evidence into these clear actionable zones. And this is helpful to further, you know, not just we have mapped the restoration, but to further in the decision making process to empower the local communities, the practitioners and the stakeholders. And so this becomes a special bridge between evidence and investment ready interventions. Okay, so now we know, you know, we have these maps. What do we do with them? What is the next step? Where, the question that one would ask is, where can one, you know, build or create interventions that would have the maximum impact or bring maximum impact for both nature and the people? And this is where we come to the second part of our story, which is restoration opportunities. So we move from data to actual action in terms of restoration. And we do not see the three dimensions that I mentioned before, that is clean energy, water security, and grassland restoration as competing land uses. We actually look at them as complementary efforts. And we know that, you know, degraded land because of the reduced vegetation cover has high runoff, increases the soil erosion, and we look at these three different opportunities, like I said before, in a very co-existence manner, and we, you know, screen, prioritize, and implement it for further planning. Now, how we do that is We look at these different objects as going a little bit in detail of what these opportunities mean. When you look at solar, we basically look at two different components. So by responsible solar, we mean that co-locating a solar plant with livestock management. So what that means is that with sustainable grazing management approaches, one could even limit the livestock per unit area. And then alternatively, we also look at solar co-located with nature-based solutions. Mongolia has a very strong native shrub system culture. And the idea, we designed this keeping in mind this reality. And so, you know, we could use solar and the native shrub systems, you know, the growth of these shrub systems together. And it gives a dual benefit of productive land use as well as clean energy. Second opportunity is the off-road water ponds. Now interestingly, when we talked to Bulgama, she told us that most of the grazing pressure is very much concentrated near the water areas, which is not surprising because most of the animals need water for their sustenance, right? But this means that the degradation was also concentrated and these areas faced high pressure and stress. Now, in order to diversify, the idea was to understand which other lands also have significant potential in becoming water ponds or for water interventions so that we could diversify or dilute this current existing pressure on the lands. So I'll talk more in detail about these, of course, in the upcoming slides. And the third opportunity was the roadside, using the roads as an opportunities, as opportunities for blue-green infrastructure. So I talk in detail again later. So how we did that is we kind of look from broader opportunity to more potential and implementable sites, right? So we narrow down the landscape using this three-step screening process. And eventually, after applying the suitability criteria in terms of environment, social pressures, we get a priority ranking, which basically means which sites should be prioritized first to bring these interventions in the grasslands and drylands. So that was, of course, our focus. Now, how we did this is using this tool, LILA. It's a geospatial tool that I developed a few years back in collaboration with Oroville Consulting, where we use different data sources and different data layers. And there's a geospatial analysis that is done, a suitability criteria, which is customized based on the land or region of interest that you're looking at. And the outputs are then the recommendations as to where the restoration opportunities can be planned, which are the priority sites, so to say. So I'm not going into details of this. is just a visual rendering of the tool, but I will show you the results that we got from it. So coming to responsible solar. So what are the benefits of solar? You know, of course, one is co-location, like I mentioned earlier, so you have productive land use as well as clean energy, but at the same time, underneath the panels, you're also creating microclimate, and this microclimate is beneficial for soil moisture, right, and the soil moisture is very beneficial for, especially for grasslands and dryer lands. So an increase, uh, soil moisture content would also means that you're regenerating land in the long run. So, um, let's look at, before we actually look at the context, solar context in Mongolia, I also want to mention that we look at the commercial viability of these solar projects. And for that, we also took into account the distance to roads and to the power evacuation infrastructure because that had to be taken into account for the commercial viability of these sites. So surprisingly, here is a high level take on the energy demand for Mongolia. And Mongolia is a solar powerhouse. Maybe unrealized right now, but hopefully it will change in the coming years. But this place is a potential solar powerhouse because also if you see the current demand, if you look at it, you know, the solar capacity needed to fulfill the current demand is about seven gigawatts. But if you look at a World Bank projection by 2050, taking into account the growth scenario, the solar capacity required is about 12 to 15 gigawatts. We just keep that in mind as I show you what is the potential that we actually came up with. So if you just look at this map, this is the first level of that from the three-step criteria that I talked about before. And here the total, we take look at the total area and then we screened which lands had the basic potential to meet the solar site. We take into account elevation and other, you know, factors and here we got 14% of the land. Then we look at the technical potential, like I said, looking at the land size, the solar irradiance and the infrastructure distance. So both power evacuation as well as access to roads. So looking at that, I'm not showing that particular map right now, but I'll straight away go to the highest potential, which was about, you know, taking into account actually the, uh, the social pressure index, like where is the grazing pressure is the highest because we want to co-locate it with those sites. So that comes to 0.5%, so about 400 square kilometers of area. So let's look at this other high potential sites. What does it mean? 25 gigawatts of solar capacity only in TOV and Ulaanbaatar regions. And this could not just fulfill the entire