The pre-processing performed on the available imagery is focused on checking metadata associated with the image file, looking at the image size and the pixel resolution. Such a preliminary survey can identify failure in the data corruption and hopefully some issue in the server accessibility. The second step is the analysis of RGB values and their image patterns with the major goal to identify images characterized by low illumination, as well as situation with intense cloud cover. Finally, additional lens interferences (rain drops or ice crusts) are discarded.

The monoplotting of the image on the surface require two key information about the camera orientation and the sensor specifications. The preliminary process consists of calibrating the time-lapse camera using a reference chessboard. This operation is based on having at least 6 images shot by the camera system with the reference chessboard in a close-range distance. Having different chessboard orientations in front of the camera is possible, using standard computer vision techniques [Zhang, 2000], to obtain the intrinsic lens features useful for correcting the lens distortion.

The orthorectification module is based on the geometrical correction of the perspective view. This step is implemented following [Corripio, 2004]. The available digital elevation model [Porter et al., 2022], with a 5 m spatial resolution and 1 m vertical resolution, provided several topographic points that were projected on the camera view. The effectiveness of the correction was estimated considering the available ground control points.

The Segmentation procedure is based on the Spectral Similarity approach, proposed by [Salzano et al., 2019]. The module is based on measuring the spectral variations in a 3D colour space where reference endmembers are a theoretical “white” snow and a theoretical “black” target. The parameters estimated in this vector system are the spectral angle defined by [Kruse et al., 1993] and the Euclidean distance [Jensen, 2015], respectively calculated considering white and black references. While the parameter based on the Spectral Similarity represents an independent spectral feature, the Euclidean distance of the vector can be defined as a brightness-dependent feature. The involvement of all the three-color components will support the increase of surface types that can be discriminated: snow, shadowed snow and not snow. The proposed approach was developed in the R programming environment [R Core Team, 2022].
The first step consists of rearranging the three-color components of each pixel into a new two-dimensional vector space, mathematically defined as follows: the spectral angle which represents the relative proportion of the three-pixel components in relationship to the reference composition. The angle varies from zero, which can be associated with a “flat” behaviour of colours (R = G = B), to π2, referring to a very dissimilar behaviour from the theoretical “white” reference. The spectral distance is conversely an estimation of the vector length in the RGB space. It can range from 0 (black) to 1.73 (white) and it can be associated with the Euclidean distance from a “black” reference RGB composition. While this parameter is sensitive to the brightness of colours, the spectral angle is invariant with brightness [Van der Meer, 2006]. The outcome of this step consists in the frequency counting of pixels considering the two spectral components with a 0.05 resolution. Furthermore, the total number of included pixels and the area included in the cluster perimeter can be estimated.
The second step of the procedure consists in discriminating clusters from the obtained frequency distribution, and a watershed algorithm [Vincent et al., 1991] can support this segmentation phase. Each cluster was fitted with a normal distribution to retrieve modes and deviations. If clusters are very close to each other, they can be combined in one larger group depending on their probability to be discriminated using the Mahalanobis distance. The criteria adopted for the definition of the cluster perimeter was based on the pixel frequency higher than the Poisson error of the adjacent pixel. The procedure for the delimitation of the cluster perimeter was implemented using a per-pixel method following [Seo et al., 2016].

The final step consists in the identification of the surface type (snow, not snow and shadowed snow). This step was defined observing the frequency distributions of pixels in the defined spectral space (Figure 1). It was possible to detect that snow covers were generally characterized by higher spectral angles and lower distance values than not-snow covers. Snowed centroids (defined by μΔ and μθ) were generally positioned where angles were higher than 0.9 and distances were lower than 0.1.

Furthermore, the range of cluster values (Δmax, Δmin, θmax and θmin) are characterized by short distance variations compared to angles in the case of snow-covered surfaces. From this point of view, clusters with limited perimeters and a high number of included pixels describe surfaces with homogeneous reflective behaviour, as expected for snow-covered surfaces. The second rule that can be considered includes clusters with limited perimeters and consistent number of included pixels. The optical behaviour of those clusters must be coupled to their centroid position that must have low spectral angles. These constraints describe, also in this case, clusters characterized by a homogeneous spectral behaviour coherent with a snow-covered surface. The third rule that completes the classification procedure consisted of estimating the range of Δ between the defined clusters in the image and on defining a threshold (TΔ) that discriminates snow and other surface types. Two situations can occur for defining clusters above the threshold as snow-covered surface: one with multiple clusters and one with a single polygon.
The identification of the snow clusters has been checked considering also another automated method, which is a linear classifier based on thresholding of the blue channel (BT), introduced by [Salvatori et al., 2011] in the Snow-noSnow software. The method is based on the frequency counting of the blue component, and its hardness is associated with the definition of snow-not-snow limit looking at increments in the blue-channel histogram. This method has been used in several studies and it has shown some limitations. The illuminating conditions, the surface roughness and the distance from the camera are critical issues that affect the performance on retrieving snowed covers [27].