 So, next up, I'm pleased to introduce Gavin McCormick, who is the co-founder and executive director at Watt Time, the non-profit tech company that first developed Automated Emissions Reduction Technology. Gavin's also one of the 10 founding members of Climate Trace, a global coalition of tech companies, NGOs and universities working together to combine satellite imagery, artificial intelligence and to make global greenhouse gas emissions transparent. And Gavin's going to talk today about tracking the whole world's carbon emissions with satellites, machine learning and data fusion. So a warm welcome from the open group, Gavin McCormick. Over to you, sir. Thank you so much, Steve. As you mentioned, I have an interesting background here, wearing two hats both at Watt Time and Climate Trace. It might be helpful to mention that I have a slightly different background as a result of this. I don't come to this work from the world of carbon accounting and transparency. I come to it from the work of carbon optimization. And I thought that maybe starting with sort of two slides on how is that world related to this world and how is that a similar competition or a different one might add a lot of context for the work we do at Climate Trace. So, I work on my day job at the audition Watt Time where I'm executive director. And what we really focus on is the fact that although a lot of corporate reporting has focused on annual emissions from power plants. In reality, power grids change their emissions factors every 5 minutes. So, although it would be a nightmare to try to add up the electricity that every organization used every 5 minutes and use an emissions factor for that. In the real physical world, this is actually how power plants dispatch at certain times of day. You will have more carbon intensive results if you actually use electricity. And that's relevant because what we do at Watt Time is we scan power grids for electricity every 5 minutes. So that we can do things like shift electric vehicles and smart thermostats and batteries to consume energy at times that have lower emissions factors. I tell you this because this is a very different ecosystem of carbon expertise that I think in the coming years is going to start intersecting a lot more with the world of carbon accounting. And as we are figuring out how do these 2 worlds come together, the intersection really started with the fact that there are very few power grids in the world where a person can in practice get their hands on 5 minute granularity emissions factors. So we know that this is how power grids actually operate, but good luck in most places in the world getting high quality annual emissions factors for every power plant, let alone the 5 minute level. And so organizations like mine that spend a lot of time, our environmental NGOs, they're very tech savvy, but are thinking about carbon in a totally different way. We started thinking, what would it look like for us to be able to expand our own capabilities to be able to dispatch smart devices at cleaner times worldwide, not just in the few places where these data already exist. And we started thinking in practice about what would it look like to use satellites and artificial intelligence, particularly computer vision to actually reproduce from space and emissions monitoring system for every power plant and other form of emissions that would go around the need to have an emission sensor in every stack and have that be available from the government. So, if you look at this world, climate trace is a group of NGOs, universities and companies who wanted to have access to really high quality emissions factors and activity data for the whole world. And so we got into that without realizing how that might actually be quite useful to the world of carbon accounting. Now what you're going to be able to see from space is never going to be quite as high quality as having a CO2 emissions sensor on every ship, every car, every factory. But the fact that these systems are inherently global might make them of interest to a lot of folks in the audience today. So, I think it's helpful to sort of start with a physics based view as I'm saying mass is conserved. A very different view of emissions than a typical carbon accounting world would say satellites for many years have been able to see in the atmosphere, how much emissions are there. So, for decades, I mean, this is how we know climate change is happening. We've been able to monitor that increase. What's new and different is talking about not what is the level of emissions in the atmosphere, which is sort of really straightforward to see how a satellite could see that. But rather, how did it get there? That of course is a very different problem that you have to solve if you want to actually know the emissions of individual activities. So, if you wanted to tackle that kind of problem, you might want to start with a photo like this. This is a snapshot of a power plant in the United States that has a CO2 emission sensor in its stack, open public data telling us the hourly emissions. So, we can see from space, this is what that power plant looks like when it is emitting a little bit over 3,900 tons of CO2 per hour. But we can also see this is what that power plant looks like when it's emitting zero tons of CO2 per hour. Obviously, the unedited human eye can actually see the difference. And so, what my organization and a number of other nonprofits hit upon is, could we train a computer