Mongolia's energy demands, but Mongolia could also potentially become the solar exporter to the world for the clean energy. So just imagine the potential. Now coming to the water opportunity, we look at what are the benefits, of course, the design intent was that you are able to capture the runoff before it degrades the lands, right? So you're harvesting the rainwater. Now we know that overall the rainfall is very low in Mongolia, right? When it comes to, if you compare it to the global average. But that's not a problem because you can still harvest this water, you can still build water ponds if we design it strategically, right? So if we have obviously accurate maps, one can even do restoration very strategically and accurately so that it's effective. So, um, and the logic being that it would support long-term vegetation recovery and, you know, you are diluting the pressure again, like I mentioned before, from the current sites which get degraded because of the high concentration of grazing pressure in the current, you know, areas near the current water ponds. So how do we do it? So first, if why it matters, I think, I don't have to go through this again, but there's a huge opportunity again in Mongolia. Most of the rainfall is captured in summer, and the livestock dependence on water is very high. So we are looking at almost 71 million livestock, and the idea is capture the seasonal water where it falls and make it available for the grazing. So how we did this is using, you know, the hydrological approach where we start with the elevation, we look how water flows through this land. So that is shown by the water streams here. So we map the different water flow. And then we also look at the different units, the hydro sheds or basins, however you want to call it. And those are then looked at which are having the highest potential in terms of capturing this runoff. Those are marked in blue over here. And once we have this information, which is again based on a lot of other layers, once we have this data, we look at which lands within these hydro sheds are best for accumulating water or holding water. And so once you have these maps, local communities can decide if their land has a water potential, how big they want the water pond to be, and the further other details. So Coming to the water upon opportunity maps, we have the theoretical potential. Again, we started with, you know, 13.7%, which is huge because Mongolian, the area that we're looking at is also fairly huge. And then as we filter down from technical to the highest potential lands, there is again a huge potential. Now, this is just a potential for, you know, of course, it's up to the communities to further take this and use it for decision making on the ground. And that was the whole idea. We keep it flexible. We want people to use these maps for their benefit. And we would be making all these maps open because at Highgate, we do believe in open maps and open technology. So these are the water ponds with highest potential. Again, the color is ranked as per, you know, which are the zones that require urgent attention and which are the zones that are at risk, again. So combining all the priority zone information for, you know, actionable, further actionable data, so to say. Now coming to the third opportunity, this is a roadside blue-green infrastructure where we, use road drainage to slow, spread and retain water across degraded landscapes. And we also know that a lot of grazing settlements are actually situated in close proximity to the roads. So the idea is to stabilize these corridors using the native shrubs or trees to reduce the erosion and to increase the moisture. And this starts Basically, there's a connection to the previous water pond module that I presented because where we look at all the land and where is the potential and here we look where are the roads intersecting with those maps and we come up with, you know, which are the potential or the highest relevant sites for this particular opportunity. So I think I've already spoken about why it matters. We have a total road network of that is mapped perhaps even more, this is probably an underestimation, which is something like 112,000 kilometers, where the paved road network is about 9,000 kilometers. And the idea is that one could potentially construct swales, check structures, or roadside ponds as part of this intervention. So coming to the technical potential, here we see the possible sites where one could plan these structures and here are the sites with the highest potential and it is just one percent of the suitable area but still it would make a huge difference to the overall regeneration of the area because we are almost looking at 743 square kilometers of land. Summarizing my talk, I would like to say in short words that we have looked at spatial intelligence to answer three key questions. Where is the need? Where is the opportunity? And where could one plan intervention where there could be multiple benefits? Which multiple benefits in terms of not just ecological benefits, but also in terms of economic value for the Mongolian people. And this, of course, requires diagnosis, prioritization, and delivery, and so we move from mapping to regeneration. Earlier, I showed you what were the possible interventions that one could look at, you know, the different global efforts that were happening. Now, the idea is not just to learn from these success stories around the world, but we want Mongolia also to be part of these successful stories. And this is where we show that, you know, the possible, we want to pin Mongolia on this world restoration map. Yeah, and this is how we move from global experience to local action. And one thing meaningful that struck me along this journey was that this journey could be actually symbolized very uniquely through a Mongolian symbol, the soyombo, which represents continuity, harmony, and aspiration for the people of Mongolia to flourish. Now, our technologies might not be able to relate in traditional sense to these ideas, but the objective is strikingly similar. We want that, you know, the Mongolian people restore the, we want to improve the relationship of the Mongolian people, the land and water, so that Mongolia remains resilient in the future and flourishes. So with that, I would like to thank you for for your attention. And last but not the least, I would like to acknowledge the wonderful team without their effort, their hard work, none of this would have been possible. And also I would like to thank the UNG20 team for funding this initiative and for supporting us all along. So thank you very much once again.

