Snow or plant

An SLF employee records vegetation data at an IMIS station in the Bäretälli near
An SLF employee records vegetation data at an IMIS station in the Bäretälli near Davos. (Photo: Christian Rixen / SLF)

Machine learning helps researchers to use weather stations to investigate the consequences of climate change for the growth of alpine vegetation.

This text was automatically translated.

Six days earlier than in 1998, on June 14 instead of June 20, plants begin to grow on average at higher altitudes in the Alps. Michael Zehnder, biologist at the SLF, explains this with climate change and the sharp rise in average temperatures in the mountain region. The growth rate and the point of maximum growth have also changed due to the warmer mountain spring over the past three decades.

He investigated this with the help of the weather stations of the Intercantonal Measuring and Information System IMIS. Since the end of the 1990s, around 190 stations have been measuring numerous weather data such as wind speed and temperature every half hour, two thirds of which also measure snow depth. The federal government, cantons, municipalities and other interest groups support this nationwide project, which is managed and analyzed by the SLF.

In addition to snow, the ultrasonic signal from the snow depth sensor measures the size of every object under the sensor. Zehnder makes use of this: "We can use the snow depth signal to track plant growth in summer and observe how it changes over the years without having to be on site ourselves."

Unless it is snowing. This also happens more frequently in summer, as the stations are located throughout the Swiss Alps, usually above the tree line between 1800 and 3000 meters. Algorithms then help to distinguish whether the sensors are measuring snow or grass. The stations themselves cannot do this.

Jan Svoboda, an expert in machine learning (ML), has trained a model with a large amount of data in collaboration with the Swiss Data Science Center (SDCS). "By linking it with other sensors from the measuring stations, the algorithms can separate snow from plants," he explains. For example, on summer days, the temperatures of the ground and air run in parallel at well above zero degrees. A blanket of snow, on the other hand, always has a maximum temperature of zero degrees, whereas the air above it fluctuates, even in the plus degrees range. The algorithms have learned such correlations.

Measuring the height of plants using the IMIS stations is nothing new, says Zehnder: "But with the new ML approach, our results are more accurate." This saves time on the manual readjustments that are still required.

However, the system has its limits. "At stations that are too high up, the vegetation is too small to reliably measure growth," the biologist points out. In addition, the sensor does not differentiate between plant species. Zehnder therefore wants to record the vegetation on site again this summer to investigate whether new or different species are contributing to the change in growth.

It is interesting to note that the growth phase of the plants begins earlier than 25 years ago. On average, however, the thinning process still takes place at the same time in the same places. "This means that the plants need less time to sprout after the snow cover has disappeared," says Zehnder.

This article first appeared in the Davoser Zeitung on 25 June 2024.