Abstract. We explored the potentiality of time-lapse photography method to estimate the snow depth in boreal forested and alpine regions. Historically, the snow depth has been measured manually by rulers or snowboards, with a temporal resolution of once per day, and a time-consuming activity. In the last decades, ultrasonic and/or optical sensors have been developed to obtain automatic measurements with higher temporal resolution and accuracy, defining a network of sensors within each country. The Finnish Meteorological Institute Image processing tool (FMIPROT) is used to retrieve the snow depth from images of a snow stake on the ground collected by cameras. An “ad-hoc” algorithm based on the brightness difference between snowpack and stake’s markers has been developed. We illustrated three case studies (case study 1-Sodankylä Peatland, case study 2-Gressoney la Trinitè Dejola, and case study 3-Careser dam) to highlight potentialities and pitfalls of the method. The proposed method provides, respect to the existing methods, new possibilities and advantages in the estimation of snow depth, which can be summarized as follows: 1) retrieving the snow depth at high temporal resolution, and an accuracy comparable to the most common method (manual measurements); 2) errors or misclassifications can be identified simply with a visual observation of the images; 3) estimating the spatial variability of snow depth by placing more than one snow stake on the camera’s view; 4) concerning the well-known under catch problem of instrumental pluviometer, occurring especially in mountain regions, the snow water equivalent can be corrected using high-temporal digital images; 5) the method enables retrieval of snow depth in avalanche, dangerous and inaccessible sites, where there is in general a lack of data; 6) the method is cheap, reliable, flexible and easily extendible in different environments and applications. We analyzed cases in which this method can fail due to poor visibility conditions or obstruction on the camera’s view. Defining a simple procedure based on ensemble of simulations and a post processing correction we can reproduce a snow depth time series without biases. Root Mean Square Errors (RMSE) and Nash Sutcliffe Efficiency (NSE) are calculated for all three case studies comparing with both estimates from the FMIPROT and visual observations of images. For the case studies, we found NSE = 0.917 , 0.963, 0.916 respectively for Sodankylä, Gressoney and Careser. In terms of accuracy, the first case study gave better results (RMSE equal to 3.951 · 10−2 m, 5.242 · 10−2 m, 10.78 · 10−2 m, respectively). The worst performances occurred at Careser dam located at 2600 m a.s.l. where extreme weather conditions occur, strongly affecting the clarity of the images. For Sodankylä case study, we showed that the proposed method can improve the measurements obtained by a Campbell snow depth ultrasonic sensor. According to results, we provided also useful information about the proper geometrical configuration stake-camera and the related parameters, which allow to retrieve reliable snow depth time series.
Webcam network and image database for studies of phenological changes of vegetation and snow cover in Finland, image time series from 2014–2016