Yeah, can you hear me?

Can you turn the microphone on?

Can you hear me?

Yeah. Yeah, sure.

So how can we share the PowerPoint?

I'm student. Okay, okay. So you've just seen now what geospatial AI can actually do and find and quite impressive results. But I want to spend the next 10 minutes or so on a question that rather follows up on this. And the question is, what does it take so that results, as we've just seen, can actually be actionable so that someone does something about that? So it's not working? Okay. So, there you go. Thank you. So yeah, now it's better. Did you hear the starting words?

No.

So, so I will start again, so because I think you couldn't hear me before. Okay. So you have just seen with Sukanya's presentation what geospatial AI can find and do and how impressive the results are. And I want to spend the next 10 minutes or so on an often neglected question that follows up on the results we have just seen. And the question is, what does it take for results like we have just seen to change what someone actually does about it. So how can we actually become, create agency and impact from it? I'm Kirsten van Everfelt, and I'm a professor for geography, and I'm working at Highgate in the climate action team. So a lot of Research on climate change and land consumption or land degradation is based on a rather comfortable assumption. And the assumption is that information, so data and facts, will increase awareness. And then from this awareness, there will result action. So the idea is give people better data, they will understand the data and the problem, and then they will act. But latest evidence from neuroscience and behavioral research shows this is not the case. It's actually vice versa. You have to act in order to believe. So this is happening through something researchers call self-persuasion. And Christian Meyer's group, you can see the notification there, and if you want to learn something really fascinating, please check it out, because he shows it over and over again that it's really the other way around. People have to act, and then they will start to believe, and their world views will change. There's a rather famous biologist, late biologist, who was called E.O. Wilson, and he has put much earlier this same sentiment. in a very famous quote. He said, "We are drowning in information, but starving for wisdom." And I think it's sadly very true, and we find that when we watch the news every day, because we are informed, but somehow we are not getting any wiser. It's sad, but I find it rather true. And two things follow from this. First, our data on its own will not create action. Data and information alone doesn't help. So the second part of this conclusion is that what's usually not missing is awareness. We are aware. People know. People know the numbers. People know what is at risk, but it doesn't create action. And they don't need more information on how dire the situation is, they don't have to be more afraid than they already are. So the whole large picture is not enough, so we need to give them something else so that they can act. So this is one example for a large picture data where we, you know, everybody knows about this, knows about the CO2 concentrations in the atmosphere. And it's scary, it's alarming. I personally am very afraid. But the problem is that fear usually triggers three actions in humans. It's fight, flight, or freeze. And the majority of people is freezing, so not acting. Part of the population is fleeing into denial and denying that anything is amiss. And part of the population is actually acting. But you can see that this leads to polarization. And this is something we also observe a lot in society at the moment. So this is... This is the backbone why HighGIT follows a different approach. We don't want to add to this picture. We are building a dashboard with local data and local people so we can ensure usefulness. People, the idea is that people won't feel detached and alarmed, but they feel empowered to act so that they know where and how to act. So in a nutshell, we want to da- turn data into something people can actually act on. So land degradation and climate change have a lot in common. For example, they are so-called creeping catastrophes. What does it mean? A, they are human made, and B, they are slow enough that it seems as if a single day doesn't matter. They are not the big bang where everything changes and you know like in a wildfire or flash flood, you know you have to act now to protect your home. They are slow, they are creeping. They don't look dangerous, although they are, and the most dangerous part is that we don't realize how dangerous they are. So the damage doesn't announce itself and it accumulates until the decision has already been made for us and we can't decide anything anymore. But the good news is there's something we can do about this. There's a way out of this dilemma. Local action is fast and it's very effective. So the decisions that actually stop this damage accumulation can be local. A herder, a ministry, a municipality can not only lead by example, but they can make a difference, a real difference. We have heard that in the opening statement today, in the opening ceremony too. So what's the problem is that most earth observation dashboards show good and robust information, but they stop just there. They stop at the information. They offer high resolution and global data, but they don't bridge towards the local dimension so people can see how to act. This bridge can be created by working on the dashboards with local people and on solutions together. So even when we build something really, really great, even something like we've just seen before with the global AI, special AI, E.O. Wilson would say, oh, this is just another droplet in the sea of information. So to do, to make it work, we need to be working on it, on the tools together so that they can trigger action and impact. And this isn't just a theory for us. So you can see people here working in the real life together on something, and you can transfer this to the digital realm. So, People don't want the solution being served to them. They don't want to be told what to do. They want to have something they can argue with, they can push back on, or which they could fit to their needs and also constraints they have. And if I'm honest, I used to find that rather frustrating in the past because I was a researcher in my ivory tower before I came to Heiged. And we had robust research, we found great stuff, and people would say, oh yeah, that's great what you're doing, but no, we don't really need it. Yeah, fine, but, right? And I think now, since I came to Highgate, I really learned a lot, and I think this yes, but is really the starting point, not the end point. So What I learned is that the important question is not how to build the right tool, it's how do we build up know-how, know-how to act in a process that sees each partner as expert for their own living environment, because we are all experts for the environment we are living in. And people are capable, more than capable, to find the right solutions for them if, yeah, together. So