Foto: Marc Katzenmaier
Lukas Drees works as a postdoctoral researcher in the EcoVision Lab with Jan Dirk Wegner at the Institute for Mathematical Modeling and Machine Learning (DM3L) at UZH. His research focuses on the interface between machine learning and environmental sciences. Within the DIZH Public Data Lab, he is involved in developing an indicator that describes the impact of green spaces on our well-being.
Lukas, why are you taking part in the DIZH Public Data Lab (PDL)?
Lukas: When Jan (Dirk Wegner) told me about it, I was immediately on board. The topic I’m working on at the PDL—how the living environment around me influences my quality of life—is something I find both fascinating and extremely important. But there’s another reason why the PDL excites me: I’ve long been interested in science communication, because I believe there’s generally a gap between how people perceive machine learning and what machine learning actually is and can do. Often, machine learning models are described as magic or a black box. I’d like to explore how we can explain to people, in an understandable way, how a model makes decisions and what’s really happening. There are tools like Explainable Machine Learning that can, for example, tell me why the model rates the quality of one green space higher than another. Especially when we want to develop results that urban planning experts will use to make decisions for the future, it’s essential that at least these experts understand how the model works.
What is your role within the PDL?
II’m supervising the PhD student who will join us in January 2026 to work on Flagship Project 3. Together, we’re developing a strategy for how to approach and advance the project. The aim is to link green spaces with people’s well-being. Based on the various green spaces in the Canton of Zurich, we’re trying to estimate where people feel particularly comfortable. Our input—the green spaces—consists of different sensor data, such as image data and point clouds. The critical point (laughs) is finding data on people’s well-being, which we need as a reference to train our model. For this, we could conduct surveys asking people how satisfied they are with the green spaces in their surroundings. But we could also use proxies, i.e., an indirect index: if we assume that property prices in the Canton of Zurich correlate with well-being, we could factor out all other influences (proximity to the city centre, transport connections, noise pollution, etc.) to obtain an adjusted price indicating: the higher the price, the better the green space for well-being. Alternatively, we could use existing indices as proxies: an index for air quality, heat stress, or even a happiness index. Another approach could be to generate hotspots using freely available movement profiles, like those from Strava, to infer the value of green spaces from how frequently they are used.
Where do you see further challenges?
I can imagine that different groups of people prefer different types of green spaces. As an example: Students might be interested in large green areas where they can do sports, while families with small children may prefer smaller, scattered green spaces with a playground. Working people might favour larger green areas or a forest outside the city where they can go for walks or practise forest bathing. So, there probably isn’t one single green space where everyone feels comfortable. Rather, well-being depends on people’s current everyday or life situation.
Another challenge is that green spaces can look very similar from above, on satellite images or aerial photographs, yet in reality they are not equally accessible to everyone. That’s why we will use different types of sensor data: aerial imagery, but also Street View images taken from cars.
With regard to the Public Data Lab as a whole, as my colleagues have already mentioned, finding a common language is a challenge.
What is your contribution to the PDL?
My contribution is of a technical nature. At the EcoVision Lab, we focus a lot on developing technical models that can infer insights from multimodal input data. Multimodal means that the data comes from different sources and can be in different formats—besides images, for example, point clouds from a laser scanner and text when someone describes a green space. We develop technical solutions that can efficiently process these different types of data together.
"I imagine a platform where people can create ‘their’ map and discuss it with us."
Lukas Drees
What is your goal?
The ideal outcome would be an interactive map where everyone can select specific options, such as age or demographic group. Based on these selections, the map would indicate which green spaces in the canton are most suitable for well-being. I hope the PDL will provide space for public engagement. I imagine a physical space or an online platform where people can create ‘their’ map and discuss it with us, so we can further develop the map and the underlying model. Ultimately, I hope our model will also be used by urban and spatial planners to create new green spaces.
What are you most looking forward to?
This project is quite different from those I’ve worked on before. That’s why I’m excited about the opportunity to collaborate with artists to create a machine learning model that is accessible to the public—essentially breaking down what it really takes to understand what happens inside an AI model. Within Flagship 3, I’m looking forward to bringing together insights from the various projects in our group: we’re already working on determining the phenology of trees using webcam data, assessing regional biodiversity through species distribution models, and detecting invasive plant species in drone imagery. For Flagship 3, we’ll definitely need a bit of all these elements.
Lukas Drees works as a postdoctoral researcher in the EcoVision Lab with Jan Dirk Wegner at the Institute for Mathematical Modeling and Machine Learning (DM3L) at UZH. Through studying geodesy in Bonn, he entered the fascinating field of remote sensing—and since then, he has been passionate about using intelligent image analysis methods to help preserve our planet. He knows the diverse green spaces of the canton of Zurich not only from images but also regularly enjoys them with friends for various sports such as frisbee, pétanque, or roundnet.