Automated avalanche mapping from SPOT 6/7 satellite imagery with deep learning: results, evaluation, potential and limitations

Abstract

Abstract. Spatially dense and continuous information on avalanche occurrences is crucial for numerous safety-related applications such as avalanche warning, hazard zoning, hazard mitigation measures, forestry, risk management and numerical simulations. This information is today still collected in a non-systematic way by observers in the field. Current research has explored the application of remote sensing technology to fill this information gap by providing spatially continuous information on avalanche occurrences over large regions. Previous investigations have confirmed the high potential of avalanche mapping from remotely sensed imagery to complement existing databases. Currently, the bottleneck for fast data provision from optical data is the time-consuming manual mapping. In our study we deploy a slightly adapted DeepLabV3+, a state-of-the-art deep learning model, to automatically identify and map avalanches in SPOT 6/7 imagery from 24 January 2018 and 16 January 2019. We relied on 24 778 manually annotated avalanche polygons split into geographically disjointed regions for training, validating and testing. Additionally, we investigate generalization ability by testing our best model configuration on SPOT 6/7 data from 6 January 2018 and comparing it to avalanches we manually annotated for that purpose. To assess the quality of the model results, we investigate the probability of detection (POD), the positive predictive value (PPV) and the F1 score. Additionally, we assessed the reproducibility of manually annotated avalanches in a small subset of our data. We achieved an average POD of 0.610, PPV of 0.668 and an F1 score of 0.625 in our test areas and found an F1 score in the same range for avalanche outlines annotated by different experts. Our model and approach are an important step towards a fast and comprehensive documentation of avalanche periods from optical satellite imagery in the future, complementing existing avalanche databases. This will have a large impact on safety-related applications, making mountain regions safer.

Automated avalanche mapping from SPOT 6/7 satellite imagery: results, evaluation, potential and limitations

“…Code and data availability. The manually mapped avalanche outlines from 24 January 2018 and 16 January 2019 used by us for training, testing and validation are available on EnviDat (Hafner andBühler, 2019, 2021). The code is available on GitHub: https:// github.com/aval-e/DeepLab4Avalanches.git (last access: 22 August 2022) and Zenodo: https://doi.org/10.5281/zenodo.7014498 (Barton and Hafner, 2022).…”

“…5). The area contains avalanches in the shade and in illuminated terrain, as well as all outline quality classes in the initial mappings (Hafner andBühler, 2019, 2021). The mapping experts did not see another mapping before having finished theirs.…”

“…In contrast to previous studies, our work is the first to attempt to use deep learning for the detection of avalanches in optical satellite data. This is of major importance as the largest avalanche mapping from remotely sensed imagery to date, with 24 778 single avalanche polygons (Bühler et al, 2019;Hafner andBühler, 2019, 2021), relied on optical SPOT 6/7 satellite imagery. Furthermore, there have been investigations with external data into the reliability and completeness of mappings from SPOT 6/7 .…”

Automated snow avalanche monitoring for Austria: State of the art and roadmap for future work

“…Compared to automated detection, manual detection however is very time consuming and may often be subjective. Hafner et al (2022) conducted a study on the interobserver variability by comparing expert agreement on manual avalanche mapping in SPOT6/7 optical data. They found that the expert agreement is considerably lower than expected.…”

“…In the recent study of Hafner et al (2022) an adapted DeepLabV3+, a state-of-the-art DL model, was applied to automatically detect and map avalanches in SPOT 6/7 satellite images from January 2018 and January 2019, which makes it the first attempt to apply a DL model to optical satellite data. For training, validation and testing of the model, a dataset comprising 24,778 manually annotated avalanche polygons was used.…”