I'm a researcher for nearly all my adult life, and I can honestly tell you that since I'm at Haigat and we are working together with partners in workshops, many, many workshops online and in person, and I've never learned as much as in these workshops when discussing with partners and with NGOs and learning from them much more than there's ever to be found in textbooks. So what we did at Hyget was building the so-called Climate Action Navigator. So far it has seven assessment tools, as you can see listed there. For example, the CO2 budget, heating emissions, traffic emissions, emissions from land use and land cover change, and land consumption, to name only a few. Our goal is to build usable and useful tools, and it's not necessarily our goal to have a dashboard. That's not our end point. It's the starting point. And wherever we have partners and the database for it, we try to be global with the tools, but you can see that we don't always achieve it yet. All of this is work in progress. It's not a finished answer. The dashboard is not what we want to deliver. The deliverable is the thing, the method behind it. And I want to spend the rest of my time explaining the method behind it. Why do we think it's not the dashboard which is important, but the method? Because we don't believe that dashboards or maps or data are the success measures for projects. They are the starting point to create impact and a project is successful when our data and information is being used and impact is being created. So our project will be successful if the decisions made based upon that are successful. So because what really counts is impact and not our output. Our output is so to say, somewhat irrelevant. So how do we put this into practice? How do we co-create? Most importantly, we start as equals. Partners are experts for their living environment, and they bring the question and they bring the local knowledge. What we then bring in is our methods, like geospatial AI, which Sukanya has just presented. The first steps then are ideas incubation workshops. We are using these to turn an open problem into a first prototype, and this prototype is a rough prototype, but it's a rough prototype on purpose because then we can work on that together. And people can still argue with the prototype and also with us. And then the iteration starts. Partners test the tool, they criticize it, we then reshape it, and most importantly, the tool is not finished when we say it's finished, it's finished when our partners say it's finished. So, and the process is the same for an NGO, for a municipality, a ministry, or a research team. So the method stays the same, but really what changes is the product every time. In several cases, we have seen that globally applicable tools means offering a so-called living product, so it has to be changeable still. For example, it is really highly context dependent on whether you think your city is walkable. Walkability is one of our tools. So, walkability in Lagos, in Nigeria, means something completely different from walkability in New Delhi, for example, in India, or walkability in Heidelberg, where we are working. So, context matters. And the solutions that are needed in the context will vary every time, or very often at least. And the same is true for land degradation. Data, but even more so the solutions, cannot be prescribed but need to be co-created. So I brought you one concrete example, this about land consumption. We co-created this with the WWF Austria, and they are working hard against land consumption rates, high land consumption rates in Austria. And doing this, they face several challenges. The first challenge is that they don't have a data set with which they could validate the official data set. And the second challenge is that it's very difficult to break down land consumption in different sectors. For example, land consumption in residential area or for residential areas, industrial areas, or for transportation. And third, the third problem is a major problem too, because every country measures land consumption in a different way. So there's no way to compare land consumption rates internationally. So WWF's main intention when cooperating with us was to explore new lines of argumentation and to explore also new data sources which they could use so that they have new perspectives and reasoning. And this is what we built with them and the result is still work in progress but you can see here the result for land consumption in Ulaanbaatar. So displaying the results as a so-called tree map on the right hand side is their idea. Breaking it down into different sectors is what they need. The categories is what they are interested in. And we didn't hand them over our analysis. We built it with them, workshop after workshop after workshop. And I think this is really important because this will make the tool actionable for them as soon as it is finished. So, but of course, I mean, we are a research institution. We also have our own ideas what might be interesting or what might pose a challenge for humanity or society. And one of them is assessing carbon emissions from land use and land cover change. And right now it's a Germany only prototype, as you can see here. The example is from Tesla factory building being built or that has been built near Berlin. And it needs a partner to be transferred to different contexts, to other contexts, to ask new questions or different questions. And maybe The partner for carbon flows from land use and land cover change is in this room right now and you are interested in this and we could scale this up to other areas. So according to the experiences we made with our project so far, there are a couple of key principles that are really enabling co-creation. One is openness, so open data, but also mental openness, and actually, which is not there, it's open and free, so nobody has to pay for that, neither for working with us nor for assessing or accessing the results. Transparency, that it's global in principle, or at least can be global in principle, that it's useful and usable. And as I mentioned before, that it is a living product. And I think with these key principles that are guiding our work, we hope or we think that our work will really make a difference. So if you're working on land degradation or on climate change mitigation or adaptation, We want to help getting the most out of your work. So the idea is that your work is not only providing information and data and is very insightful, but that it's always also creating impact and that people can act on that. We do not primarily want to produce a dashboard like the Climate Action Navigator. It's a great tool, but what we want is to help people generate impact, to help them to make a difference. Our contact details are on the slide. Me and my colleagues are here in the room and at the COP for the remainder of the week. So please feel free to contact us and also to ask questions now if we have time and if you have questions. Thank you very much. Oh yeah, you have a question. Oh, two questions. Maybe I'll take the second microphone. I would do ladies first if it's-- Oh. Oh.

