Regional snow-avalanche detection using object-based image analysis of near-infrared aerial imagery

Korzeniowska, K.; Korzeniowska, K.; Bühler, Y.; Marty, M.; Korup, O. · 2017 · DOAJ

DOI: 10.5194/nhess-17-1823-2017

archive: archived pipeline: cataloged verified

Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)

Summary

This study addresses the challenge of systematically mapping snow avalanches over large regions to improve hazard forecasting and inventory completeness. Manual mapping is time-consuming and biased toward accessible or damaging events, leading to under-reporting. The authors propose an automated method using Object-Based Image Analysis (OBIA) on high-resolution near-infrared (NIR) aerial imagery to detect and classify avalanche release zones, tracks, and run-out zones. The goal was to develop a transferable algorithm that relies on spectral properties rather than just brightness, which can be ambiguous in snowy environments. The researchers utilized 0.25 m resolution NIR aerial images acquired by an ADS80-SH92 sensor in the Davos region of Switzerland during the winter of 2012–2013. The OBIA workflow, implemented in eCognition software, segmented the imagery and classified pixels based on brightness, the Normalized Difference Vegetation Index (NDVI), the Normalized Difference Water Index (NDWI), and the standard deviation of NDWI (SDNDWI). The algorithm distinguished vegetation, dark objects, and snow, then identified "rough snow" indicative of avalanche debris using SDNDWI thresholds. The method was trained on three 4 km² test sites and validated against manually digitized reference polygons. It was then applied to a larger 226.3 km² area to assess scalability. The automated classification achieved high accuracy on the training sites, with user’s accuracy exceeding 0.9 and Cohen’s kappa values between 0.79 and 0.85. When applied to the larger test area, the method yielded a user’s accuracy of 0.78 and a producer’s accuracy of 0.61, with a Cohen’s kappa of 0.67. The algorithm correctly identified 78.7% of the total reference avalanche area. Run-out zones were detected most reliably, while release zones proved more difficult to distinguish. Analysis of topographic factors revealed that most avalanches occurred between 1900 and 2600 m elevation on slopes of 20–40°, predominantly on north-eastern and south-western aspects. The study also demonstrated that detection accuracy correlated with the brightness and roughness (SDNDWI) of the avalanche deposits, but not significantly with shape or size. The findings indicate that OBIA using NIR imagery is a viable tool for large-scale avalanche mapping, offering a systematic alternative to manual methods. The use of normalized indices allows for broader applicability across different regions. However, the lower producer’s accuracy on the large scale suggests that older or less distinct avalanches may be missed. The method provides valuable data for updating avalanche cadastres and validating hazard models, though further refinement is needed to improve the detection of release zones and older deposits.

Provenance

The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed.

Summary generated by qwen3.6-27b-prismaquant on 2026-06-26; verification: verified.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

• naturalistic crash near crash