Can machine learning and artificial intelligence shore up integrity in the carbon markets?
Lawrence Xiao, co-founder of Nika.eco shares how his company is moving the carbon market forward into the digital age.
If there’s one good thing that came out of increased scrutiny of the voluntary carbon market, it’s the recognition that there are urgent gaps on the supply side that must be remedied. Specifically, the gaps in how REDD+ credits under Verified Carbon Standard (VCS) are issued which impacts about 25% of all carbon credits being sold worldwide.
Singapore-based startup Nika.eco appears to be in the right place at the right time to capture this particular opportunity. The company provides pre-project feasibility reports for carbon project developers and investors.
It also provides digital measurement reporting and verification (dMRV) services at two stages of the carbon credit issuance process: to establish the accuracy of the carbon claims prior to issuance, and continued monitoring of the project after the carbon credits have been issued.
What sets the company apart from the competition is the use of machine learning and artificial intelligence (AI) to crunch forestry data from a wide variety of sources ranging from satellite data from space agencies, historical deforestation data updated by non-profit organisations, and jurisdiction-specific databases of local forests.
Fancy stuff.
Is this the way forward for REDD+ credits? At Verra, which maintains the VCS, that certainly seems to be the case.
Verra published the VM0048 methodology last year, an updated methodology for REDD+ carbon credits that aims to set jurisdiction-specific baselines as a ‘single source of truth’ for all projects within the location. This will prevent project developers from overstating their environmental claims, or over-issuing carbon credits.
The data set for these baselines will be sourced from a single data service provider (i.e. companies such as Nika.eco) and incorporate technologies such as satellites and remote sensing, as well as ground-truthing. Nika.eco has written an explainer about it here, and you can read another explainer from the Verra website here.
This post showcases the highlights from the AMA with Lawrence Xiao, co-founder of Nika.eco, at the ReFi Singapore meetup on 19 April 2024. During the meetup, Xiao gave us a sneak peak into PlanetGPT, an AI platform for forestry projects that the company is building in stealth mode.
The audio recording of the session is available on Spotify and Apple Podcasts.
Han: Hi Lawrence, thanks for coming. This is such a new field that requires a very multidisciplinary approach. You need a data scientist thinking like a forestry expert. How did you manage to get into this field in the first place?
Xiao: I started on the software side. I graduated with a computer science degree from the National University of Singapore, then I worked at Amazon Web Services, the cloud company. I was mainly helping the Singapore government to move their data to the cloud. It was then that I met my co-founder Johann, who’s not here today.
Han: He’s in the jungle in the Philippines now.
Xiao: Yeah, he’s going through forests with the government. But it’s good. We had many conversations about starting companies because we both wanted to build things on our own, rather than making rich people richer.
Our first attempt failed. But because it was a sustainability startup, we got to know people in the carbon space. They told us to do something in the space because Singapore is trying to be a carbon hub. They also told us that there is a big gap in terms of the quality, integrity, and also the transparency of the market.
That’s when we started looking at using technology to solve those problems. The first thing we did was to learn about the different GIS (Geographic Information System) software. That does spatial mapping. Then we tried to figure out the differentiation rate, basically to see whether we could track whether certain trees were still there over the years.
We learnt a lot through that process.
Han: You started the company in 2022, and back then, I don’t think integrity in the supply side of things was part of mainstream conversation. It was only when The Guardian published its story about how 90% of REDD+ credits were worthless that shit hit the fan. So how did you manage to find investor interest in 2022 given that Verra still hadn’t woken up to the idea that their system was broken?
Xiao: We (and the rest of the industry) have known that the system was broken for quite a while. But at the time, our tech skills were still limited. We could only do a very minor part of the entire value chain; only the pre-feasibility portion which is a lot easier. That was our starting point.
But through that, we found out how to solve the problem of calculating additionality using historical satellite data. So if it’s a reforestation project, we could see if the developer cleared a forest in order to replant the trees. And if it’s an avoided deforestation project, we could see clearly if there was a deforestation threat or not.
Then we moved beyond pre-feasibility towards the whole spectrum of dMRV when Verra shifted to the new wave of using satellite data imagery. By then we already had more experience compared to other companies.
