I first became aware of the ETH Studios projects at a meeting where ETH researchers presented their work to students. I was immediately intrigued by the prospect of working on a research project in the picturesque setting of Davos, a place I had so far only associated with skiing and the World Economic Forum.

As an applied mathematics student with an interest in using statistics and machine learning for environmental sciences, I found the perfect project: modelling river-runoff at the Snow and Avalanche Research Institute (SLF). The relevance of this task could not be more evident. With floods becoming increasingly frequent, studying their evolution under different climate scenarios is crucial. To achieve this, capable models that can use weather data to predict river levels are essential. Davos seemed like the ideal location to study rivers and streams, with its abundance of water and the lingering reminders of the severe 2005 flood in neighboring Klosters underscoring the importance of the task.

As I rode the train up to Davos, Zurich was already on the brink of spring, but the higher elevations told a different story. Snow piled up along the tracks, and sunshine was scarce. The first few days were filled with snow and a wintry mood in town, but eventually, the sun broke through, revealing breathtaking views from the office. I enjoyed having lunch outside and taking walks in the surrounding valleys.

I was warmly welcomed at SLF,  quickly getting to know my colleagues from the institute as well as visiting researchers from abroad. The communal lunches in the institute’s small cafeteria provided a cosy opportunity to speak to researchers from various backgrounds.

One highlight of my stay was a tour of SLF on my final day in Davos.  The variety of research conducted at the institute, from field-testing the effects of vegetation to advanced modelling of rockfalls and avalanches, was quite intriguing. During the tour, we visited the cold rooms on site, where researchers work at temperatures as low as -25 degrees Celsius to analyse snow samples from around the world. Our guide showed us his workstation, including a snow-making machine and an avalanche simulator.

Meanwhile, the project progressed rapidly, and as I returned to Zurich to continue my work remotely, we already had the first model results to analyse. It quickly became clear that machine learning is very well-suited for modelling rivers in the Swiss domain, predicting even extreme events with impressive accuracy and significantly outperforming state-of-the-art models. This opens up exciting possibilities for future modelling: better precision and faster inference mean that changes in river levels, floods, and droughts under climate change can be more accurately studied.

A good understanding of these changes is of great importance for hydropower generation, flood defences, and climate adaptation. While additional work is needed to ensure the reliability of machine learning models, even in a changing climate, the future is promising.

Interested in ETH Studios? Discover more here.