vision algorithm to take hundreds of thousands of photos like this all over the world, learn to recognize where are the smokestacks in different power plants, and learn to recognize what does a photo look like of a smokestack when emissions are happening at a certain level. So, you can actually begin to scan the entire world's power plants. And so, again, this is not quite as accurate as having a smokestack on site, but the advantage is that it is available for all power plants worldwide. And so, with support from the Google AI impact challenge, my organization and a number of others was able to develop algorithms that are now able to scan power plants worldwide without self reporting and are beginning to make that data available to the whole world. The reason that we are doing this is to try to get the maximum accuracy and granularity. So, every extra percentage point of accuracy is very valuable to us. And so, we're always playing around with other methods. So, for example, this is a thermal infrared view of the same power plant just like in the movies, we can look for heat instead of CO2 directly or steam. The reason that matters is that that means we can use even more satellites in the sky and even more sensor techniques to cross-revenge and get a higher accuracy algorithm. We can also do things like look for other co-pollutants such as NOx pollution that is downwind of the power plants an hour later, which allows us just to cross verify another way. We can even do slightly obscure things you might not see coming. For example, a lot of power plants cool by sucking in large amounts of water from the nearby river lake. You can actually see ripples in the water from space, and that turns out to be a really good tool in training an AI algorithm to recognize what does a power plant look like when it's polluting. So, we do is we combine all of these different estimates into ensemble AI model that is able to provide an estimate trained on those power plants that do have high quality emission sensor data for all the power plants in the world and make that available to the public. The original version of the project, this was the only idea, but we were then approached by Al Gore and Generation Investment Management Fund, who said, would it be possible to apply this to all forms of emissions worldwide? And we thought that is way too big a project for our small number of NGOs. But what we realized is that if we continued partnering with more and more tech savvy nonprofits, universities and companies, it actually was possible to collectively measure essentially every form of emissions produced by humans. So another example of that would be transition zero based in the UK, which is applying a different class of algorithms trained to recognize pollution from steel mills. And so they are able to classify different industrial processes. Here we're again using thermal infrared because the heat signature of a steel mill is quite different than the heat signature of a power plant. But they've been able to train a different class of algorithms to recognize what does emissions look like from a steel mill. We can then combine that into a single database so that anybody interested in third party objective estimates of the emissions very steel mill or power plant in the world has access to that through the climate trace coalition. But we have a number of other ones as well. So for example, a group called synthetic is using computer vision algorithms to recognize what does a factory farm look like Rocky Mountain Institute is detecting oil and gas field emissions. Blue sky analytics is looking at fires from crops and forests. Johns Hopkins University is looking at cars and transportation emissions. So all of us and many more are looking at physical properties of emissions from space that are objective and third party and verified using multiple different techniques and sharing them in this open climate trace data set. I'd like to emphasize again how unusual it is that this is a truly global project. My experience in emissions factors is that the vast majority of projects have high quality data in a few countries. And these techniques are fundamentally global. So you're looking here for example at our partner ocean minds model of all the emissions from all large ocean going vessels wherever they are anywhere in the world. This one is based more on radar. And so these techniques are possible because they're a growing number of public or commercially available satellites orbiting the earth. We're using things like NASA satellites, European Space Agency satellites, Chinese Space Agency satellites and commercially available satellites. We're combining them all into a single tool. So this is all possible of course because the rise of big data in AI so I don't need to explain to everybody here that we have a large number of data scientists involved in the project. There's any data scientists listening to this we are always hiring and would love to talk to you. But what we are really able to do is to combine the fact that many different organizations were already building tech stacks to solve different piece of this problem. And we're basically stitching together these micro services to build a system that would not be possible for any one organization. An example I really like is that Ocean Mind already had AI algorithms to track the locations of all ocean going vessels to look for signs of piracy and illegal fishing to take that tech stack and modify it