Han: Could you tell me about the initial conversations you had once The Guardian article was published?
Xiao: I think it’s bad for us all. Because imagine if the whole market crashed and nobody wasted to finance carbon projects. If all the developers say they’re not investing time and money anymore, then we have no clients and there’s no reason for us to exist.
The people that suffer the most are actually the project developers who are doing good work on the ground. They rely on carbon funding but now they cannot even sell the credits at the initial issuance price. Now their credits are worthless, right?
But we’re hoping that the new way of doing projects as shown in the VM0048 methodology can bring confidence to the market again.
Han: As the market moves towards digital MRV because of the proliferation of the data available, do we still need people around?
Xiao: Just to caveat, I think we’re referring to forestry projects specifically. For the other types of projects like biochar or cookstoves, that’s not so much you can do with satellite data.
But even for forestry stuff, the satellites cannot see through what’s below the canopy because that’s very granular. For example, if I told you that below the canopy there are many fallen tree trunks and fallen leaves which become very good carbon sinks. You can’t see that from satellite data. All that will degrade and become part of the carbon-rich soil.
At the same time there’s also mangroves. Mangroves are in the water, but the soil is also very rich in carbon. All these things are not measurable by satellites or lasers from the sky.
They still require humans to go down to verify how wide the trees are, taking samples of the soil to test how the composition of the soil has changed over time, and so on.
I think that’s still necessary. That’s still another part of the data set that’s used to do the carbon calculation. We need data from both underground and from the sky.
Han: You were mentioning earlier that this new way of doing things is so hard and so technical that it’s very hard to find people to come into the industry. What sorts of skill sets would people need to have to enter the carbon market given that the industry is moving in this direction?
Xiao: Because of the recent shift towards using more remote sensing stuff under the VM0048 methodology, I do think that in future, people doing forest carbon need to have experience doing coding and machine learning, but at the same time can really sit down and read all the science behind how people calculate carbon stock.
I would say it’s very multidisciplinary and they have to be very strong on each of these fronts. Unfortunately, those talents are so hard to find.
Audience: Have you gone through any accelerators?
Xiao: We have been through three accelerators already. The biodiversity accelerator last year by Silverstrand, there was the startup engine by Shell, and there’s one by our first investor, MiraclePlus.
Audience: Are you looking for Series A now?
Xiao: No. We are just at the seed round.
Audience: So you help landowners calculate the carbon credits that they can issue, and sell them to the market? Is that your function?
Xiao: We don’t help them to sell to the market. We’re more on the supply side and people like Rene [Velasquez] do the advisory and then the sales and brokerage. We help them calculate how much carbon credits they can get out of the land without actually having to deploy heavy capital first.
Johann and I concluded that our industry is very similar to mining and real estate. Before you build a building, you need to know how many units you can sell in total, and the position of your building in the context of the overall residential area, then you can start planning to get money to invest and then start building it.
Similarly for mining. You own the mine but you have to poke some points to calculate how much say lithium is inside this mountain before you actually go to people to say I need this amount of money in order to get this amount of lithium outside to sell.
Audience: Do you work with old or new projects?
Xiao: Both. Because the old projects will have to switch to the new method. Verra is going to announce that shift sometime later. And some smarter investors, they’re really quite anxious of what’s going to happen to their portfolio.
Audience: Are your clients aware of what they can do with satellite data, or the more advanced technologies that are available?
Xiao: The good ones, the ones that have already been developing projects, they are very aware of the technology. But the whole technical stack is still very challenging for them.
Audience: Do you have any plans to scale up your solutions to investors?
Xiao: Investors are already using our platform to understand their portfolio better and what kind of risk they may have. So that’s like the due diligence flow. Because investors also want to understand the risk profile of any of the assets they’re holding, or any assets they are going to hold.
We can do the due diligence very quickly in order to help them to understand how much carbon credits either they will lose from the current portfolio or how much they are going to get from the amount of investment they are putting into.
One of our customers is Carbon Growth Partners. It’s an Australian fund that deploy capitals into both REDD+ and mangrove restoration in a few parts of the world. They are using our services to first understand if they should invest in certain new projects. We do the project feasibility a lot faster than the traditional technical consultants. We’re also working with some of their existing projects which were issuing carbon credits under the old Verra methodology. They’re asking the projects to switch over to us for the next batch of carbon credit issuances.
