Using satellite imagery to study snow conditions on the Walker’s Haute Route

*Snowy section between Cabane Mont Fort and Cabane de Prafleuri on WHR in 2016. Photo by Tuomas Vuorinen*

I solo hiked the Walker’s Haute Route from Chamonix to Zermatt in June-July 2016. It was my first hike ever in the Alps. In Chamonix, before even starting, the mountain info told me it cannot be done because the route is covered in snow.

I still went, and glad I did, but all useful information regarding snow conditions I learnt firsthand on the route. On the most difficult and snow-risky section between Cabane Mont Fort and Cabane de Prafleuri, I had to turn back mid-way as I didn’t dare cross a snowfield where there was a stream underneath and was running out of time. When I got back to the hut, I couldn’t drink my beer in peace as a dozen hikers were asking me about the snow conditions on the route.

With easy access to Sentinel 2 imagery at the time, everyone’s life would have been much easier. Today, we test how satellite imagery can help understand snow conditions on high-altitude routes.

Area of study

We study the below area near Verbier and specifically the route between Cabane Mont Fort and Cabane de Prafleuri, where I turned back. I consider it the crux when it comes to route’s snow risk due to the distance, steep slopes and lack of alternative walking routes.

*The section of Walker’s Haute Route under study.*

What NDSI and our script does

Our script finds up to 4 usable Sentinel 2 scenes from June 20th to July 12th for all years 2019-2025, capturing late-season snowpack conditions.

The script filters out clouds, cloud shadows, and defective pixels using Sentinel-2's Scene Classification Layer. It then calculates cloud coverage percentages for the remaining images and selects the four clearest observations from each year.

The analysis approach is based on NDSI - Normalised Difference Snow Index. It is based on optical bands so requires cloudless skies for a usable scene. Copernicus Browser tells me it relies on snow’s high reflectance in visible green light, and low reflectance in shortwave infrared. With certain thresholds (0.4 for us today) pixels are classified as snow in our scenes. Visual layers are generated.

The script then calculates the total snow-covered area by multiplying snow pixels by their ground footprint (100 square meters per 10-meter resolution pixel) and summing across the study area.

Example results for hiking intel

The most interesting data are the visual layers the script produces, as it clearly shows snow cover on the route for a visual reference. It shows remarkably well the snow areas and how the to snow recedes, aligning with my own experience of the snow lines. See below two examples from 2019, a year with a lot of snow.

*Snow coverage on the study area between Cabane Mont Fort and Cabane de Prafleuri.*

Annual snow coverage results

While the above-kind visuals are the most useful for hiking, I also wanted the script to show annual variations in snow coverage.

There were some scenes that were partly blocked by cloud-cover, but it wasn’t obvious which of all pictures were affected. I manually reviewed every layer and eliminated the ones without a clear scene. For a visual analysis, they could still be helpful.

Below are the results showing varying late June to early July snow coverages. Year 2022 shows very low snow coverage. Google Search reveals this was an exceptionally low-snow winter. Years 2019-2020 with snowy winters show snow extent that still continued well into the early July.

Applications for forecasting and hiking decisions

This analysis was easy (with help of AI) but gives excellent snow condition insight on the route. Even simpler, anyone can go to Copernicus Browser and review latest scenes with the built-in layers while hiking.

I got tempted to build an automated tracking of key alpine passes to serve hikers, but will refrain from it this time. Because the snow levels are directly driven by spring snowfalls with fairly consistent melting curves below certain snow coverage extent, it would be possible to start publishing forecasts already in May-June.

The only difficult part would be to find a vertical slice somewhere in the Alps that correlates well with high-altitude snow depth. This is quite similar to how we estimated Lake Gatún water levels for Panama Canal.

WIth that, hikers that could adjust their calendar and plans to avoid snow-covered sections.

Conclusion

This was a clean example of how we can measure snow coverage in a key alpine pass, but also simply leverage the visual outputs to enable real-time decision guidance for hiking. It would have helped me and various others hikers when doing the Walker’s Haute Route in 2016. I will use it to understand typical snow extents in my future hiking routes.

See you,
Orbital Vantage