**2. Materials**

## *2.1. Satellite Datasets*

## 2.1.1. Sentinel-2 Imagery

S-2 is a polar-orbiting, multispectral high-resolution imaging mission of the European Space Agency (ESA) for land, ocean and atmospheric monitoring. With the aim of fulfilling revisit and coverage requirements and providing robust datasets, the constellation consists of two identical satellites, Sentinel-2A (S-2A) and Sentinel-2B (S-2B), which were launched on 23 June 2015 (operational in early 2016) and 7 March 2017, respectively. Since the twin satellites are in the same sun-synchronous orbit with a phase delay of 180◦, they guarantee an effective revisit time of 5 days at the equator and 2/3 days over mid-latitudes, with a 290-km swath width. Multi-Spectral Imager (MSI) instruments provide fine spatial resolution optical images (Figure 1) having 13 bands spanning from the visible and the near infrared to the shortwave infrared, covering wavelengths from 0.4 to 2.2 μm (Table 1) [66]. Depending on the spectral band, the spatial resolution varies from 10 to 60 m. Four visible and near-infrared (VNIR) bands at 10 m for optical measurement, four NIR bands at 20 m for vegetation red-edge, two shortwave infrared (SWIR) bands at 20 m for snow, ice, and cloud discrimination, three coarse bands at 60 m in the aerosol, water vapor, and cirrus domain designated for atmospheric correction [67,68]. However, it is noteworthy that S-2 does not have a thermal band, which is of key importance for cloud detection, as cloud pixels are much colder than clear-sky pixels [69]. By December 2015, the acquisition of S-2 Level-1C (L1C) top-of-atmosphere (TOA) reflectance data is available and currently also S-2 Level-2A (L2A) bottom-of-atmosphere (BOA) reflectance data product is available to the remote sensing community worldwide.

**Figure 1.** Sentinel-2 RGB image—Tile T32TLR (Italian Alps), 6 December 2017.


**Table 1.** Spatial resolution and central wavelength of Sentinel-2A and -2B spectral bands.

#### 2.1.2. H-SAF H10 Product

H-SAF H10 (SN-OBS-1) is a daily operational product of snow extent generated from the visible (VIS) and infrared (IR) radiometry of the Spinning Enhanced Visible and Infrared Imager (SEVIRI) instrument on board the geostationary Meteosat Second Generation (MSG) satellites. The high temporal resolution and wide aerial coverage of SEVIRI imagery make it highly suitable for snow-cover mapping, since cloud cover is continuously monitored. Indeed, the daily snow cover product is derived for a multi-temporal analysis of SEVIRI 15-min images, that are processed as new data are available to collect the largest possible number of cloud-free pixels. The sampling is performed at 3-km intervals, which degrade to ~5 km over Europe. The resulting daily map has a spatial coverage delimited

between longitude 25◦ W–45◦ E and latitude 25◦–75◦ N [56,70] and it consists of four different classes: snow, cloud, water and bare ground (Figure 2).

**Figure 2.** H-SAF H10 product (25◦ W–45◦ E, 25◦–75◦ N), 5 April 2017.

#### 2.1.3. H-SAF H12 Product

H-SAF H12 (SN-OBS-3) is a daily operational product of FSC based on the multi-channel analysis of the Advanced Very High Resolution Radiometer (AVHRR) on board National Oceanic and Atmospheric Administration (NOAA) and meteorological operational (MetOp) satellites. FSC is generated at pixel resolution by exploiting the brightness intensity, which is the convolution of the snow signal and the fraction of snow within the pixel.

The sampling is carried out at 1 km intervals over the same H-SAF area of H10 product. The thematic map includes cloud and water classes, and percentage classes of fraction snow cover ranging from 0% (i.e., snow-free condition) to 100% (i.e., full snow cover) (Figure 3).

**Figure 3.** H-SAF H12 product quick-look image (25◦ W–45◦ E, 25◦–75◦ N), 5 April 2017.

#### *2.2. Test Sites and Data Collection*

With the aim of properly investigating the reliability of satellite snow products and their consistency under different topographical conditions (i.e., mountainous and flat areas) and vegetation cover, this study includes three case studies located in Finland, the Italian Alps and Turkey (Figures 4–6). In each country, eight S-2 tiles of interest have been selected to properly ensure a sizeable sample of satellite images. The selection of S-2 tiles has targeted those providing significant datasets over the analyzed period by minimizing possible overlapping. Furthermore, when selecting S-2 tiles the location of in-situ monitoring instruments has been considered to allow the validation of S-2 imagery against ground-based data. It is noteworthy to consider that in this study both snow extent and FSC are referred to the snow cover viewable over the satellite field of view, and not at the ground level. Because of the significant impact of vegetation on satellite snow detection, ancillary information on the vegetation cover of each S-2 tile has been derived from ESA GlobCover 2009 land cover map to support the assessment of the comparison results. This GlobCover map is derived from an automated classification of the Medium Resolution Imaging Spectrometer Full Resolution (MERIS FR) time series and it consists of 22 land-cover classes at a 300 m spatial resolution. Among the classes of natural and semi-natural terrestrial vegetation, two main categories have been defined (Table 2). The first main category embraces the vegetation classes having the highest impact on snow detection (V\_1) (i.e., evergreen or semi-deciduous forest), while the second one includes those having a lower impact (V\_2) (i.e., deciduous forest).

