Multitemporal Accuracy and Precision Assessment of Unmanned Aerial System Photogrammetry for Slope-Scale Snow Depth Maps in Alpine Terrain

Pure Appl. Geophys.  2017 The Author(s) This article is an open access publication https://doi.org/10.1007/s00024-017-1748-y Pure and Applied Geophysics Multitemporal Accuracy and Precision Assessment of Unmanned Aerial System Photogrammetry for Slope-Scale Snow Depth Maps in Alpine Terrain MARC S. ADAMS,1 YVES BÜHLER,2 and REINHARD FROMM1 Abstract—Reliable and timely information on the spatio-temporal distribution of snow in alpine terrain plays an important role for a wide range of applications. Unmanned aerial system (UAS) photogrammetry is increasingly applied to cost-efficiently map the snow depth at very high resolution with flexible applicability. However, crucial questions regarding quality and repeatability of this technique are still under discussion. Here we present a multitemporal accuracy and precision assessment of UAS photogrammetry for snow depth mapping on the slope-scale. We mapped a 0.12 km2 large snow-covered study site, located in a high-alpine valley in Western Austria. 12 UAS flights were performed to acquire imagery at 0.05 m ground sampling distance in visible (VIS) and near-infrared (NIR) wavelengths with a modified commercial, off-the-shelf sensor mounted on a custom-built fixedwing UAS. The imagery was processed with structure-from-motion photogrammetry software to generate orthophotos, digital surface models (DSMs) and snow depth maps (SDMs). Accuracy of DSMs and SDMs were assessed with terrestrial laser scanning and manual snow depth probing, respectively. The results show that under good illumination conditions (study site in full sunlight), the DSMs and SDMs were acquired with an accuracy of B 0.25 and B 0.29 m (both at 1r), respectively. In case of poorly illuminated snow surfaces (study site shadowed), the NIR imagery provided higher accuracy (0.19 m; 0.23 m) than VIS imagery (0.49 m; 0.37 m). The precision of the UASSDMs was 0.04 m for a small, stable area and below 0.33 m for the whole study site (both at 1r). Key words: Unmanned aerial vehicles, terrestrial laser scanning, manual snow depth probing, digital surface models, validation, error. 1. Introduction The spatial distribution of snow depth in alpine environments is highly heterogeneous (Elder et al. 1 Department of Natural Hazards, Austrian Research Centre for Forests (BFW), Hofburg Rennweg 1, 6020 Innsbruck, Austria. E-mail: 2 WSL Institute for Snow and Avalanche Research SLF, Flüelastrasse 11, 7260 Davos Dorf, Switzerland. 1998). This is mainly owed to the complex interaction between alpine terrain and meteorological factors, such as precipitation and surface energy fluxes, as well as the redistribution of snow by wind, sloughing or avalanche activity (Cline et al. 1998; Elder et al. 1991). Area-wide approaches to determine snow depth [e.g., based on automatic weather station (AWS) data combined with medium-resolution satellite imagery (Foppa et al. 2007)] are not able to capture its high local variability (Ginzler et al. 2013). However, detailed information on slope-scale snow depth distribution plays an important role for many applications in snow science and practice, including numerical modelling of snow drift (Durand et al. 2005; Beyers et al. 2004), ecological studies on alpine flora and fauna (Bilodeau et al. 2013; Peng et al. 2010), planning avalanche hazard mitigation measures (Margreth and Romang 2010; Fuchs et al. 2007), avalanche forecasting and warning (Helbig et al. 2015; Vernay et al. 2015), avalanche event documentation, e.g., for hazard zone mapping (Holub and Fuchs 2009; Decaulne 2007), prediction and assessment of flood hazard resulting from snow melt (Painter et al. 2016; Schöber et al. 2014) or as an input for the optimisation of numerical simulation models in avalanche dynamics research (Fischer et al. 2015; Teich et al. 2014). Manually measuring this information in situ is labour-intensive, potentially hazardous or even impossible (Nolin 2010). Therefore, a wide range of terrestrial, airborne and spaceborne remote and close-range sensing techniques have been applied to retrieve digital surface models (DSMs)/snow depth maps (SDMs) at the slope-scale (Deems et al. 2013; Dietz et al. 2012; Rees 2006). One of the most recent techniques is unmanned aerial system (UAS) photogrammetry, M. S. Adams et al. which has quickly become a wide-spread method for geodata collection in different fields of earth science (Colomina and Molina 2014; Nex and Remondino 2013). This development has been fostered by the proliferation of easy-to-use UAS platforms and sensors, as well as recent progress in the field of computer vision [structure-from-motion (Koenderink and van Doorn 1991) and multi-view stereopsis (Furukawa and Ponce 2009)], considerably reducing requirements for photogrammetric processing of aerial imagery (Mancini et al. 2013). Despite some drawbacks (e.g., range limited to slope-scale, legal regulations, necessity for stable flight weather conditions), UAS photogrammetry offers many advantages over established techniques for snow depth mapping: compared to manned aircraft campaigns, UAS can acquire imagery at a much lower cost (e.g., for equipment, training, maintenance, operation) (Harder et al. 2016), higher operational flexibility (Vander Jagt et al. 2015), as well as higher flexibility and choice regarding the sensors’ spatial and radiometric resolution, including an option for UAS-based laser scanning (Whitehead and Hugenholtz 2014); compared to terrestrial laser scanning (TLS), UAS photogrammetry is more flexible regarding deployment in alpine terrain [high-accuracy UAS positioning or point cloud registration routines as presented by Miziński and Niedzielski (2017) make georeferencing targets obsolete] and it does not suffer the limitations of the line-of-sight due to acute viewing angles or occlusions (Marti et al. 2016; Harder et al. 2016). However, while the abovementioned techniques are well-established, their quality and repeatability well-known (Hartzell et al. 2015; Müller et al. 2014), crucial questions regarding the accuracy and precision of UAS-based snow depth mapping are still under discussion (Avanzi et al. 2017). Several contributions have recently been published, reporting on the application of UAS photogrammetry to snow depth mapping, using both multicopter and fixed-wing UAS. In all of these studies, the UAS results were validated with reference data including: i. Global navigation satellite system (GNSS) measurements of the snow surface and/or manual snow depth probing (MP) (Miziński and Niedzielski Pure Appl. Geophys. 2017; De Michele et al. 2016; Harder et al. 2016; Lendzioch et al. 2016; Bühler et al. 2016; Vander Jagt et al. 2015). ii. Very high resolution optical satellite imagery (Marti et al. 2016). iii. A large-frame aerial camera mounted on a manned aircraft (Boesch et al. 2016). iv. A multi station in scanning mode (Avanzi et al. 2017). However, all these assessments were made based on a comparatively smal (...truncated)