How can youth participate in your idea and contribute to our land.

Okay, so sorry, once again.

How can youth people contribute to your plan in the future?

New people or young people?

Young people.

Okay, I will answer, come back to you.

Yes, thank you. So I listened to the presentation, especially Lila is what I'm interested in. So I'm from the Caribbean and I work with seeds basically. So our islands are basically 350 to 500 kilometers square. So what is the potential for the application of Lila within the Caribbean region? We know the issues we have, well we have with high resolution data. So we are just trying to develop our own resolution products in terms of land cover and land productivity. All right. Um, well, I represent PISLM. Um, we work with APACETA within the Caribbean. But what is the potential for LILA to be applied within a region such as ours working with SIDS?

Okay. Thank you. So I will come back to your question. So young people, um, can work with us like any other people, um, and can get in touch with us. I can give you my contact details too. And then, yeah, bring in your perspective and we can try to build something together which is especially useful maybe or applicable to, for young people or important for young people. Maybe later.

Okay, coming to the Lila tool for the Caribbean, I'm sure there's a lot of potential, but that has to be investigated using the different data layers that are available for that particular region. So we can definitely take this forward after this session and we can, you know, talk about it, connect and take this further. Thanks. Sorry.

Okay, sorry. Sorry for interrupting the interesting questions. But now we will continue with a panel discussion with our speakers and two more that will bring interesting perspectives to the discussion. And after a brief set of questions that I have prepared for them, I will open the floor for questions from everyone. Is that okay? So in addition to Kirsten and Sukanya, who you already know, we have here Dr. Bulgama. Densambu, the program director of Rangeland Research and Development at Green Gold, the Mongolian Rangeland Research Center, who was our partner from the beginning of the project and guided us to make our results relevant for the country. In addition, we have Debasree Miraula, who is an environmental scientist and an expert on remote sensing and GIS, and that is also a researcher at the G20 Global Land Initiative, who also was involved from the beginning of the project and brought us here. So thank you so much for participating in our panel discussion. To begin, I want to ask something to Sukanya. You showed us how modern geo-AI techniques are flexible enough that they can be tailored to the specific situation and context of a country and a region which allows making the results relevant for the local situation. So what were the most interesting aspects of the geology and ecology of Mongolia that you took into account during this work?

Yeah, there were of course many different aspects which we learned. you know, through interaction with the local stakeholders with Bulgama who, you know, and also by reading some very interesting facts about Mongolia. I mean, one interesting story that I could share, Bulgama told us that, you know, it's not the problem that there's no rain, then there is water, and then it does rain, even though the precipitation seems very low as compared to global average, there are still flash floods that are happening in a lot of these grassland areas that last, you know, so there's a lot of water accumulation that happens for up to few and then the water completely disappears. So it's not that the water's not there, it is just that it's not harvested at the right time and it just disappears. So this is a very local, very sensitive ecosystem with very special demands and features. So we of course had to take this into account and It was very surprising. And also, I think just what I showed earlier that, you know, the greenness, how it changes so rapidly, you know, as response to rainfall over the years, and you cannot use that as a degradation metric. So every land, I think, has its own challenges, you know, and own demands. So you can look, you know, satellite imagery can show you the pixels, it can show you what a land looks like, but it, but it's, you cannot just use that you know, for making practical interventions or for restoration that makes sense for the local people. So, I mean, there are many such examples, but I just shared one of them.