Audience: Can you use AI to do predictive measures? Say predicting the risk of the carbon being dug up and released into the atmosphere?
Xiao: No, we don’t do that. The methodology already takes into consideration that scenario. When Verra issues credits, they don’t issue 100 % of credit to the project developers. They only issue a certain percentage and then the rest is basically a buffer. The buffer is to account for natural disasters or any unforeseen circumstances.
Han: Do you work with insurance companies though? Because this predictive modelling can be quite valuable to assess the risk of a project.
Xiao: I think we have a chance for the climate insurance field, but we don’t have a chance for the carbon insurance field because carbon insurance companies use a carbon pool. What they do is they buy a certain amount of carbon credits.
And imagine your projects happen to be of very poor quality. Maybe The Guardian exposes you like Kariba. Then what happens is the buyers of your projects can then now use credits from the pool, which come from other projects, to compensate for the loss.
That’s how carbon insurance works. They don’t really need to do the predictive analysis of how much carbon credits they will lose in the future because they base their risk on the verification reports they are receiving already. So they only get clients who are already issuing carbon credits with Verra.
What we do comes before the carbon credit issuance. But in this case, those projects are already way beyond that. They’re already on the market doing sales.
But for climate insurance definitely yes. Because people need to understand what’s the chance of some places getting burnt.
Han: If you’re building a beach hotel you want to know the chances of a tsunami hitting right?
Xiao: Exactly. There will be more breakthroughs in AI and machine learning technology as companies such as Nvidia build the next generation HPC (high performance computing) systems in order to calculate weather and disaster forecasting. I do think that what we are building can allow more scientists to start building things in Python in order to predict better analysis for the future.
And that we can basically monetize it by selling those models to individual corporations that really want to understand the risk of all this climate stuff.
Han: How difficult is it to replicate exactly what you do?
Xiao: Very difficult. In terms of the skill sets we already built from when we first entered the market until now, the technical consultant piece is very, very hard. Because it requires the knowledge of a lot of different disciplines in order to perform the end-to-end flow.
In terms of the product that we are building currently in stealth (PlanetGPT), that is also very hard to replicate. Because what you guys saw was basically just the surface. But behind the scenes we had a lot of optimizations that required putting different technologies together in one piece.
It’s very easy to use. You just need to ask questions and the platform shows you the answer. But actually behind the scene who knows how many models are talking to each other in order to produce the results that fast?
It’s a whole lot of science and research and trial and error. So if we spend like one year feeding these [models], I think others also will spend one year feeding these.
Audience: What is the limiting factor at this stage for the carbon market? Is it really the technology and measuring that is too costly?
Xiao: I do think that the bigger problems lie on the supply side, not so much on the demand side or the buyer side. The old method under VM007 was not too scalable. People doing 10–20 Excel sheets in order to calculate how much carbon sequestration you have.
That is part of the reason why it takes so many years to create a project. If every project takes two months instead of two years to launch, then more people will be interested in getting into the market. There’s also the complexity of the entire methodology and the standard. But that’s also important because being rigorous prevents low quality credits from destroying the market.
The second problem is the talent pool. We need multidisciplinary people who know about mathematics, statistics, and nature. People who have all these skill sets are so rare.
Whenever we try to hire someone from Singapore universities, people who know how to do geography and geospatial, they don’t know how to code. And people who know how to code don’t know what GIS is. As for people who know climate or ecology, it’s the same thing.
They basically are trained in one area, but not in all three areas together at the same time.
The last problem is the tools. The tech stack that people are using is just too old.
People in other industries are already using ChatGPT to do brainstorming, then using some other software to help them write a code, and another platform to help them to draft their legal documents.
But in our space, people doing spatial mapping usually use five different pieces of software. They ask you to send files to email. We are trying to change that by introducing one platform where you can do everything.
Han: We’re almost out of time. Any last thoughts?
Xiao: After all the scandals and the sagas of last year, I think we all need to learn from the mistakes and still have hope for the future.
