**1. Introduction**

Changes in glacier mass balance are commonly used as indicators of global climate change [1]. However, contrary to central Europe or Scandinavia, regular glacier observations for most of Asia are sparse to very sparse [2]. One parameter contributing to the annual mass balance of glaciers is the amount of solid precipitation (snow) and the snow cover extent. Actually, most glaciers worldwide, rely on the input of solid snow to grow glacier ice [3]. To measure and monitor temporal and spatial changes of snow extent, remote sensing technology from space is considered as an optimum tool (e.g., [4]).

In particular, optical systems are commonly applied to map temporal and spatial changes in snow cover extent (SCE) (e.g., [5,6]). The lower limit of the SCE for glaciers or ice sheets is defined as transient snowline, a measure of the extent of the snow-covered area at "any instant, particularly during the ablation season" [7]. However, in mountainous areas, optical images are often of limited

71

suitability due to cloud coverage or illumination. Space-borne synthetic aperture radar (SAR) sensors provide data independent of prevailing weather and illumination.

Dry snow of up to few meters thickness is considered as transparent for SAR data at C-band [8] frequencies. However, in wet snow conditions, the high attenuation characteristics [9] of water and the significantly increased dielectric permittivity of melting snow (e.g., [10]) reduce the backscatter coefficient significantly in comparison to dry snow or snow free conditions. Numerous studies used this effect to monitor the extent of wet snow with SAR data for C-band (e.g., [4,8,9,11–14]). In opposition to optical remote sensing data, which are capable of monitoring the SCE, SAR systems are solely sensitive to the area being covered by wet snow, which restrict data to the wet snow covered area fraction (WSCAF) per glacier. Several studies used the extent of wet snow to derive the transient snowline on glaciers (e.g., [4,12–14]). The SCE has large impacts on the energy balance and, consequently, on the mass balance (B) and runoff of a glacier [8]. However, previous studies (e.g., [12,13]) mentioned that the discrimination of snow and firn for C-band SAR data is impossible for utilizing co-polarized channels. In consequence, these studies failed at monitoring the temporal evolution of the accumulation area ratio (AAR) once the transient snowline retreated above perennial firn areas. Such retreat occurs primarily in years of strong negative mass balances. To overcome this deficit, it is possible to support with optical data (e.g., [6,13]), which involves the named illumination and cloud problems. Utilizing high winter scenes to analyze for retained liquid water in firn as proposed by Brown [15] does not allow for discrimination of snow and perennial firn either. Hence, a robust method to derive the transient snowline from active microwave remote sensing data—which do not have the restrictions of optical sensors—is beneficial for increasing databases of ablation processes for alpine glaciers.

Several authors describe a direct relationship between annual AAR, the equilibrium line altitude (ELA) and B (e.g., summarized by [16]). The prerequisite to establish this relationship are long-term observations of AAR and B for each specific glacier or a general approximation for this relation. Once the relationships have been established and evaluated, reliable AAR estimates from SAR data enable predictions of B solely from remote sensing. Such data help to assess and quantify runoff from glacierized catchments as snow, firn and ice have different surface albedos and, hence, melt rates are varying [17].

This study introduces a two-step workflow for C-band SAR data enabling monitoring of the WSCAF. We analyzed for backscatter distributions of SAR scenes, which were acquired during wet snow conditions over entire elevation ranges. Multi-annual wet snow scenes allowed for correction of topography-related signal effects and the determination of a wet threshold. If areas classified as being wet fall below 50% of the entire glacier area, we further discriminated wet snow and firn by a subsequent threshold. For quality assessment of determined transient snowlines, we used visible and shortwave infrared data from Sentinel-2 (S2), Landsat-7 (L7) and Landsat-8 (L8) missions. Results of the minimum seasonal wet snow extent were used as annual AAR. For Vernagtferner and Hintereisferner, the thus identified relation between AAR and B were compared with the existing field observations.

#### **2. Materials and Methods**

#### *2.1. Study Area and Data*

For this study, we used SAR data for the Hinteres Ötztal, Tyrol, Austria acquired from January 2015 to October 2018. We included twelve individual glaciers within our analysis, namely: Gepatschferner (GPF), Guslarferner, Hintereisferner (HEF), Hintereiswände, Kesselwandferner (KWF), Rofenberg West and East, Vernagtferner (VF), Vernaglwandferner North and South and Weissseeferner (Figure 1). Details for all individual glaciers are listed in Table 1. All twelve glaciers differ strongly in size, elevation range, slope angle and exposition. Further details for all glaciers can be found in [18]. For simplicity reasons and to prevent subpixel analysis of the SAR data, we summarized all individual glaciers into three areas of interest (AOI) named after the largest glacier per AOI. Both Guslarferners, both Vernaglwandferners, Weissseeferner and Hintereiswände were grouped together with Gepatschferner into the GPF AOI. The HEF AOI consists of both Rofenberg glaciers and Hintereisferner (Figure 1) and VF consists just of Vernagtferner.