Thank you, thank you. So we keep claiming that our results are locally relevant, but I want to ask you, Bulgama, how exactly does this type of geo-AI analysis can change what a range land manager actually does on the ground?

Yeah, thank you. So this is really very important that how to really move the data, existing data to the action. I am a person who was doing the research on the rangeland degradations. How is it degrading? Where is degrading? Where are the hotspots and what can we do? Whether is it recoverable or not. And then in same time also we are working with the local herders and local officials who are responsible for this whole range land management issues. And then they are working on the planning that how to rest, where to rest, and then where are the key, the problems, you know. And then you, as I see that your products, so this map is, uh, very relevant because even though it doesn't, uh, show that what is the degradation level and how is it going, it's more the short term. So it's showing that what is exactly is going on on this particular ecosystem at right now. And because, you know, the range land management planning is usually for one year and then usually it's planned the previous year, end of the previous year. And then sometime we are facing problems that we cannot implement that because for that year, depending on the precipitation, how it was the rainfall in the beginning, of the summer or in the middle of the summer, sometimes they cannot implement this rotational grazing management. So for that case, I find it is quite useful that it looks very green. And then also the map will show that where are the spots where we can accumulate the water and then use it for the certain time as a water point, you know. And then we can use this information then we can update our management plan, you know, because as we were mentioning earlier that the summer and fall grazing area is quite busy and it's having more degradation because that's the only open water where everybody is coming and settling there until the vegetation is sensing, you know, so that's why that. we can, we will have a more opportunity to direct the rotational grazing and resting through those short-term water points, right? And also using this information on this greenness. It looks green, but maybe we can't define that what plants are there, what is degradation, but we can see that there is the grazing opportunity. So this way we can, update the plan and make it more liable, the plan. Yeah.

Thank you very much. And Devashree, you are an expert on this field as well. So could you share a few examples of other situations in which remote sensing or AI analysis have actually changed decisions about land management on the ground? And also perhaps some of the factors that have limited that happening more often, like why? AI and modeling approaches not always are successful in shaping policy.

Thank you so much, colleagues. First of all, the very interesting case study that we saw of Mongolia, that is one of the examples that I would say not of implementation, but of basically showcasing what can be done. for everybody you heard that I do represent the UNCCD secretariat so I'll give some of the examples from the UNCCD secretariat in the room you'll see the G20 global land initiative that I represent right behind you the pavilion space and we also have IDRO which is our second pavilion space these are the two let's say institutes or two platforms that have showcased some of these examples I'll just give an example of the International Drought Resilience Observatory, where within their platform, you can go inside and also, please, colleagues, you can actually go inside and also look at their beam, which is very interesting, where they're showcasing how they have mapped and have case studies of different countries and their methodology. Super interesting. But what I think is... more interesting than that is the fact that they have a work stream on the forecasting that emphasizes preparedness, that emphasizes forecasting, that emphasizes risk as well. And also there is adaptation where they're looking at how countries can adapt. So that is one of the examples where you can take a direct link with, let's say, implementation in a country. In the G20 Global Land Initiative, this year we are in this COP itself on the 27th on the food security day, we are trying to launch the RAIS mapping tool. This is a mapping tool of UNCCD that will be launched. RAIS mapping tool is mapping opportunity for restoration for businesses, specifically where return in investment for businesses is high and mapping it in pixel level. It's challenging, but that's one step to first understand where economic activity is high, where yield might be high, understanding the soil type, understanding the water usage, understanding the community's level of coping capacity or adaptation in general, putting it together in pixel level and trying to understand or give a picture of where restoration would make more sense in terms of range land and cropland for food security. So this is another example. UNCCD in general, we work specifically with country parties and we work specifically with policy. So none of these initiatives that I told you about happens without country parties engagement or countries engagement, where these are the countries that come to us, request for let's say a tool or a platform or actually like how IDRO and G20 came about as well is countries coming together, discussing understanding the importance, and then bringing about these initiative and actionable items where technology, where remote sensing and GIS, understanding in a bird's eye lens how any of these environmental issues are was one of the major, let's say, issues are one of the major outcomes that came out of the decision that led to the growth of these initiatives. Something that was a question from the floor, which is engaging youth. And I would like to bring the engagement of youth towards what G20 Global Land Initiative is doing and potentially any other future work that we do together is youth are heavily engaged as citizen scientists, as ecopreneurs, as if you're in media and content creation, that is another offering that we have under G20 Global Land Initiative. We give trainings to a lot of youths in different sectors and in different aspects. So youth is the core as our communications head Wagaki always says, we are catering to Gen. Zs and we are catering to that generation to empower them in using technology, in using AI machine learning. So that is one of the, let's say, opportunity for young people. Thank you.

