- An introduction to Matt Gray, co-founder and CEO of TransitionZero.
- What TransitionZero does and why it is important for reducing emissions.
- The role of machine learning in data analysis.
- TransitionZero and the Climate Trace organization.
- How TransitionZero validates models.
- Why domain expertise is indispensable.
- How the approach differs between different facilities.
- Measuring impact in accordance with TransitionZero’s mission.
- Examples of the impact the organization has had in China and Japan.
- The challenge of finding the right people to join the team.
- Policies that enhance non-salary benefits to the team.
- His advice not to lead with AI.
- How TransitionZero is approaching the Future Energy Outlook Project.
- His hope for the future of TransitionZero’s impact.
“Another application we are just embarking on is using data science to estimate the productivity of wind and solar assets.” — Matt Gray
[INTRODUCTION]
[00:00:03] HC: Welcome to Impact AI, brought to you by Pixel Scientia Labs. I’m your host, Heather Couture. On this podcast, I interview innovators and entrepreneurs about building a mission-driven, machine learning-powered company. If you like what you hear, please subscribe to my newsletter to be notified about new episodes. Plus, follow the latest research in computer vision for people in planetary health. You can sign up at pixelscientia.com/newsletter.
[EPISODE]
[00:00:33] HC: Today, I’m joined by guests, Matt Gray, co-founder and CEO of TransitionZero, to talk about the transition to a zero-carbon economy. Matt, welcome to the show.
[00:00:42] MG: Thanks for having me, Heather. It’s great to be here.
[00:00:45] HC: One of the projects that TransitionZero works on and that we’ll touch on today began as a collaboration with one of my clients, Watttime. I’ve known a few members of Matt’s team for several years now. But TransitionZero has a lot of other work going on, so I’m excited to learn more about that today.
[00:01:00] MG: Happy to share it.
[00:01:01] HC: Matt, to start, could you share a bit about your background and how that led you to create TransitionZero?
[00:01:06] MG: Yeah, for sure. My background isn’t technical per se. I started my career in investment banking. I was a carbon trading analyst and forming trading decisions within the carbon market in Europe. Did that for a number of years, and then got old into oil and gas analysis, and researching which oil and gas companies to buy and sell, very typical investment banking work. I sort of became a bit demoralized by, I think, serving fossil fuel companies, and empowering them through our research and analysis. I sort of decided to resign from that line of work, and start working for not-for-profits.
Within the not-for-profit space, I joined an organization called the Carbon Tracker Initiative, which looks at the risks associated with continuing to invest in fossil fuel companies. What I was doing at Carbon Tracker was building out their power utility team. Doing analysis on the risks associated with continuing to invest and operate coal-fired power plants throughout the world.
How the story came about, both founding TransitionZero, but also my interactions with what Simon and you hear there was. We were building out economic models, and these asset-level financial models were looked at the operating costs and the financial viability of coal-fired power plants throughout the world. We started building our models in Europe and America. There was lots of data around the productivity of these assets and the emissions that they release.
But what we found is, once we tried to build our models in Asia, and in particular China, the data was very sparse, and the data around the productivity of the asset, and the emissions that there were producing was just not available asset level. This is what led us to exploring the use of satellite imagery to estimate the productivity and emissions from these assets through the use of satellite imagery and machine learning.
We did a study a Carbon Tracker called Nowhere to Hide, estimating the productivity and emissions of coal-fired power plants using satellite imagery and some basic ML techniques, release that report. It was really well received, ended up collaborating with what time on a Google AI Impact Challenge grant, which we were lucky enough to win. That was sort of the catalyst for starting TransitionZero. The team that Carbon Tracker was growing and sort of warranted a new organization. So yes, two years later, here I am with TransitionZero.
[00:03:51] HC: It’s been an exciting path. What does TransitionZero do and why is this important for reducing emissions?
[00:03:58] MG: TransitionZero is a climate data analytics, not-for-profit. We were co-founded in 2020. We’re based in London, and entirely grant-funded by philanthropic institutions, such as Bloomberg Philanthropies, Google.org, the Quadrature Climate Foundation, amongst others. What we do is we provide Open Data Tools, modeling, and analysis to support the phase-out of fossil fuels, and power, and heat industry, and the transition to cleaner sources of energy, such as wind, and solar, and battery storage, et cetera.
How we do that is we provide what we call investment-grade data to our partners working on the ground in emerging markets and in particular, Asian emerging markets such as China, India, and Southeast Asia. They take that data to help support policymakers, investors, and corporates to make decisions to avoid the risks associated with losing money as we transition, and embrace, and take advantage of the opportunities. Both economic and financial opportunities as we embrace these new technologies that I just mentioned.
