Snoskred Tromso Lyngen 1419 web

Satellites improve avalanche forecasting

Research scientists at NORCE have in recent years developed a method for detecting and mapping snow avalanches based on satellite images. These images will now be used by the Norwegian avalanche forecasting service varsom.no.

Snoskred Tromso Lyngen 1419 web
The map covers much of the municipalities of Tromsø and Lyngen in Troms County. All avalanches recorded using satellite images during the last five winters have been plotted on the map. Illustration: NORCE.

On 1 December, the winter season starts with daily avalanche forecasts on warning on varsom.no. Right from day one, high snow avalanche danger is forecast in Northern Norway and in North-western Norway. Last winter, avalanches claimed the lives of 13 people in Norway. The number of fatalities over the past 10 years is 75. Of these, 90 per cent were taken by avalanches while ski touring in the mountains.

Five years ago, a research group at NORCE in Tromsø (formerly Norut) began analysing satellite images with avalanche forecasting in mind. The images were provided by the European environmental satellites in the Sentinel series.

The researchers wanted to contribute to better avalanche forecasting, which would increase the safety of skiers and others in the mountains and in other avalanche-prone areas. The first milestone has now been reached.

Supplying images to varsom.no

“We have now developed a system for avalanche detection from satellites. We have recently transferred it to the Norwegian Water Resources and Energy Directorate, which will operate the system on its servers. The objective is for them to use it in some test regions this winter. Next winter, we will scale this up so all of Norway gets automatic snow avalanche monitoring by satellite,” says NORCE Senior Research Scientist Markus Eckerstorfer.

The Norwegian Water Resources and Energy Directorate (NVE) is responsible for the avalanche forecasting on the website varsom.no. The Head of Section of NVE’s Norwegian Avalanche Centre, Rune Engeset, is looking forward to using the avalanche forecasting system. The centre is now working on ICT integration of the new service.

“NVE’s researchers and developers have worked closely with NORCE to develop the method and put it into practical use with avalanche forecasting. The project, which is supported by the Norwegian Space Agency, shows how useful it is for Norway to participate in the Copernicus programme. This enables us to put in place daily monitoring of most avalanches that occur in Norway,” says Engeset.

Skjermbilde varsomno 20191203
The screenshot shows the avalanche forecast on 3 December 2019. High avalanche danger was forecast for large parts of Northern Norway and North-western Norway. Source: Varsom.no.

Using images from radar satellites

The twin satellites Sentinel 1A and 1B are radar satellites that are well suited for avalanche monitoring. The images are not affected by overcast conditions or poor visibility and, as they are polar-orbiting, the northern parts of Norway are especially well covered by frequent overflights.

This provides a lot of good data to analyse and improve the avalanche images, in addition to fresh images of avalanches throughout the winter season.

“We use a method that involves measuring changes in reflection from the Earth’s surface to the radar between two satellite image recordings of the same area. This enables us to detect and map snow avalanches,” says Senior Research Scientist Eirik Malnes, who is also part of NORCE’s avalanche group.

“The rough surface of avalanches changes the reflection significantly and, by comparing images from six days earlier, we can detect new avalanches continuously,” says Malnes.

Found 12,000 avalanches

During the winter of 2016-17, the research scientists analysed an area the size of the municipality of Tromsø. A total of 800 avalanches were registered via RegObs, a website and app used for registering observations. However, by analysing the radar images, they found around 12,000 landslides in the same area.

The research scientists have now collected data from the past five winters, which gives an even image of the avalanche-prone areas. This is the first time that data about avalanches have been collected over several winters in a larger area. When combined with meteorological data, they can create models that predict where avalanches can occur, given particular weather conditions. This will make the avalanche forecasts even more precise in the future.

The unique dataset is described in the publication Near-Real-Time Automatic Snow Avalanche Activity Monitoring System Using Sentinel-1 SAR Data in Norway. The publication was published on 2 December.

Big data and artificial intelligence

In the longer term, the huge amounts of data can be used to improve the registration of the avalanches. In another recent publication, Snow avalanche segmentation in SAR images with Fully Convolutional Neural Networks, the NORCE research scientists demonstrate that big data and artificial intelligence provide greater accuracy when it comes to avalanche mapping.

The research scientists conclude that the new method only overlooks small avalanches.

Unique competence and local fieldwork

The five NORCE research scientists involved in the new studies come from five different European countries. An interest in snow and avalanches has attracted several of them north to Tromsø, where the snow season is long and the distance to snow – for fieldwork or ski touring – is short. Tromsø is an ideal place to combine professional research careers and passion for ski touring. Moreover, it’s clearly a unique resource for NORCE to have expert satellite data analysts, who also have good knowledge of avalanches on the ground.

Work to develop a nationwide avalanche warning service will continue in 2020. The research and development work is financed by the Norwegian Space Agency, Norwegian Water Resources and Energy Directorate (NVE) and the Norwegian Public Roads Administration.

New publications

Markus Eckerstorfer, Hannah Vickers, Eirik Malnes, Jakob Grahn:
Near-Real-Time Automatic Snow Avalanche Activity Monitoring System Using Sentinel-1 SAR Data in Norway. (Remote Sensing)


Filippo Maria Bianchi, Jakob Grahn, Markus Eckerstorfer, Eirik Malnes, Hannah Vickers:
Snow avalanche segmentation in SAR images with Fully Convolutional Neural Networks. (pre-press)