The selected tiles are reported in Table 3, where the percentage values of the main vegetation categories are reported according to the GlobCover 300 m land cover map.



**Figure 4.** Selection of S-2 tiles over Italian Alps.

**Figure 5.** Selection of S-2 tiles in Finland.

**Figure 6.** Selection of S-2 tiles in Turkey.

The analysis period extends throughout two winter seasons, namely 2016/2017 and 2017/2018. With the aim of properly taking account of the local climatology, the duration of the snow season has been independently set from October to May (eight months) in Finland and over the Italian Alps, and from November to April (six months) in Turkey.


**Table 3.** Selection of S-2 tiles at each test site and characterization of vegetation cover according to GlobCover 2009 land cover map. Two main vegetation classes having high (V\_1) and medium (V\_2) impact on snow detection are reported.

Since this study is focused on assessing how satellite products succeed in detecting snow cover, cloud free scenes or scenes with minor cloud cover are primarily selected. Indeed, only S-2 images with cloud cover lower than 20% have been included in the analysis.

The resulting datasets of S-2 imagery for the analyzed case studies are reported in Table 4. It is noteworthy to consider that the effective number of S-2 images in the snow season 2017/2018 is significantly higher than in the previous one (Table 4), since S-2B data has become available in March 2017.

**Table 4.** Seasonal effective number of S-2 images at each test site.


Throughout the analyzed period, only one daily H10 image is missing during the first snow season and 7 images are missing in the second one. Likewise, H12 product is not available for 7 and 16 days in snow seasons 2016/17 and 2017/18, respectively.

## *2.3. Ground-Based Datasets*

The validation of S-2 imagery relies on both ground-based dataset of snow measurements in Turkey and digital observations in Finland and over the Italian Alps.

#### 2.3.1. In-Situ Webcam Imagery

In Finland and in Italy, in-situ webcam imagery has been used to assess the consistency of FSC maps based on S-2 data (S-2-derived FSC), which are derived by counting the number of S-2 snow pixels versus the total number of S-2 pixels over the camera FOV. Webcams have been selected according to two main criteria. The first constraint requires a sufficiently wide webcam field of view (FOV) enabling the comparison with S-2-derived FSC. Secondly, webcams providing a properly representative dataset of observations have been primarily selected. With the aim of complying with these conditions, five webcams have been selected (Table 5), only one of which located over Italian Alps, mainly due to the complex topography, which strongly limits the extent of the webcam FOV.

The four cameras selected in Northern Finland are part of the camera network deployed in the frame of the MONIMET project [24]. MONIMET monitoring network consists of 28 cameras in 14 locations in Finland. The images are free and open. Those cameras produce images at each half an hour during daytime. For the study, midday time images are used since the snow cover does not change significantly during the day. One of the cameras is located in Kenttärova looking over a large evergreen spruce forest, another one is located in Lomppolojankka, a peatland site, and the other two in Sodankylä, located in a Scots pine ecosystem and in a wetland site, respectively. The FOVs of those cameras is shown in Figure 7a–d.

The webcam located in Aosta Valley (north-western Italian Alps) is at the experimental site of Torgnon, which belongs to the Phenocam network [71]. The camera is pointed north and it looks over grassland with mountains visible at distance. Camera images are provided every hour from 10 a.m. to 4 p.m. [21]. The FOV of the camera is shown in Figure 7e.

(**a**) (**b**)

(**c**) (**d**)

**Figure 7.** Webcam field of views: (**a**) Kenttärova canopy camera, (**b**) Lompolojankka peatland camera, (**c**) Sodankylä canopy camera, (**d**) Sodankylä peatland camera, (**e**) Torgnon camera.

(**e**)


**Table 5.** Selected in-situ cameras in Finland and Italy.

#### 2.3.2. In-Situ Snow Measurements

In Turkey, binary snow maps derived from S-2 imagery have been validated against ground-based measurements for the winter season 2017/18. Snow data from automatic weather stations (i.e., AWOS: Automated Weather Observing System, and SPA: Snow Pack Analyser) operated by Turkish state meteorological service (TSMS) have been used. Daily snow depth (SD) values have been obtained by processing and filtering the raw data supplied by these stations (e.g., removal of possible false snow detection due to grass). This analysis relies on SD measurements provided by 75 ground stations and 205 S-2 images available between November 2017 and April 2018 over Turkey. The validation has been performed over 25 S-2 tiles and 286 in-situ SD observations have been analyzed. Relative positions of S-2 tiles and the ground stations are shown in Figure 8.

**Figure 8.** Locations of ground-based monitoring stations in S-2 tiles.
