Comment on egusphere-2024-498

Elisabeth D. Hafner, Theodora Kontogianni, Rodrigo Caye Daudt, Lucien Oberson, Jan Dirk Wegner, Konrad Schindler, Yves BühlerOpenalex Unknown 2024 · DOI · OpenAlex

Abstract For many safety-related applications such as hazard mapping or road management, well documented avalanche events are crucial. Nowadays, despite research into different directions, the available data is mostly restricted to isolated locations where it is collected by observers in the field. Webcams are getting more frequent in the Alps and beyond, capturing numerous avalanche prone slopes several times a day. To complement the knowledge about avalanche occurrences, we propose to make use of this webcam imagery for avalanche mapping. For humans, avalanches are relatively easy to identify, but the manual mapping of their outlines is time intensive. Therefore, we propose to support the mapping of avalanches in images with a learned segmentation model. In interactive avalanche segmentation (IAS), a user collaborates with a deep learning model to segment the avalanche outlines, taking advantage of human expert knowledge while keeping the effort low thanks to the model's ability to delineate avalanches. The human corrections to the prediction in the form of positive clicks on the avalanche or negative clicks on the background result in avalanche outlines of good quality with little effort. Relying on IAS, we extract avalanches from the images in a flexible and efficient manner, resulting in a 90 % time saving compared to conventional manual mapping. If mounted in a stable position, the camera can be georeferenced with a mono-photogrammetry tool, allowing for exact geolocation of the avalanche outlines and subsequent use in geographical information systems (GIS). In this way all avalanches mapped in an image can be imported into a designated database, making them available for the relevant safety-related applications. For imagery, we rely on current and archive data from webcams that cover the Dischma valley near Davos, Switzerland and capture an image every 30 minutes during daytime since the winter 2019. Our model and the associated mapping pipeline represent an important step forward towards continuous and precise avalanche documentation, complementing existing databases and thereby providing a better base for safety-critical decisions and planning in avalanche-prone mountain regions.

Introduction

Accurate documentation of avalanche events is vital for hazard mapping, road management, and validating simulation models. However, current data collection methods face significant bottlenecks. Field observations are limited to specific, accessible locations, while satellite imagery is often expensive, hindered by cloud cover, and lacks the temporal resolution to pinpoint the exact timing of an event.

Webcams offer a promising alternative, providing high-frequency updates (every 30 minutes) across the Alps. However, manually tracing avalanche outlines in these images is labor-intensive. This research introduces the Interactive Avalanche Segmentation (IAS) pipeline, a "human-in-the-loop" approach that makes webcam-based mapping efficient enough for large-scale, operational use.

Methods

The IAS model treats segmentation as a collaboration between human experts and deep learning. Unlike fully autonomous models that struggle with the complex textures of snow and debris, IAS leverages human intuition to guide the AI.

  • Backbone Architecture: The system utilizes a state-of-the-art HRNet+OCR (High-Resolution Net + Object Contextual Representations) model.
  • Interactive Clicks: A user provides "positive clicks" on the avalanche and "negative clicks" on the background. These clicks are encoded as discs and fed into the model to refine the predicted mask in real-time.
  • Refinement Loop: The model iteratively updates the avalanche boundary based on these corrections until a satisfactory outline is achieved.
  • Georeferencing: Once a 2D outline is created, a mono-photogrammetry tool projects these coordinates onto a Digital Elevation Model (DEM), allowing the avalanche to be exported directly into a Geographic Information System (GIS).

Datasets and Evaluation

The study utilized several distinct datasets to ensure model robustness:

  • SLF Dataset: Approximately 400 annotated avalanches from the Dischma Valley webcam network.
  • UIBK Dataset: Over 3,000 images featuring different perspectives and coarser annotations.
  • Out-of-Distribution Sets: "WebNew" and "GroundPic" datasets (including handheld mobile phone photos) were used to test how well the model handles unfamiliar viewpoints.

Key Results

The IAS pipeline demonstrates a significant leap in both efficiency and accuracy over traditional methods:

  • Massive Speedup: Mapping an avalanche with IAS takes seconds, resulting in a 90% reduction in time compared to conventional manual tracing (roughly 15 seconds vs. 2.5 minutes per avalanche).
  • High Precision: The best-performing model (AvaWeb) achieved an F1 score of 0.97 on the UIBK test set with just one click, significantly outperforming previous fully automated benchmarks (0.64).
  • Low Effort: To reach an Intersection over Union (IoU) of 0.90—indicating near-perfect agreement with ground truth—the model required an average of only 3.31 clicks.
  • Consistency: A user study with eight participants showed that final masks were remarkably consistent (within 5% IoU of each other), suggesting the tool effectively reduces the subjectivity inherent in manual mapping.

Discussion and Limitations

While the IAS pipeline represents a major advancement, certain challenges remain. The model still struggles with images captured under diffuse light or when avalanches are partially shaded, as these features are difficult to distinguish from surrounding snow. Furthermore, the current georeferencing pipeline relies on cameras being in stable, fixed positions.

Future work aims to expand this into a mobile pipeline. This would allow backcountry users to take photos of avalanches on mobile devices, which could then be automatically georeferenced using GPS and orientation data, facilitating real-time uploads to national databases.

Conclusion

The IAS model and its associated mapping pipeline represent a vital step toward continuous and precise avalanche documentation. By combining ubiquitous webcam data with efficient AI-assisted mapping, this system provides a superior foundation for safety-critical decisions and planning in avalanche-prone regions.

Summary updated

Abstract

<strong class="journal-contentHeaderColor">Abstract.</strong> For many safety-related applications such as hazard mapping or road management, well documented avalanche events are crucial. Nowadays, despite research into different directions, the available data is mostly restricted to isolated locations where it is collected by observers in the field. Webcams are getting more frequent in the Alps and beyond, capturing numerous avalanche prone slopes several times a day. To complement the knowledge about avalanche occurrences, we propose to make use of this webcam imagery for avalanche mapping. For humans, avalanches are relatively easy to identify, but the manual mapping of their outlines is time intensive. Therefore, we propose to support the mapping of avalanches in images with a learned segmentation model. In interactive avalanche segmentation (IAS), a user collaborates with a deep learning model to segment the avalanche outlines, taking advantage of human expert knowledge while keeping the effort low thanks to the model's ability to delineate avalanches. The human corrections to the prediction in the form of positive clicks on the avalanche or negative clicks on the background result in avalanche outlines of good quality with little effort. Relying on IAS, we extract avalanches from the images in a flexible and efficient manner, resulting in a 90 % time saving compared to conventional manual mapping. If mounted in a stable position, the camera can be georeferenced with a mono-photogrammetry tool, allowing for exact geolocation of the avalanche outlines and subsequent use in geographical information systems (GIS). In this way all avalanches mapped in an image can be imported into a designated database, making them available for the relevant safety-related applications. For imagery, we rely on current and archive data from webcams that cover the Dischma valley near Davos, Switzerland and capture an image every 30 minutes during daytime since the winter 2019. Our model and the associated mapping pipeline represent an important step forward towards continuous and precise avalanche documentation, complementing existing databases and thereby providing a better base for safety-critical decisions and planning in avalanche-prone mountain regions.
BibTeX
@inproceedings{hafner2024comment,
  author = {Elisabeth D. Hafner and Theodora Kontogianni and Rodrigo Caye Daudt and Lucien Oberson and Jan Dirk Wegner and Konrad Schindler and Yves Bühler},
  title = {{Comment on egusphere-2024-498}},
  booktitle = {Openalex Unknown},
  year = {2024},
  doi = {10.5194/egusphere-2024-498-rc1},
}
Figure