Time-Lapse Photogrammetry of Distributed Snow Depth During Snowmelt

Abstract

Characterizing snowmelt both spatially and temporally from in situ observation remains a challenge. Available sensors (i.e., sonic ranger, lidar, airborne photogrammetry) provide either time series of local point measurements or sporadic surveys covering larger areas. We propose a methodology to recover from a minimum of three synchronized time-lapse cameras changes in snow depth and snow cover extent over area smaller or equivalent to 0.12 km2. Our method uses photogrammetry to compute point clouds from a set of three or more images and automatically repeat this task for the entire time series. The challenges were (1) finding an optimal experimental setup deployable in the field, (2) estimating the error associated with this technique, and (3) being able to minimize the input of manual work in the data processing pipeline. Developed and tested in the field in Finse, Norway, over 1 month during the 2018 melt season, we estimated a median melt of 2.12 ± 0.48 m derived from three cameras 1.2 km away from the region of interest. The closest weather station recorded 1.94 m of melt. Other parameters like snow cover extent and duration could be estimated over a 300 × 400m region. The software is open source and applicable to a broader range of geomorphologic processes like glacier dynamic, snow accumulation, or any other processes of surface deformation, with the conditions of (1) having fixed visible points within the area of interest and (2) resolving sufficient surface textures in the photographs.