[00:05:11] HC: What role does machine learning play in creating this analysis?
[00:05:14] MG: For us, machine learning plays a host of roles. There’s the tracking role that machine learning plays with regards to estimating the productivity and the emissions of fossil fuel power plants. And indeed, heavy industry facilities as well, such as cement and steel plants. That’s a big part of what we do through the Climate TRACE initiative, which was co-founded by Watttime, TransitionZero, and a number of other organizations.
What we do there is we use satellite imagery, both publicly available in commercial imagery. Depending on the application or the sector, we take signals, whether they be visual signals or heat signals from the imagery, and we apply ML techniques to estimate the productivity of the assets. Then, we overlay that with emissions factors based on best available data. That allows us to predict the emissions from those assets. That data is then fed into economic and financial models, which allow us to tell investors and policymakers how those assets are going to perform as we transition away from fossil fuels towards cleaner sources of energy.
The other application that we’re currently working on is identifying specific assets in the electricity grid. Within the electricity grid, there’s the generation assets themselves, but those generation assets are all connected by distribution, and transmission lines and things called substations, which change the frequency of electricity. Assets or stuff like transmission lines, distribution lines, substations, these are all assets that we can pick up from satellite imagery to identify, and size, or get an estimate of the size. That’s another application that we’re exploring. Another application that we’re just embarking on as well was using data science to estimate the productivity of wind and solar assets as well.
[00:07:22] HC: It sounds like remote sensing imagery is one of the key inputs to these models. But they also require ground truth in order to train and validate them. For the emissions, projects were for power plants, and heavy industry, where you’re trying to estimate what each plant is producing, ground truth generation and emission state is important. For the latter piece, what forms of data do you rely on and how do you obtain it? How do you ensure your models are properly validated?
[00:07:52] MG: Yes. I guess just taking the example of fossil fuel facilities, so coal and gas, electric power plants, and steel and cement facilities. What we do is typically, there is good data availability in some markets. For coal and gas power, for example, there’s the European emissions trading scheme, which provides good data on emissions. That’s a way of training our models, and there’s also the grid operators in markets, such as Europe and in North America that provide generation data, a reasonable level of accuracy, and time-frequency. That allows us to train the models and then release those models in markets, such as China and other parts of Asia, where this data typically isn’t shared in the public arena.
That said, I think we are acutely aware, I think of the need to validate our estimates with local partners. At TransitionZero, we have what we call in-country market analysts who are on the ground and some of these jurisdictions that we are modeling. They help validate the results, help engage with policy officials, and investors, and some of the operators of these assets just to make sure that we are since checking our models to ensure that they are as accurate as possible. Because I think one thing that we’ve learned over the last two to three years is domain expertise is indispensable when you’re building these models.
You can’t underestimate how important it is to have the experts facilitating the development of these models just to make sure they do reflect what is happening on the ground.
[00:09:35] HC: I know you’re producing estimates for both power plants, and heavy industry, and possibly other types of industrial things. How do your approaches for these things? How are they similar and how did they differ depending on what it is you’re trying to – what type of facility you’re trying to estimate emissions from?
[00:09:55] MG: Yes. The similarities I think is in the process. Typically, you need training data. As I mentioned, we typically get our training data from either Europe or North America, because there is a history of being more liberal with data availability, a number of key regulations. Both in Europe and the US mandates that operators make data available. That allows us to train the models, and that is something that we do, of course, across both electricity and heavy industry. I think the differences across power, and heavy industry for power, we’re typically relying on visual signals. We take optical imagery from commercial data providers like planet. Because it’s relatively high resolution and very timely, we can detect a signal from the plume of steam coming out of the facility and correlate that with the extent that is being utilized.
With regards to heavy industry assets, such as cement and steel, we typically rely on heat signals. If we’re using thermal imagery to correlate, or use it as a proxy for understanding production, and therefore emissions. For a steel facility, for example, they emit a lot of heat, because the process to create steel was very energy intensive. As a consequence, there’s a trend tremendous amount of heat being generated at these sites. We can pick that up with publicly available thermal imagery. The size of the plume or the size of heat signal is a relatively good proxy for understanding to what extent the facility is being used. That is what we then use to estimate production.
[00:11:59] HC: It sounds like domain knowledge is critical. Basically, understanding how power plants are different than heavy industry. They produce different signals that can be seen from different types of satellites. It sounds like that’s key to success for these projects.
