Researchers turn to artificial intelligence to model shrinking snow cover

In a leafy courtyard in the northern Italian city of Bolzano, children chase each other as educators look on, interns sip cappuccinos and researchers jostle to get to the lab.

In the distance, pine-clad mountains rise in all directions like majestic gatekeepers. The famous Dolomites of the Italian Alps are breathtakingly beautiful, but they are also a stark reminder of how climate change is making snow-capped peaks more unpredictable.

In July, 11 hikers were killed when record high temperatures contributed to a massive piece of the Marmolada mountain glacier break away. Shrinking glaciers and less snowfall have also led to the drying up of the Pothe longest and most important river in Italy for agriculture and hydroelectric power.

This week, as world leaders prepare to gather in Egypt for the COP27 climate change conference which begins on Sunday, a UN report warns glaciers around the worldthe last of which in Africa, will have disappeared by 2050.

Here in Bolzano, researchers from the private clean energy research group Eurac have pieced together a long-term picture of how the world’s snow cover has already changed, using modeling and artificial intelligence.

Their study, published in Nature’s Scientific Reports, found that globally it has declined over the past 38 years, with 4% less mountainous areas covered in snow and an average of 15 more snow-free days. per year.

In the Rockies, the study found that the number of days without snow cover reached up to 30 at certain times and in certain regions, with a slight increase in snow in tiny microclimates.

« The warming of the minimum temperature, as well as the decrease in winter precipitation and more [rain] … can speed up the fusion phase, » said Claudia Notarnicola, a scientist at Eurac’s Earth Observation Institute who led the study.

“The strongest effect we see is the anticipation of the melt season, [spring temperatures] come sooner. »

From aluminum to the development of clean energy

Eurac’s work takes place in a facility called Nature of Innovation (Noi) Tech Park, which a century ago was the site of Italy’s burgeoning aluminum production, one of the most energy-intensive and polluting, launched by the fascist dictator Benito Mussolini.

At its peak, the region produced a third of the country’s aluminum, until production dried up due to global competition and ended in the 1980s.

Today, the converted factories, along with modern buildings, are part of the growing hub of environmental innovation and research – home to everything from start-ups and clean energy labs to environmental agencies, a university campus and a daycare centre.

« In this [region of] South Tyrol, nature has always had an important place in the way we live and do things,” said Wolfram Sparber, renewable energy manager at Eurac, one of the main occupants of Noi. “The idea was to provide a place with a high level of work value, a pleasant place to live, a good combination of work-life balance. »

Sparber shows a lab where scientists spend entire days in large refrigerator-like rooms testing equipment and human response to extreme weather on mountain peaks reaching 9,000 meters.

In another building, solar panel tests are underway, with a technician tasked with detecting malfunctions to increase efficiency. Eurac is involved in several large-scale European projects to develop high-performance solar panels to help revive production in Europe after Chinese manufacturers undermined European manufacturing.

But – unusual for clean energy technology centers – Eurac also conducts climate change research, in a sleek, elevated glass structure nearby.

A global view extended over time

The recent study follows another by Notarnicola published in 2020 which looked at snowfall going back two decades and showed evidence of a decline in snow cover in 78% of the world’s mountainous areas.

What’s different this time is that researchers used artificial intelligence (AI) to explore what was happening with high-altitude snow cover before consistent satellite data became available in 2000.

The 2022 study used MODIS satellite data available from 2000 and, using artificial neural networks, modeled data up to 1982.

« What Claudia has done here is really innovative, » said Chris Derksen, a research scientist with Environment and Climate Change Canada’s Climate Research Division.

« For climate studies, what we really want is as many years as possible – 30 to 40. »

Derksen says mountain studies tend to be regional, with researchers in North America focusing on the Rocky Mountain or Sierra Nevada ranges, for example, or in Switzerland, Austria or Italy, on the Alps. .

“From a climate change perspective, the more we can look at the whole hemisphere, it just gives us a more powerful signal of how things are changing,” Derksen said of the need for global studies. .

Ground data required

Still, using MODIS satellites to study snow has its limitations, said John Pomeroy, Canada Research Chair in Water Resources and Climate Change at the University of Saskatoon.

He said the low resolution of satellite data and the inability to see through thick forests, thus missing the snow below, can lead to errors. It can also confuse cloud cover with snow.

« I don’t dispute the results, » Pomeroy said of Notarnicola’s study. « It’s useful to have a global analysis like this – the way she tried to fill in the gaps and uncertainties with artificial neural networks was clever.

« But there are also problems with these in that they are trained for the past and are data-driven, so they can be pulled in the wrong direction. »

Pomeroy is not against the combined use of satellite technology and AI, but he would like to see other controls involved, such as the field study of snow carried out by field stations and sites. mountain research, snow reports and other data. sets.

More and more observational data are becoming available worldwide, with the establishment by Pomeroy and others of the Common Observational Period Experiment (COPE), a network of sites for intensive observation of high mountain areas. of the whole world.

Brian Menounos, Canada Research Chair in Glacial Change at the University of Northern British Columbia, agrees that direct observational data can only help enrich model-refined satellite data. complex mathematics. Especially, he says, with the challenge of coping with smaller, multi-year climate trends of drought or increased precipitation, as part of the larger trend of global warming.

“We have to think about different time scales. … It’s really this decadal variability that is much harder to predict and will greatly influence water availability,” he said. « And that’s really what we’ve seen around the world. »