How does the Lyon Metropolis use AI to plant trees?

The Lyon Metropolitan Area is developing an artificial intelligence tool to better adapt its territory to climate change. Called IA.rbre, this project draws on public data and collaboration with researchers to guide choices regarding vegetation and de-impermeabilization.

In a territory already constrained by underground networks and street furniture, planting a tree can be a logistical challenge. This difficulty is addressed by the IA.rbre project, led by the Lyon Metropolis. "The Metropolis is one of the territories most exposed to climate change," says Anthony Angelot, project director. "One of the strategies is greening. But for that, you need to know where you can plant trees." Born from an initial tool called the "plantability layer," IA.rbre cross-references dozens of geographic datasets to identify areas suitable for greening or de-impermeabilization. The challenge: to enable elected officials to make decisions based on precise analyses of the terrain. "We mainly use existing data, such as gas or electricity networks, and we produce new data, for example on vegetation, thanks to aerial photos analyzed by AI."

Developed with Lyon 2 University and the Téléscope cooperative, the tool is intended to be transparent and replicable. "The entire project will be open and documented so that other communities can replicate our approach," insists Anthony Angelot. But the use of artificial intelligence also raises the question of its ecological impact. "We want a frugal AI, which we use only where strictly necessary," emphasizes the project director. The team plans to evaluate the carbon footprint of the system, taking into account in particular the positive effects of the trees planted.

As for the risk of seeing digital tools supplant political decision-making, Anthony Angelot reassures: "It is the agents of the Metropolis who define the parameters of AI, and the human always retains the decision."

Read also: Immersive exhibition in Lyon: "Thanks to AI, you will be able to talk to Napoleon"

Read also: Economic meetings in Lyon: artificial intelligence at the service of Franco-Swiss companies

Read also: "AI systems are starting to do unexpected things," says Flavien Chervet.

The full transcript of the show with Anthony Angelot:

Hello everyone, welcome to the 6 Minutes Chrono program, the daily meeting of the Lyon Capitale editorial team. Today, we're going to talk about artificial intelligence, community, and political management, since the Lyon metropolitan area, accompanied by the SCOP Téléscope, the Lumière Lyon 2 University laboratory, the CNRS, IRIS, and the State, is developing a new AI-enhanced tool to help with decision-making in adapting the city to climate change. And to talk about it, we welcome Anthony Angelot, who is director of the IA.rbre project for the Lyon metropolitan area. Hello Anthony Angelot. Thank you for coming to our set. We're going to get to the heart of the matter. What is the genesis of this project and, above all, what is it for? How will it work?

Thank you for having me. So, the Lyon metropolitan area is one of the areas that will be most affected by climate change. And so, we must adapt the territory for its residents. One of the strategies is greening. And for that, we need to know where we can plant trees in an area that is already very constrained. So, that was the genesis of the project. It's really this question: where can we plant, knowing that we take into account a very significant number of constraints, between underground networks, street furniture, and even existing vegetation.

The project was born in the Lyon metropolitan area before all the partners I mentioned? Is it a metropolitan initiative?

It's a project that we called the "plantability layer," which aimed to say, in the metropolis, where it would be easy to plant or, on the contrary, where it would be much more complicated due to the constraints that I mentioned.

Okay. So, it's really focused on ecological and environmental issues; you're talking about greening. It's not going to go beyond that area. It's important to clarify that as well.

The starting point is vegetation. And the idea was, beyond this tool that helps us determine where we can plant, to go further and address all the challenges of adapting the territory. So, today we're also going to work on issues of de-impermeabilization, that is, how to get rainwater into the soil to feed this vegetation, but also on social vulnerabilities and biodiversity issues. So, we're expanding the scope, but always thanks to data and artificial intelligence.

So, you were talking about data. That's a good segue for me. What data does your tool rely on? Do you communicate about it? I'm mentioning this because many generative AIs used by the general public don't communicate about the data that generates images, videos, or text. What collection, what data base does yours rely on?

We mainly use geographic data to map the territory. This is often already existing data, particularly network data: we know where the gas and electricity networks are, etc. And other data is produced by us. For example, regarding the territory's vegetation, we need to know where the upper and lower layers of vegetation are located. And to do this, we will collect aerial photos across the entire metropolis and use AI to detect all the trees and shrubs present.

Okay, so this isn't behavioral data about the people of Lyon or the Greater Lyon area, or about who they are, necessarily. This is public data.

Absolutely. This is public data. Most of it is already open and available on the Open Data platform.

And how does it work? Are you creating the tool internally? Or are they aggregates? How will this tool work, broadly speaking?

The first step is to gather all the data to produce maps: where to plant, where to de-impermeabilize, where to act in the face of climate change. So, within the metropolis, we produce a lot of data with our partners: GRDF, Enedis, the fire department, etc. Then, we aggregate this data to create a real decision-making tool. It's not very useful to have an aggregate of 35 or 60 data sets if we don't know what they tell us. The tool is therefore developed internally with the University of Lyon and the Téléscope cooperative, which have all the data expertise. The entire project will be open and documented so that other communities can reproduce our approach.

It's quite unprecedented in France for local authorities, clearly in conjunction with the research community, to create their own artificial intelligence tools. Or are there other scenarios?

There are other cases. Artificial intelligence is a growing topic. More and more local authorities are tackling it. Regarding adaptation issues, we responded to a national call for projects. The goal is for Lyon to truly be a driving force in AI for the ecological transition, and for this project to be replicated in other local authorities. This is already the case in Bordeaux Métropole, for example.

Okay. It's already started, after all. The Lyon model is already being exported to France. You could say that.

Exactly. They were the first on the subject of greening.

There's also an issue with AI, which is ecology... or rather the impact of AI on the environment. Since you've created an AI that addresses environmental issues, how do you integrate pollution, heat, and the energy demand that artificial intelligence requires? How do you integrate these issues into your tool?

You're right to point that out. It's an extremely important issue for us. The idea is to have frugal AI. And by frugality, we mean: using it only where it's strictly necessary. There are many things we can do without AI, and in those cases, we don't use it. And when it's essential, we try to develop the most eco-designed algorithms possible. If we do a carbon and energy assessment of the project, we can imagine that planting trees—we'll do a review at the end of the three years—will partially offset the emissions emitted at the launch of the project.

So, okay, we're doing a bit of political fiction here, but isn't there a risk in the long term, by constantly developing these projects, even in the political decision-making process? We're talking about tools that will help elected officials make decisions. Isn't there a risk that it's the tool that ends up making the decision, saying "this is the best, this is the worst," and that humans or elected officials end up having to follow these choices?

So, two levels of response. First, we want to have explainable AIs, that is, not black boxes, but systems whose reasons for recommending a particular solution are understood. It is agents of the metropolis who will define the parameters of the AIs. So we can explain where this decision-making support comes from. Then—and this is necessary for all AIs in the metropolis—it is the human who retains the decision.

Very well. That will be the final word. It's already the end of the 6-minute timer . Thank you very much for coming to our set to explain this AI.rbre tool. More details on the latest news in artificial intelligence and the ecosystem that creates it can be found on the Lyon metropolitan area website and on lyoncapitale.fr. See you soon.