Learning to read the snow: avalanche awareness in the Pyrenees

A turbulent start

Before the actual internship at the Centre Lauegi began, my friend Alex and I had planned to spend a few days in the mountains. He would then drop me off in Vielha, the largest town in the Val d'Aran and my new home in the Pyrenees.

After his unit day, we started far too late in Toulouse. By the time we arrived at the southern end of the Val d'Aran two and a half hours later, it was already starting to get dark. We had prepared ourselves for the fact that the first ascent to Refugi Besiberri would be strenuous, both in terms of time and weather, but we were not prepared for what actually awaited us.

I lost a ski pole in the river. By the time we gave up the search, it had really got dark. We started the steep climb through the increasingly wild snowfall, taking our headlamps with us. The closer we got to the hut, the stronger the gusting wind became, which we actually wanted to avoid. As we could barely feel our hands, we zig-zagged our way forward without looking at the map more than every few hundred meters. When we finally found our way through the last steep section to the hut after 11 pm, we were both more than relieved.

But the wild stories about this 3-day trip, which was actually intended as a little warm-up, continued. The next day, I triggered a small snow slab. We had misjudged both the spiciness of the avalanche situation and the complexity of the passes, and after a long day we were both glad to arrive at Refugi Restanca at sunset.

After two avalanche noises in the distance and poor visibility, we decided in the morning not to try our luck again, but to take the direct route from the hut to the forest road and then down into the valley.
A pretty wild start to my time in Spain and definitely one that knocked my confidence a bit when it came to planning and safety in the (snowy) mountains.

The first few days - powder and language barriers

The first few days were a mix of language challenges, a warm welcome and a good dose of snow luck. The snow, which had presented us with problems at the weekend, had become more stable in the meantime, so we were able to combine daily observations and snow profiles with brilliant, easy descents. Linguistically, however, I was quite overwhelmed. Especially when we were out and about with other locals who spoke Catalan, Spanish and the local language of the valley, Aranès, but hardly any English or French, communication became really difficult.

I had promised to learn as much Catalan as possible before I arrived, but it wasn't that easy from Germany. As a result, I was often overwhelmed at first - even though I already knew a lot of vocabulary in theory, I could hardly follow conversations. On the one hand, I didn't want to keep asking what the conversation was about or whether someone could translate something for me. On the other hand, I didn't want to be the trainee who just stood there and obviously didn't understand anything.

Little by little, however, things got better. I invested time in online courses and vocabulary flashcards in the evenings and tried to keep up with everything as best I could and talk away, even if it must have sounded pretty confused at times. The team here was very understanding in this respect. Depending on the topic and vocabulary, we sometimes switched between Catalan, English, French and German in a wild mix of languages.

Overall, the three of them made it really easy for me to arrive. I was involved in the entire process right from the start: from the observations on the mountain to the dissemination in the daily bulletin and in blog and social media posts to the data analysis for scientific publications.

What is special about the work in Val d'Aran is that the small region has its own forecasting team. In contrast to many other regions in Europe, data and profiles from local observers are not transmitted to a central office, which then publishes meteorological analyses and the bulletin for a large region. Instead, everything is done locally and from a single source. In the mornings, the team goes out into the field with different members, digs up snow profiles and makes observations. In the afternoons, the bulletin and the publications for the local administration, such as the risk report for the pass roads, are produced from this.

The work in the avalanche warning service

Once I had gradually got to grips with the language, things got really exciting! My tasks were as varied as the job itself: from maintaining weather stations and webcams to public relations work with the local adult education center and creating social media posts to the core of snow science, the field work. I spent a lot of my time in the snow, observing avalanche activity, assessing signs of instability and digging snow profiles - usually with a crystal loupe at the ready.

Another aspect that I have learned to appreciate very much is the attention with which you move around the terrain as a forecaster. I've been ski touring myself for a long time and have assessed many individual slopes along the way, but avalanche forecasting requires a much broader perspective.

It's not just about the slope you're on, but about gathering information about the whole area. You try to understand how current and past conditions affect different exposures, elevations and slope characteristics. All this information is later incorporated into the bulletin. Questions are asked along the way such as: To what elevation did the snow get wet today in which exposures? Is the snow transport on the ridge strong enough to form relevant drift snow accumulations overnight? Where did we find surface frost that could form a weak layer under later snowfall?

The indoor side of avalanche forecasting

But life as a forecaster isn't just about field work - I also spent a lot of my time in the office. In addition to analyzing the weather conditions and the meteorological development of the season, my main area of responsibility was data analysis and processing. In addition to some visualizations of the snow profiles recorded during this season, I spent a lot of time verifying an AI system that Centre Lauegi is slowly integrating into its analysis processes. An Italian company, Alpsolut, provides AI-based analysis of the snowpack and avalanche situation for our two main weather stations. However, as their models were trained on data from Livigno, Italy, they do not yet perfectly match the conditions in Val d'Aran.

My role was precisely at this interface - I developed a simple system to monitor model performance by comparing the AI outputs with the human-generated avalanche bulletins. This involved overcoming a few technical hurdles, from automated data retrieval and processing to the compatibility of the two forecast sources. Without a real background in computer science, the learning curve was quite steep, but in the end almost everything worked out.

For example, this visualization compares the main hazard levels estimated by humans (red) and calculated by the AI (blue) at the Sasseuba weather station over the season. You can see that both follow a similar pattern, which indicates that the model recognizes the rough correlations. Nevertheless, the predicted hazard level deviates by one level from human judgment on a large proportion of days.
The graph on the right shows a "Confusion Matrix", which illustrates the hazard level predicted by the model for days with a certain level assessed by humans. It is noticeable that the model has particular difficulties with hazard level 2 and classifies it almost equally often as level 1, 2 or 3. This shows that improvements and fine-tuning are definitely still needed for medium hazard levels.

Cyclone Jana

Before the end of my season, things will get really exciting again in terms of avalanches in mid-March. After a winter with little snow so far and characterized by angular crystals, the low-pressure system Jana is expected to bring up to 70 cm of fresh snow within 3 days, triggering a rather tricky avalanche situation. At the peak of the storm, we are issuing avalanche warning level 4 for the areas above the tree line, as the cyclone's strong winds will also promote the accumulation of drift snow.

However, despite the unfavorable conditions, the deep snow is attracting freeriders to the mountains in what has been a rather disappointing snow year so far. There have been several avalanche accidents in the area around the ski resort. In the evening, we learn from the local rescue services that an accident on Montlude in the north of our region has ended fatally.

As a team, we discuss what conclusions we should draw from the situation. A day on which accidents occur more frequently always raises the question of whether the situation has been correctly assessed and communicated. However, the avalanche report actually painted a clear picture of the situation: "Due to the fresh snow and wind, many medium to large avalanches are expected. A lot of experience and caution is required to move around in unsecured areas".

When we visit the southern, hardest-hit areas in the days following the storm to assess the activity during the peak and map the avalanches, the picture we had predicted is confirmed: two size 4 avalanches, i.e. avalanches that can destroy even large buildings and forest areas, reach the valley floor and several other size 3 avalanches now dominate the valley landscape.

Almost all the steep slopes show avalanches, with deposits piling up below them. I am deeply impressed and affected. Of course, I've seen pictures of large avalanches before, and I've even seen one or two myself. But the combination of following (and predicting) the situation so closely, being out and about in the region almost every day, and knowing about the fatal accident two days earlier, makes me feel the consequences much more clearly.