AI for the environment

Profile photo of ETH Zurich master's student Kenza Amara, developing new usages of AI for the environment
Kenza Amara is a Master’s student at ETH Zurich and is passionate about applying artificial intelligence to solve the big global problems of our time – such as climate change.

By: Kenza Amara, master’s student at ETH Zurich, part of Crowther Lab and DS3Lab

Artificial intelligence (AI) for the environment: It first seems somewhat paradoxical. But could the rise of these new technologies be in line with forest landscape restoration and environmental protection? Seeking out an answer to this question is a challenge that I have set for myself. But why and how?

From Paris to Zurich… and OneForest

With a strong background in mathematics, physics and computer science, I have always been motivated to be involved in solving the great challenges of our era. This motivation became even stronger when I was studying my first Master’s degree in Data Science at the Ecole Polytechnique Paris. I still remember one specific day: I was attending a conference on the damages humanity has caused on the environment – and left it in tears. Seeing the environmental distress our planet is under was the final straw: I had to do something!

I decided to do a second Master’s degree in environmental sciences at ETH Zurich. I wanted to learn more about our environment and the processes underlying it, and with full knowledge of those facts, put my technical skills in deep learning, and more broadly in computer science, to practical use. The following study year was very different from the past years at the French Grand École: I read a lot of articles on climate policies, studied conflicts over international waters and learned about international and European environmental laws. But for my Master’s thesis I decided to go back to my favourite field: neural networks and their many applications. I contacted David Dao from DS3Lab to start a multi-disciplinary project with Crowther Lab. This gave rise to OneForest, a fantastic initiative to merge drone imagery and citizen science to build informative maps of our forests worldwide.

Putting drone images and citizen science on the map

We started with the disappointing observation that one major issue for forest monitoring was the lack of information on the forests. There is no existing high-resolution labelled dataset on trees at scale. Therefore, we decided to build a global map of our forests that would gather all existing trees and document their exact position and characteristics. This large-scale data source would enable scientists, farmers and policymakers around the world to reference the same global database and monitor forest landscapes in even the most isolated parts of our planet.

From having to create our own species training dataset, to dealing with the various noise sources in the data, or developing an algorithm that could reliably match drone images with citizen-made ground measurements: our project was for sure riddled with (exciting!) challenges. Now, OneForest is able to return stable mappings and offer up rich information about the trees’ attributes such as name, species, genus, height and trunk diameter for any given forest landscape area. With OneForest we can begin to better understand and answer questions about forest ecosystems on a global scale.

Photo collage of Crowther Lab and DS3 Lab members gathering test data in the forest.
Who says training algorithms is desk work only? To gather more input data, the team ventured out into the field to measure trees and take some test drone images!

Global maps for global impact

What simply began as my Master’s thesis project has, after its completion, drawn the attention of multiple NGOs and academic research groups. Back when I was developing this end-to-end method, I definitely did not expect so much enthusiasm for OneForest from the scientific community! But I soon realised that there was great potential to create wider impact. As an integral part of Crowther Lab, OneForest also became a promising tool for the new open-data ecosystem restoration platform Restor. OneForest provides automated ecological insights on individual tree level: therefore, it can generate localised information about trees no matter where on the planet – even in the most isolated places where data is difficult to collect in large amounts!

To go back to the question in the beginning: AI and environmental sciences are a pair not to be missed out on. Artificial intelligence has long been used and explored for robotics, marketing or the industry. But working on OneForest and seeing the feedback it received has reinforced my belief that there is also a lot of potential for AI to generate a deeper understanding of our environment and identify global ecological patterns. From checking how much carbon storage is in a specific group of trees, or observing the evolution of a forest in terms of tree size, to detecting whether the right species have been planted – I can envision many future applications.

A backpack full of new experiences

At Crowther Lab I discovered an academic environment with unwavering team spirit that encourages participation in diverse activities and the sharing of ideas and experiences. It’s still very rare to have computer and environmental scientists working hand in hand and I’m glad that exceptions like Crowther Lab made this collaboration possible! This teamwork between us scientists results in innovative solutions that seem to be truly appreciated by NGOs, private companies and political actors. It shows how vital interdisciplinary work is to bringing our understanding of the world further.

But now my Master’s thesis and my time at the lab has come to an end. Where to go from here? With a backpack full of new, enriching experiences, I decided to continue my interdisciplinary research at the ETH AI Center. This brand-new centre encourages AI to be put to service for society in domains such as sustainability, health and robotics, opening up new horizons to collaborate with the industry, public administrations and international universities.

And what about OneForest? My hope is to see OneForest grow, be developed further, improved, and of course, used by scientists, restoration agencies and farmers. It will hopefully open the doors to faster and easier forest monitoring, necessary for effective forest landscape restoration. And maybe, just maybe, it will serve as an inspiration for the next generation of students wanting to take part in solving the global problems of our time.