**1. Introduction**

The knowledge of the extent and location of snow cover is of key importance to enhance the understanding of the present and future climate, hydrological cycle, and ecological dynamics, at both local and global scales [1,2]. Indeed, snow-dominated regions serve as an active reservoir for water supply during the melting period [3,4], and the seasonal presence of snow cover significantly modulates

7

a surface energy balance because of its high albedo and thermal properties [5]. Therefore, information on the spatial and temporal distribution of snow cover is critical for several research purposes and operational applications [6]. However, the monitoring of a snow-covered area is generally hindered by the complex interactions among site-dependent factors, especially in mountainous and forested regions. Meteorological forcings (i.e., precipitation regime, average air temperature, solar radiation) [7–9] and local topography (i.e., elevation, slope orientation and mean aspect) [10–12] are the most explanatory variables affecting spatial and temporal variability and persistence of snow cover. The definition of the topographic control on snow distribution is made challenging by the presence of vegetation, which intercepts snowfall and impacts the intensity of meteorological forcings [13–15]. Furthermore, wind-induced erosion and deposition phenomena are the main control factors driving the snow's spatial redistribution [16,17]. In-situ automatic measurements provide continuous and direct observational data allowing the retrieval of a temporal evolution of snow cover. However, they are site-dependent, generally subjected to distortions (e.g., wind action, vegetation interactions), and they do not succeed in catching the spatial variability of snowpack due to the heterogeneity of both climate and terrain with respect to the network density [18,19]. The collection of in-situ measurements at a large scale necessarily faces a general widespread lack of instrumental records, especially for steep slopes and remote high-elevation areas, where harsh environmental conditions usually entail a high operating cost [20].

Among in-situ gauges, the time-lapse camera is renowned for being a cost-effective device to monitor many environmental variables for scientific purposes [21–25]. Several webcam networks are currently operational worldwide, such as the European phenology camera network (EUROPhen) [26], the PhenoCam Network [27], and MONIMET camera network [24]. Recently, a growing interest aims at using webcam photography to detect snow cover from digital images to monitor its variability in space and time, even though the use of these observations is restricted to limited spatial scales [12,28–34].

Remote sensing represents a suited and powerful tool to monitor snow properties at larger scales and to overcome the gradual decrease of the representativeness of the gauging network with the increasing altitude. Under specific conditions (e.g., day-time, absence of cloudiness) [35], the snow cover detection is relatively straightforward through satellite-based optical observations, because of the high albedo of snow with respect to most land surfaces and the higher near-infrared reflectance of most clouds compared to snow-covered surfaces [36,37]. As well as cloud cover, the vegetation can obstruct visible and infrared information about snow, especially where forest canopy protruding above the snowpack reduces the surface albedo [38] and partially or completely shades the underlying surface [39,40]. Nevertheless, since satellite-based data are indirect measurements of snow-related quantities, they require a quantitative understanding of their accuracy, mainly depending on the uncertainty in retrieval algorithms [37,41]. Therefore, the comprehensive validation of satellite snow products is of key importance to properly assess and quantify their reliability, to identify possible errors and to provide input for further improvements. Indeed, the availability of information on the quality control of remotely-sensed data is critically needed by the scientific community, as one of the main key criteria for the selection of the most proper dataset to be effectively used, according to the final purpose.

Numerous studies have addressed the validation of satellite snow products at local and global scale by assessing the accuracy of remotely-sensed observations against ground-based data, which is one of the most widely used validation procedures [42–54]. Lacking any available in-situ reference data, a common approach relies on a cross-sensor comparison among different satellite-derived snow products by assuming one of the analyzed datasets as the reference truth [55–58]. This approach is even necessary when assessing the accuracy of satellite-derived products of fractional snow cover (FSC) requiring spatially distributed observations of reference [33]. Even though currently there is no agreed-upon methodology to perform a cross-sensor comparison, the most commonly used approach assumes the high-resolution satellite imagery as the reference effective dataset to assess moderate-resolution remotely-sensed observations, since it is supposed to provide the most reliable

information on the actual snow cover [59]. Nowadays, the Sentinel-2 (S-2) mission of European Copernicus Earth Observation program provides high-resolution multispectral imagery with an operational short revisiting time (~5 days) and free, global and systematic availability. Because of its meaningful payload, several studies have already experienced the potentialities of S-2 data in different fields of application [60–65].

This study aims to investigate the potential use of S-2 data to assess the reliability of moderate-resolution products of snow extent and FSC, that are the snow-related quantities most commonly used as input for hydrologic, meteorological and climate modelling [2]. Indeed, H10—Snow detection (SN-OBS-1) and H12—Effective snow cover (SN-OBS-3) supplied by the EUMETSAT's H-SAF project are compared against S-2 imagery. The interest in H-SAF snow products is focused on investigating the potential of these datasets and their suitability for hydrological purposes [56]. Over past decades, operational H-SAF snow products have been continuously validated against ground-based snow measurements [53,56]. Even though the high-resolution imagery can be reasonably used to establish reliable ground truth, a finer spatial resolution does not necessarily entail a higher accuracy of the satellite product, since its accuracy strongly depends on the retrieval algorithm used to derive snow-related information. Furthermore, no existing study supplies detailed information on the accuracy of S-2 imagery in detecting snow. For these reasons, the study has therefore the dual objective of validating this high-resolution dataset against in-situ snow measurements and webcam photography in order to properly assess its consistency and to guarantee the reliability of the comparison analysis. Therefore, before addressing the cross-sensor comparison of snow satellite products, a comprehensive validation of S-2 data is performed against ground-based data. With the aim of testing and assessing the satellite snow products under different climatological and topographic conditions, three study areas located in Finland, the Italian Alps, and Turkey are analyzed.

After introducing the context of this study, its motivation and research purposes, the article consists of four main sections. Section 2 is focused on data collection through a comprehensive description of the analyzed remotely-sensed (i.e., S-2, H-SAF H10 and H12) and ground-based (i.e., snow depth measurements and webcam imagery) datasets. The selected case studies in Finland, Italy and Turkey are presented and characterized. In Section 3, the methodology is extensively explained into the details of the retrieval algorithms for the generation of satellite products. The procedures implemented within both the validation of S-2 imagery against in-situ data, and the comparison between S-2-based and moderate-resolution snow products are widely described and their main assumptions are discussed. Results are reported and assessed through several evaluation metrics in Section 4. Lastly, conclusions are outlined in Section 5.