**Figure 1.** Study area Hinteres Ötztal with named glaciers. Area margins are color coded for the three areas of interest Vernagtferner (VF), Hintereisferner (HEF) and Gepatschferner (GPF). The background image is a Landsat-8 band 8 composite from September 2016. The red rectangle within the inset displays the location of the study area. Coordinates are given in UTM with datum WGS 1984.


**Table 1.** Glacier names and elevation, exposition as well as size for all individual glaciers observed in this study. Glacier data are taken from [18] with glacier margins from 2006.

The remote sensing data basis for this study consists of 82 Sentinel-1A (S1A) and -1B (S1B) scenes (Table A1), which were acquired in the Interferometric Wide Swath Mode (IW) with dual-polarization (VV/VH). Scenes cover an area of 250 km × 200 km with ground resolution of 10 m × 10 m. The absolute radiometric accuracy is given by 1 dB [19]. Further details are listed in Table 2. Not every envisaged date of acquisition was achieved in 2015–2018. In addition, right after the launch of S1A and S1B, data acquisitions are usually sparse (Table A1). However, in summer 2017 and 2018, we were able to download almost all theoretically possible S1 scenes with a return cycle of six days. In addition, we collected eight optical and near infrared imagery during the ablation seasons in almost cloud free conditions.

**Table 2.** Parameters of the acquired Sentinel-1 (S1) data. Values for the noise equivalent sigma zero (NESZ) were derived from [19]. The inclination angle of the respective orbits vary for locations within the areas of interest and consequently are given as approximate values.


To relate data interpretation from SAR scenes to prevailing meteorological conditions, we used continuous ablation measurements and related meteorological observations from the monitoring program of the Geodesy and Glaciology group of the Bavarian Academy of Sciences, Munich, Germany [20]. On Vernagtferner, at 2930 m a.s.l., a ventilated thermometer records air temperature and an ultrasonic ranger measures ablation and accumulation continuously. For this study, we made use of the hourly data set. We used the ultrasonic data to determine whether new snow per day occurred. In a first step, we calculated the mean ablation rate per season (1 June–1 September each year), which for all four years is strongly negative. Next, we calculated the diurnal trend in surface height from 6:00 UTC to 6:00 UTC the subsequent day and divided each diurnal trend by the respective ablation trend. To relate positive quotients to new snow events, we multiplied by −1 and set all negative trends to zero. This results in a simple solid precipitation index with an indication of increases in surface height per day (new snow event).

To compare mass balance projections with field data, we used the long-term records for mass balances for VF (since 1965) and HEF (since 1952) [21,22]. The glaciological mass balance method relies on stake observations at specific points [23]. Uncertainties for stake readings are rather small (<3–5 cm). However, measurements have to be interpolated in between the stakes. Depending on the number of stakes and interpolation techniques, errors for specific areas on the glaciers can become significantly larger, especially for ice margins or areas difficult or impossible to probe (crevasse zones, etc.). Zemp et al. [24] summarize numerous sources of errors for the the glaciological method and present literature on uncertainties of 100 mm w. e. [25] to 600 mm w.e. [26]. Zemp et al. [1] conclude that systematic errors in glaciological mass balance assessments are usually below 100 mm w.e., while random deviations per year can reach values of "a few hundred mm w.e.".

On 20 September 2018, we identified the transient snowline in situ over a length of about 300 m separating wet snow and firn. This in situ transect represents only a small part of VF. Locations of the transient snowline were recorded with a conventional handheld GPS. We compared this GPS transect with derived snowlines from remote sensing data (SAR and optical results) by intersecting the respective lines and calculating for average offsets.

#### *2.2. SAR Data Workflow*

For the complex terrain in the investigated area, it is important to discriminate backscatter distributions for assumed homogeneous surface conditions from backscatter dependencies due to various surface conditions such as wet snow cover, firn cover or bare ice. Under the assumption of dry snow being transparent for SAR data in C-band frequency ranges, the only period of homogeneous glacier surface conditions are after melt affected upper elevation ranges and before the lowermost glacier parts become snow free. Otherwise, contributing surfaces have different dielectric permittivities (dry, wet snow, ice) and highly variable surface roughnesses. For the Alpine region discussed here, complete wet snow coverage occurs usually in June each year followed by an upglacier retreat of the SCE during the summer. In the following, we describe data processing for all acquired SAR scenes including statistical analysis of homogeneous backscatter scenes. The workflow to derive transient snowlines from acquired SAR scenes is displayed in Figure 2.

**Figure 2.** Workflow for the classification of wet and dry snow and firn based on Sentinel-1 data. Grey shaded boxes indicate data processing.