Thank you very much. So part of the power of AI is that it allows you to make assessments at scale. But Kirsten, you showed us and argued very well that it is essential to incorporate local knowledge throughout the entire process. So if it requires so much work to organize these workshops with local, finding the local partners and engage them, how can we scale the power of AI while at the same time retaining the local knowledge.

Yeah. Thank you. Yeah, that's a tricky question. I don't think that you need locals from every locality, from every municipality. But of course, for example, when we are working with the WWF in Austria, what they are telling us is applicable to Germany, is applicable for Austria. Mostly, maybe not so much for the US, maybe even less for African countries or Asian countries. So what you, what we need is, um, finding, um, typical partners, let's, um, call it like that, and then putting, The product out there, like the assessments we make, what's making clear it's a living product, so that we are always keeping this inviting character that if you find something there which is not applicable to your region, reach out to us and we try to make it so. Of course, it's not always super fast, so an AI will give you a quick answer and this is a bit more cumbersome, but the gain you get from that is that it's actionable, that you can work with it and that you know that's the problem, but you know where to tackle the problem and how to tackle the problem. And I think And also the feedback from our partners shows that that's, um, a big asset because there are a lot of dashboards out there and many of them are not used.

I would also just add to that is, um… you know, when we talk about scaling of AI, I think if it's done responsibly, like Kristin said, through co-creation process or by integration of the local knowledge, like what we have shown for this particular project, then it also increases the, you know, the confidence with which the people accept these results. And this is very much possible now with AI because AI technologies are getting more and more flexible and it is possible to, you know, integrate this local knowledge and scale at the same time. And so earlier people always had inhibitions of trusting a lot of, you know, data that was coming out. But because we make this flexible and integrative, it is -- I feel that like, especially like what we have done, the results that we have gotten now with expertise, local expertise, one, it is possible to reach there. So we -- you know, it's good to be open about this as well.

So those were all my questions. So now I open the floor for questions, if anyone else wants to ask someone. Oh, thank you, Kristin.

Thank you so much. Hi, everyone. My name is Garima, and I'm from Delhi. While I completely understand that scaling AI is, let's say, the future, my problem with AI is that AI is technically very biased, especially when we're talking about local knowledge. And when we want to scale local knowledge via AI, it simply means that you would have to employ people at data centers who are now inputting labeling knowledge, which we, let's say, call local knowledge. My another problem with that is most of these people who are usually employed to basically input these data, especially in India, let's say, are very vulnerable women who are not being paid enough, who are not being covered, let's say, their health insurance, there's nothing that they are being given. And I feel that while we're looking at scaling AI, because yes, it is important for future, I just don't see how it becomes sustainable in nature. And while that's not exactly what, let's say, CCD does or not exactly related to rangelands or pastoralists, but I think that deep down when we're looking at scaling AI, you will eventually have to work with these women who are already very vulnerable, and then scaling AI through them eventually impacts them further. So I'm just trying to find where do we find the balance here?

Yeah, that's a good question. So I think we have to think a little broad in this term. So also how we define scalable AI. So when I mean that AI is scalable, I don't mean just create one map with, as you said, a lot of data is very biased. So just if you take example of the land cover map, most of the data is coming from certain countries. And of course, you then see a bias and then you see that it completely misses the point when you try to generate the map for Mongolia or for other places in the world. But the idea is to use the flexibility. So if you develop a technology that is scalable, like what we did, so there's an innovative model out there that combines different data streams, it is scalable What we mean is that there is going to be that input where a local would define that these are the classes that are relevant for me in this area, but now it's an automated process and then you further generate the final results, but you have the AI that is capable of scaling there. Yeah, so there are ways to, you know, make it response, to do the scaling up in a more responsible manner. And that's where I also think open data and open methods play a very important role. And unfortunately, in the field of AI, a lot of people who even came with an intention of being open, like open AI itself, you know, they are not open anymore. So I think that way, we are very unique as an organization at Highgate where we are still promoting open data. And I think it's important to think in terms of these open models and make that available for people so that they could scale it responsibly for their own regions as well. But it is possible, but it's just that, yeah, one would have to consider all these options.