[00:12:14] MG: Yes. I mean, absolutely. I think within electricity, I mean, comparing electricity to say, steel. I would say coal-fired power plants are much more homogeneous than steel facilities. But even within coal-fired power plants, there’s all sorts of nuances around how they call the facility, whether there’s control technologies, the age, which of course affects the efficiency, and therefore the productivity of the asset. Also, the type of coal that is being fed into these facilities. There’s all these factors, which yeah, inherently require a lot of domain expertise. But I think the more you move into those heavy industry sub-sectors, it does become a lot more nuanced, and these differences do start to emerge much more than within electricity.
[00:13:08] HC: You mentioned Climate TRACE previously, which is the coalition of organizations that are trying to track greenhouse gas emissions across different sectors of the economy, and doing so globally. As an organization, climate trace does strive for global coverage, but reducing emissions in some parts of the world will be more impactful than others. How do you decide where to focus your efforts?
[00:13:32] MG: Yes. I mean, I guess just to take a step back and explain Climate TRACE. So each member organization focuses on a different sector. There are numerous organizations who are part of Climate TRACE, and that coalition of organizations is growing rapidly as people become more interested. We find out different organizations have data for different sectors. TransitionZero focuses on power and heavy industry.
In terms of how we think about focusing our attention, I think, broadly, this is something that everyone, all organizations who are part of TRACE think about the problem of where to focus. We try and focus on A, those countries that are omitting the most, and B, those countries that don’t have a history of data availability, either for cultural or resource reasons. For example, in China and parts of Southeast Asia, there is either a cultural, I guess, a cultural preference not to share data, or for some of the smaller Southeast Asian countries. Particularly in Africa and Sub-Saharan Africa, there often isn’t the resources to make emissions inventories available.
That’s how we think about how we spend our time and what we focus on. We typically like to spend most of our time monitoring and trying to understand what’s happening in Asia, because it’s a huge matter, and is said to become quite an important part of the global economy over the next sort of 20 to 30 years. Within that, I think there’s I think, a broad reluctance to share data, because historically, that’s never happened. But also, in a lot of instances, there is just not the resources and the willingness to dedicate resources to monitoring emissions. Because historically, externalities haven’t been priced, and capital allocation decisions haven’t been made based on emissions.
But of course, that is all changing, and now there’s a growing awareness, particularly from international investors that these assets do need to be progressively phased out or converted into different technologies that emits progressively less CO2 in the atmosphere.
[00:15:49] HC: Thinking more broadly about TransitionZero and your goals, how do you measure the impact of your models and analysis to be sure you’re accomplishing what you set out to accomplish?
[00:16:01] MG: Yeah, it’s a really good question. We’re currently a not-for-profit organization. That means we measure our success based on the impact. Our mission is for affordable and dependable energy for everyone. What we really mean by that is we want zero-carbon sources of electricity, which we believe will be the cheapest forms of electricity in the not-too-distant future, if it is not today. We want everyone to have access to that type of electricity to benefit from, I think the benefits associated with having energy and something that the global north takes for granted and as headaches is dubious for some time.
In terms of how we measure success on a practical level, I think it really comes down to the policy implications. Alongside the models and data that we produce, we regularly release reports, and blogs with the aim of shifting narratives to highlight the cost advantages of transitioning to clean energy, and supporting policymakers, and decision-makers with the data and tools that they need to make quite technical and complex decisions around how to phase out fossil fuel assets, and how to create an enabling policy environment for clean technologies.
One practical example was the first report that we released in 2021 was on China’s coal production or coal use within the electricity sector, and whether they were on track to meet the needs or privilege, which they made a few months earlier to be net zero by 2060. That report received widespread media attention and was a report that was picked up and used by John Kerry, who was in China at the time with his climate envoy to talk to policymakers and political leaders in China, about China’s policy and investment decisions around coal. From the feedback we got from Al Gore, who was the person who handed the report to John Kerry, we are of the understanding that that report was influential with regards to some decisions that the Chinese government made around reducing the number of coal-fired power plants that they were planning to build over the coming years.
Another example was last year, we produced a company profile for a steel company called JFE, is a Japanese steel company. That company profile was based on the data that I mentioned. What it did was it showed that JFE’s climate policy was inconsistent with the goals from the Paris Agreement. Meaning, based on what they were planning to do in their company reports, they were not going to be net zero by 2050. What we did is we worked with some activist investors who negotiated or had an engagement with management at JFE Steel. That resulted in the company changing its policy to replace an aging high-emissions blast furnace with a lower-emissions electric arc furnace. That is something that we’re really, really proud of, and I think shows how data can be used constructively to engage with companies and getting them to reduce their emissions.
[00:19:22] HC: Your data science and machine learning team has been growing pretty steadily. Even though hiring for this domain is pretty challenging. Did it do the demand for professionals in the field? What approaches to recruiting and onboarding have been most successful for your team?
