This summer, I took on something completely new: clustering algorithms. I had never worked with them before, but during a six-week internship at the Swiss Federal Institute for Snow and Avalanche Research (SLF), I had the chance to dive in and learn. I was part of the Hydrology and Climate Impacts in Mountain Regions (HYCLIMM) group, which focuses on understanding how hydrological extremes and climate interact in mountainous regions.

The topic of my internship built on something I knew a bit about – droughts in rivers – which I had explored during my bachelor’s thesis. But this time, the goal was to go a step further: develop a method to group drought events from the past 40 years based on how they were formed – their hydrometeorological generation processes. These can be a precipitation, storage or snowmelt deficit, high temperatures and increased evapotranspiration, or a combination thereof. After creating the clusters, I would then explore them in more detail: when and where do they happen? How long do they last? How severe are they? And how have they changed over time?

The internship took place in Davos, and I was able to stay there the whole time. That gave me the chance to get to know the team and get a better feel for how they work. It also meant being surrounded by mountains, which made for some great weekend hikes and quiet evening bike rides.

When I arrived, my supervisor Giulia Bruno welcomed me warmly and helped me settle in. The team was kind and open, and some routines – like regular lunches together or afternoon coffee and cake – helped me feel included. I appreciated these small things, and they helped me feel at home quickly.

One of the trickiest parts of my project was the explorative nature of it. There wasn’t a clear roadmap to follow. The first few weeks were all about figuring out what kind of data the clustering algorithm needed to do a good job. As I soon realized, a clustering algorithm is only as good as the data you give it. So, I spent a lot of time selecting and preparing data – trying to feed the algorithm with the “right food,” as I liked to think of it.

What helped was being able to talk things through with the team. They took time to discuss my questions and offered new perspectives. Slowly, things started to make more sense. The clusters began to show meaningful patterns from a process-oriented perspective, and I started asking more focused questions:

• What processes lead to certain clusters?
• Are some more common in specific areas or seasons?
• Has the frequency or severity of these clusters changed over time?

Getting to that point felt rewarding, especially after the initial uncertainties. It also made me more comfortable with the idea that exploration – and sometimes getting stuck – is part of the process.

Beyond the technical skills – like learning how to use clustering methods and improving my programming – this internship also taught me something more personal: how to deal with frustration. Some days were tough, especially when things didn’t work as expected. But over time, I learned to be patient and to trust the process. By the end of the internship, I felt a bit sad to leave. The team had been so welcoming, and I had learned much in such a short time. But I was also proud of what I had achieved. The work I did opened up some follow-up questions, and maybe I’ll get the chance to explore those further, possibly in a master’s thesis.