Yes, so my question is related to the question earlier.

So as you know, Mongolia, nature itself is borderless. between Kazakhstan and Mongolia, we share a region called Altai Mountain Range. And it's both our responsibilities to.

To restore and protect these areas.

So my question is, uh, for, is probably from the policy making perspective, right? So how do you see countries, um, scaling, scale and deploy, uh, geo AI innovations such as yours?

Uh, should we consider, uh, more of a coordinated regional corporation where funding and.

Resources can be pulled together and, and also data, right? Or should there be, um, a thousand of hyper local initiatives that really relies on the local knowledge?

Yeah, so actually just like what I mentioned before, I think if, uh, we have open models and open data, then everybody could benefit from it. So then the borders don't matter anymore. But of course, because we would like that there would be local knowledge that is integrated into these models as much as possible, of course, one can always get better and better with more local knowledge, but at least to make maps that are accurate enough to kind of, you know, to map the restoration interventions effectively. So for that, Yeah, I think it's the way you design your models that if it reflects the local realities, which is what was our intention with this project, we have shown that it is possible to do that. And then to make these models open for everyone to use so that, you know, this could be adopted or taken by different communities and people. So it is, yes, again, very much possible to do that.

I would add into that, um, I don't think it's an either or question, it's both, because, um, models, also AI models, are always a simplification of reality. And sometimes, especially with AI, sorry, at the moment, harsh oversimplification of reality. And especially what I call the living environment of people. AI, I don't see that AI will capture that in any near future. And I even think it's good. And so I think I wouldn't want to make up a contrast between AI and local knowledge and all this wisdom which is out there. I think if, you know, like Suthakanya said, it's open and free, not closed and expensive at the moment, as in the moment, then I would say you can make something out of it together.

In terms of policy towards taking it towards that direction, because I think they covered specifically the tech part quite well, I would be a bit of a devil's advocate in this context. I don't think we need to integrate AI in everything. I don't think AI is necessary in everything. As of now, it might not be necessary in everything. However, the point that you mentioned is to have these little different platforms within the country or let's say a robust, let's say, tool like Lila, like what is the best approach or how do we merge these together? One of the examples that I immediately thought of is I come from Nepal. Nepal has a very interesting national disaster risk reduction portal. This is nationally what they have. And within the portal, they have like little, let's say, sub web pages or let's say sub tools like Lila, where today if there's a disaster somewhere, police can go and look at the extent of loss and damage and report it. There is the ambulance that looks at how many people went to the hospital. There is another little, let's say, tool with hydro-meteorological station that tells you about the condition of the, or the weather condition at that time, or that moment, which is all pulled together in this disaster risk reduction portal. What I'm getting to at is, in a policy lens, things can be very different. And as a grassroot lens, things can be very different. Right now, we're talking about restoration. But the same data, same information, for example, your precipitation and temperature, these two precursors, or even soil type, these can be very differently used for a construction-related work, even for policymakers. But these are the precursors and indicators that they use. However, this data is already provided by the hydro and meteorological station. I think these little portals or little sites are very important. They're important. However, let's say tools like Lila is also very interesting because they give a very good insight or very good understanding externally, especially because these are scientists to then use some of the methodologies. I know specific countries, like for example, Mongolia itself. with consultation with other platforms, other tools that we have, they're very open in using any other tool that the world has to look at, let's say, best practices and see how they can integrate, how they can improve. But I know that there are countries that are a bit more, let's say, conservative about the type of tools that they let in their country. And it's usually a common practice between what a country is specifically looking for and asking for. like let's say set rule in terms of policy making. However, if we are just to use a blanketed approach and a international, let's say, I'm not calling LILA an international platform, but I'm just saying if we are just to use, let's say, a platform that is made somewhere in US, but it is for Mongolia, it doesn't make sense. For me personally, it doesn't make sense. It would have rather made sense if, let's say, it come from local knowledge and local understanding, and perhaps with consultation as what they did with our local partners, having a local consultation, local understanding, maybe even come and do ground testing, validate their data set. I would rather prefer that than or maybe even developed in Mongolia by itself. Personally, I would rather prefer that, and this is also what UNCCD pushes, is to push what is locally available and but to also integrate some of these scientists and their methodologies together. Thank you. Do you have any input?