[00:19:38] MG: Yeah, it’s tremendously challenging. I would say it leaves a very specialist roles and skills. Currently, these people are in high demand. I think, particularly challenging for not-for-profits given it’s difficult to compete with Wall Street and Big Tech. I think how we compensate for that is we really lean on the impact side of what we’re trying to do, and try and emphasize that the purpose and impact that they will have by developing the data that we are working on. Instead of helping big tick with AD optimization, they’ll be helping governments and policymakers in some of the poorest parts of the world, really to reduce their emissions and to give them a better understanding of what they need to do, and how they can reduce their emissions.
I think that is something that does appeal to a lot of people who increasingly want to work on something a bit more impactful. But beyond that, we try and offer a good working environment with benefits such as enhanced gender-neutral parental leave, flexible working, employee assistance, et cetera. We recently introduced a policy called Neptune Days, where the tech team can spend 10% of their time on something innovative and new that they want to work on. It doesn’t have to be part of their day-to-day job. We try and bring as many, I guess, non-salary benefits as we can, given the limitations associated with being a not-for-profit.
[00:21:18] HC: Is there any advice you could offer to other leaders of AI powered organizations?
[00:21:23] MG: Generally, or on recruiting, per se.
[00:21:26] HC: More generally.
[00:21:28] MG: I think generally, my advice, which is probably quite generic is, don’t lead with AI. I think lead with the use case and the problem that you’re solving. I think there’s tremendous stuff happening with AI at the moment. But the one thing I noticed when we launched TransitionZero, a couple of years ago, as there was a lot of hype around ML and the role that ML could play for various types of applications. But I think really to be successful, you need to, I think, understand what your use case is, who the problem owner is, and how you’re going to solve the problem.
If AI is part of that, then great, I think for us, AI was just one piece of the puzzle and it needed to be complemented with physics-based models, with traditional economic and financial models. Because ultimately, for us and the stakeholders that we’re trying to influence, they are particularly motivated by the financial and economic implications of the energy transition. AI and an ML on its own isn’t going to give them that sort of insight they need to change policy and change capital allocation decisions.
[00:22:47] HC: Finally, where do you see the impact of TransitionZero in three to five years?
[00:22:53] MG: Wow! I think one thing that we’re particularly excited about is, we’re currently developing an open-source energy systems model, called Future Energy Outlook, which is a very ambitious project. The ambition of Future Energy Outlook is to eventually have global coverage and provide net zero-line scenarios for all countries in the world, as well as highly granular asset-level data insight. It will be an energy systems model, and it will cover roughly about 160 countries. Within those countries, it will provide policymakers and investors in an understanding of how the energy mix needs to change over time for that country to be indeed zero aligned.
We hope that that will inform policymaking decisions for resource planning within the electricity sector. But also, for a number of these countries in the Global South who desperately need climate finance. We hope that everyone forms decisions around finance and getting the money that has been promised by the global north into these countries, so they can build renewable energy, so they can build the infrastructure that they need to help fulfill our mission, which is carbon-free, or clean energy that’s affordable and dependable for everyone.
[00:24:14] HC: I look forward to following these developments. This has been great. Matt, your team at TransitionZero is doing some really interesting work for reducing greenhouse gas emissions. I expect that the insights you’ve shared will be valuable to other AI organizations. Where can people find out more about you online?
[00:24:31] MG: We are online at transitionzero.org. We’re on Twitter, same handle, TransitionZero. We’re on LinkedIn. We have a GitHub repository, which you can also check out. In general, we’re posting webinars all of the time, so feel free to sign up to our newsletter to find out when our data and modeling is available, or join one of our webinars which we have when publish results of the analysis that we do.
[00:24:59] HC: Perfect. I’ll link to all of that in the show notes. Thanks for joining me today.
[00:25:05] MG: Thanks, Heather. It’s been great to chat.
[00:25:06] HC: All right, everyone. Thanks for listening. I’m Heather Couture, and I hope you’ll join me again next time for Impact AI.
[OUTRO]
[00:25:16] HC: Thank you for listening to Impact AI. If you enjoyed this episode, please subscribe, and share with a friend. If you’d like to learn more about computer vision applications for people in planetary health, you can sign up for my newsletter at pixelscientia.com/newsletter.
[END]
LinkedIn – Connect with Heather.
Computer Vision Insights Newsletter – A biweekly newsletter to help bring the latest machine learning and computer vision research to applications in people and planetary health.
Computer Vision Strategy Session – Not sure how to advance your computer vision project? Get unstuck with a clear set of next steps. Schedule a 1 hour strategy session now to advance your project.
Foundation Model Assessment – Foundation models are popping up everywhere – do you need one for your proprietary image dataset? Get a clear perspective on whether you can benefit from a domain-specific foundation model.