slightly to detect emissions factors. It was very small project for a coalition that would have been far too hard for any one organization to do for every single form of emissions in the world. That's why it's fundamentally a coalition. And so we are making this all available to the public at climate trace.org. It's live now. And so as an example, you can go in there and look for any particular type of sectors. You can go look at the rice methane emissions from Malaysia or emissions from coal plants in Australia. And we were rushing to get this out the door before COP26 so that this is available for the international community at the national level. But as our algorithms improve, first of all, you can see now the different estimates we have for the emissions of different countries by sector. But where this is really going is towards emissions of individual facilities. So there is a difference in things like scope three accounting in particular between being able to see all the emissions of all the physical real world assets and have a sense of those we really live in the world of scope one emissions. There's a big difference between that and knowing your corporate carbon footprint. So this isn't a one stop shop to solve all carbon accounting needs by any means. But we do think it's a really helpful way to verify if you have a supplier that is not giving you emissions information, or if you want to confirm and verify that your emissions estimates are about right. Or if you are interested in getting global emissions estimates, or if like the original climate race co founders, you're really trying to push the bar on increasingly sophisticated emissions data for carbon optimization, putting aside any questions of reporting and partial ownership. All of these tools we hope climate rates will be really available for. Finally, we are looking at more and more companies that are reporting their carbon emissions, but we're also looking at a large number that are not meaning to do this. So we think that for many investors and others in the community, having at least pretty good estimates of the emissions of every company that doesn't do any emissions reporting will help move the world faster towards the kind of transparency that we really need to solve the climate crisis. Thanks so much for listening to me and if there are any questions I'd be happy to share. Gavin, thank you very much. Thank you very, very much for that. That's fascinating to know that that's going on and it's there for all of us to see. So I appreciate that. Thank you. Question. Can you say a little bit more about what is carbon optimization? Yeah, so we at what time and other NGOs are working increasing on the idea that you can reduce a lot of emissions by using carbon smarter. So I like to think that as a lot of us are familiar with the concept of energy efficiency. It's not just how many miles do I want to drive? Can I drive there in a car that uses less energy? We haven't seen much beyond theoretical exercises in emissions efficiency until recently, but as the quality of data is improving, we're starting to see this as a very practical solution. So there are estimates that you can reduce the emissions impact of an electric vehicle, for example, easily by 20% by charging it at better times. You can double the emissions benefit of a solar panel by citing it in a location where it replaces particularly dirty fossil fuels. Just to give you one other example, slowing down a ship by one nautical mile an hour in certain cases can heavily reduce its emissions footprint because the emissions rise with the cube of the speed. So these kind of strategies we think of as no regrets ways to reduce carbon. Wonderful. Excellent. Thank you. Thank you. And we had a earlier on in the day we had a reference to the use of AI and how it requires, obviously one of the things it requires is a lot of training for it to be to be effective. And also the speaker stressed the importance of sustainability and these kind of emissions issues being an integral part of any architecting process for doing things better. Can you say something about about how those clearly heavy heavy use of AI in the in the project you're involved in. How do you actually apply the sustainability values you have into into the use of that AI. One thing we do one thing we are not doing yet that we are planning on doing what our own electricity load becomes worth justifying it is actually timing our own optimization of electricity based on our own data. So increasingly we're seeing that there are certain grids at certain times around the world where consuming a large quantity electricity run complex training algorithms will cause an extra windmill to operate and others where it be an extra coal plant. So 1 thing we're actually looking is the carbon footprint of our own code. Another technique that we haven't ourselves been doing, but we're very interested in is the carbon footprint of the space ecosystem itself. So we're very interested in some techniques that we have been hearing about to reduce the carbon footprint say of locking launching a rocket. By about 90% 1 technique is to launch them from a plane instead of the ground. So we're always thinking about how can we make sure that we're not adding to the government for ourselves. That's the main way we've been thinking about AI. Fascinating stuff really, really is and big thank you from all of us here for your presentation Gavin if you can stay where you are right now.