Item: BINARY AVALANCHE DETECTION SYSTEM (BADS)
-
-
Title: BINARY AVALANCHE DETECTION SYSTEM (BADS)
Proceedings: Proceedings, International Snow Science Workshop, Whistler, BC, Canada, 2026
Authors:
- Michael Dvorsak [ Alpine Infrastructure, Driggs, ID, US ]
- William Tidd [ Flat Efficient Engineering, Bozeman, MT, US ]
- Christopher Casebeer [ NextLogical, Phoenix, AZ, US ]
- Eric Bressler [ Alpine Infrastructure, Driggs, ID, US ]
Date: 2026-09-28
Abstract: Detecting avalanche events in real-time remains a challenge for ski resorts, transportation agencies, railways, and avalanche mitigation professionals. Existing solutions are often cost prohibitive, labor intensive, or limited in coverage. This paper presents the Binary Avalanche Detection System (BADS): a low-cost, scalable sensor network designed to provide binary (yes/no) real-time detection of avalanche and shock events, i.e. explosive detonations from avalanche mitigation work. The core innovation and design elements of BADS is a bespoke enclosure, a pressure sensor and amplifier, and algorithm chain for detection. The enclosure features an acoustic ear with a custom membrane and orifice to modulate air flow. The customized pneumatic filter integrated into the housing rejects ambient pressure changes while capturing pressure transients characteristic of avalanches and shocks (Marcillo et al. 2012; Anderson et al. 2018). Field deployments during the 2025-2026 season at Kirkwood Mountain Resort (CA), Powder Mountain (UT), and Jackson Hole Mountain Resort (WY) yielded 132 corroborated events: 33 confirmed avalanches and 99 confirmed shocks. Avalanche events produced characteristic pressure signatures within a 30 second window, while shocks produced much shorter, higher frequency signatures. Wind noise, the primary interference source, has proven distinguishable in most scenarios. A machine learning (ML) algorithm trained on the confirmed events is in active development and has demonstrated promising early detection performance. Two deployment architectures have been tested and validated: 1) a wired Power-over-Ethernet (PoE) network with nodes spaced up to 600 meters apart, and: 2) a wireless system capable of transmitting over/around ridgelines and from under 3+ meters of snowpack at distances exceeding one mile. The wireless architecture utilizes standalone nodes running local inference and transmitting alerts to a solar-powered base station with cellular uplink. For the 2026-2027 season, a 75 node deployment is planned across the northwestern US and Alaska with nodes installed beneath remote avalanche control systems (RACS) and in well-documented natural slide paths. This expanded dataset will be used to refine the detection algorithm and explore correlation between pressure spike duration and avalanche runout distance and footprint. BADS represents a practical, affordable path toward widespread automated avalanche detection for operational safety applications.
Object ID: ISSW2026_P2.26.pdf
DOI: https://doi.org/10.15788/1790098950
Language of Article: English
Presenter(s): Michael Dvorsak
Keywords: avalanche detection, differential pressure sensor, remote sensing, avalanche mitigation, machine learning
Page Number(s): 957 - 961
-