Monitoring snow avalanches from SAR data with deep learning

TL;DR

Snow avalanches pose significant safety and infrastructure risks, and traditional in-situ methods are limited for large-scale monitoring. The paper demonstrates deep learning–based segmentation of SAR imagery to map avalanches at pixel level, introducing an FCN with a PAR-conditioned Attention Net and a rich set of input features, achieving substantial gains over threshold-based baselines (F1 ~66.6%, IoU ~54.3%). The dataset is expanded to 4,507 annotated SAR images, and a broad architecture evaluation identifies the FPN with an Xception backbone as a favorable balance of performance and compute, enabling large-scale Norway deployment that reveals spatio-temporal avalanche patterns. These findings highlight the method’s potential for operational risk mapping and real-time forecasting when integrated with meteorological data.

Abstract

Snow avalanches present significant risks to human life and infrastructure, particularly in mountainous regions, making effective monitoring crucial. Traditional monitoring methods, such as field observations, are limited by accessibility, weather conditions, and cost. Satellite-borne Synthetic Aperture Radar (SAR) data has become an important tool for large-scale avalanche detection, as it can capture data in all weather conditions and across remote areas. However, traditional processing methods struggle with the complexity and variability of avalanches. This chapter reviews the application of deep learning for detecting and segmenting snow avalanches from SAR data. Early efforts focused on the binary classification of SAR images, while recent advances have enabled pixel-level segmentation, providing greater accuracy and spatial resolution. A case study using Sentinel-1 SAR data demonstrates the effectiveness of deep learning models for avalanche segmentation, achieving superior results over traditional methods. We also present an extension of this work, testing recent state-of-the-art segmentation architectures on an expanded dataset of over 4,500 annotated SAR images. The best-performing model among those tested was applied for large-scale avalanche detection across the whole of Norway, revealing important spatial and temporal patterns over several winter seasons.

Figures (12)

• Figure 1: Observations of avalanche debris using satellite-based SAR data. (a) Radarsat-2 ultrafine mode backscatter image. (b) Sentinel-1A IW mode backscatter image, acquired three days later. (c) Oblique photographs showing the avalanche debris are outlined in green for reference. Source: eckerstorfer2016remote.
• Figure 2: Examples of slices classified by the CNN and an expert (1-6) as well as false detections (7-12) The percentages indicate the network’s confidence that the image is an avalanche (1-6) or that the image is not an avalanche (7-12). Source: kummervold2018avalanche.
• Figure 3: (a, b) the SAR features obtained from the difference in the VV and VH channels. (c) product VVVH of the squared differences. (d, e) slope and PAR feature maps. Only a small area ($1k \times 1k$ pixels) of the actual scene is depicted here. Source: bianchi2020snow.
• Figure 4: Distribution of the slope angle for avalanche and non-avalanche pixels. Source: bianchi2020snow.
• Figure 5: Computation of the potential angle of reach (PAR) $\tilde{\alpha}$. $\theta(x)$ denotes the angle between the horizontal and the line drawn from a point in a release zone, denoted $x$, to the point of interest. Source: bianchi2020snow.