Detecting Avalanche Deposits using Variational Autoencoder on Sentinel-1 Satellite Imagery

Avalanche monitoring is a crucial safety challenge, especially in a changing climate. 1 Remote sensing of avalanche deposits can be very useful to identify avalanche 2 risk zones and time periods, which can in turn provide insights about the effects 3 of climate change. In this work, we use Sentinel-1 SAR (synthetic aperture 4 radar) data on the French Alps for the exceptional winter of 2017-18, with the 5 goal of automatically detecting avalanche deposits. We address our problem 6 with an unsupervised learning technique. We treat an avalanche as a rare event, 7 or an anomaly, and we learn a variational autoencoder, in order to isolate the 8 anomaly. We then evaluate our method on labeled test data, using an independent 9 in-situ avalanche inventory as ground truth. Our empirical results show that our 10 unsupervised method obtains comparable performance to a recent supervised 11 learning approach that trained a convolutional neural network on an artificially 12 balanced version of the same SAR data set along with the corresponding ground-13 truth labels. Our unsupervised approach outperforms the standard CNN in terms of 14 balanced accuracy (63% as compared to 55%). This is a significant improvement, as 15 it allows our method to be used in-situ by climate scientists, where the data is always 16 very unbalanced (< 2% positives). This is the first application of unsupervised deep 17 learning to detect avalanche deposits.