Snow depth time series retrieval by time-lapse photography: Finnish and Italian case studies

• Published in: The Cryosphere
• Main Authors: Bongio M., Nadir Arslan A., Melih Tanis C., De Michele C.
• Other Authors: Bongio, M., Nadir Arslan, A., Melih Tanis, C., De Michele, C.
• Format: Article in Journal/Newspaper
• Language: English
• Published: 2021
• Subjects: Nash; Sodankylä; Sutcliffe
• Online Access: http://hdl.handle.net/11311/1172509; https://doi.org/10.5194/tc-15-369-2021

The capability of time-lapse photography to retrieve snow depth time series was tested. Historically, snow depth has been measured manually by rulers, with a temporal resolution of once per day, and it is a time-consuming activity. In the last few decades, ultrasonic and/or optical sensors have been developed to obtain automatic and regular measurements with higher temporal resolution and accuracy. The Finnish Meteorological Institute Image Processing Toolbox (FMIPROT) has been used to retrieve the snow depth time series from camera images of a snow stake on the ground by implementing an algorithm based on the brightness difference and contour detection. Three case studies have been illustrated to highlight potentialities and pitfalls of time-lapse photography in retrieving the snow depth time series: Sodankylä peatland, a boreal forested site in Finland, and Gressoney-La-Trinité Dejola and Careser Dam, two alpine sites in Italy. This study presents new possibilities and advantages in the retrieval of snow depth in general and snow depth time series specifically, which can be summarized as follows: (1) high temporal resolution - hourly or sub-hourly time series, depending on the camera's scan rate; (2) high accuracy levels - comparable to the most common method (manual measurements); (3) reliability and visual identification of errors or misclassifications; (4) low-cost solution; and (5) remote sensing technique - can be easily extended in remote and dangerous areas. The proper geometrical configuration between camera and stake, highlighting the main characteristics which each single component must have, has been proposed. Root mean square errors (RMSEs) and Nash-Sutcliffe efficiencies (NSEs) were calculated for all three case studies comparing with estimates from both the FMIPROT and visual inspection of images directly. The NSE values were 0.917, 0.963 and 0.916, while RMSEs were 0.039, 0.052 and 0.108 m for Sodankylä, Gressoney and Careser, respectively. In terms of accuracy, the Sodankylä case study gave ...