Sknow: How it works
As awareness about the Sknow device grows, we want to ensure that those new to our development have an opportunity to understand how the device works. This overview aims to answer the most common questions we receive about our device.
Real-time radar avalanche measuring is well-established for remote sensing and is often used for automated detection. The ability to generate real-time avalanche danger estimation is key for safe skiing in the backcountry because currently a skier often cannot count on any assessment besides his or her vision and knowledge of the terrain. Most skier-involved avalanches are triggered by the skiers themselves. Very often a persistent, weak snow layer causes the avalanches and can result in the loss of human life. Knowledge of avalanche dangers is essential for safe skiing, especially in remote, high mountain areas.
Presently, location-specific estimation of avalanche danger during skiing is not yet established. Sknow aims to fill this gap by digitally measuring a full snow profile including weak layers and snow depth down approximately 2m, as well as slope gradient and applied forces. Our technology will keep skiers constantly informed about the snow conditions under them, no matter their location. The measurements are used to provide a real-time danger assessment during skiing, and if an internet connection is established, the data will be sent to our cloud for further analysis and danger estimations.
Our radar-based sknow device is installed directly on the ski and measures the snow layer structure below the ski. This measured data is used to predict the snow-properties as described in Bradford, Harper, & Brown (2009). The first step for avalanche danger estimation is the ability to measure snow layers below the ski during skiing. We conducted simulations and extensive lab and field tests under various snow conditions of dry and wet snow. For example, as a worst-case scenario, we tested highly consolidated wet snow in the Norwegian mountains in May 2018. The air temperature was around 12 degrees Celsius and the snow temperature was relatively constant in the whole snowpack at 0.2 degrees Celsius.
The following Figure 1 shows the result after our state-of-the-art processing workflow. The figure demonstrates that our device penetrates down to 2m (512 samples with 26e-12 seconds per sample). Some possible vertical fractures and drainage zones can be seen as well. The figure shows the current resolution of 2cm snow layer-thickness.
Measurement of the snowpack to a depth of 2m is enough. If there is a weak-layer below two meters, the likelihood of triggering an avalanche is quite small. Our simulations (Figure 2a and 2b) and the results from Monti et al. (2016), Thumlert and Jamieson (2014), and Gaume and Reuter (2017) show clearly that the most risk of triggering an avalanche exists in the first 2m of the snowpack.
In order to understand how radar data needs to be processed, it is important to understand the nature of the signal. Radar signal reflections from snow layers are highly dispersive which means that the signals reflected from the snow layers are frequency dependent. We are using an ultra-wideband radar source signal over 4GHz bandwidth with a main frequency between 6 and 9GHz depending on CE/ETSI (EU), FCC (USA) and ISED (Canada) regulations.
Dry snow is less dispersive than wet snow. The dispersive nature of snow causes signal-damping for high frequencies, like a low-pass filter but less linear over high frequencies. To overcome the frequency-depending nature of the reflected snow data we developed a processing workflow where the data, after pre-processing (like de-wow, bandpass-filter…), is transformed into the frequency domain for further processing. The results of our processing workflow are used to calculate the snow properties.
Detecting snow layers using radar has been in practice for many years ( Schmid et al., 2014), but the real challenge is to process the data without significantly changing the shape, phase and amplitude of the measured signals. This is essential for an accurate snow-property estimation and later for accurate machine learning. The calculated snow properties from the radar data are used as “labels”, and the snow reflections as “features” for an automated, supervised machine learning. After training our machine learning core in our data processing cloud and uploading the core to the Sknow device, we use the real-time measured snow data from our Sknow device to estimate snow avalanche danger at the current skier position.
We hope this explanation provides some clarity in understanding how the device will work. Some specific details must be omitted as they are trade secrets. For further reading, please see our report published at the International Snow Science Workshop (October, 2018). Found on our website, link below.
