Interpretation of SAR backscatter images for glacier mapping purposes can be challenging since several factors and processes on and within the subsurface affect the phase and magnitude of the SAR signals. The snow, firn and ice are dependent on external and surface conditions and can result in similar backscatter values. Therefore, additional data, or comparison datasets, were used to check the quality and reliability of the SAR time series. Meteorological data, temperature and precipitation, were downloaded from the Norwegian Meteorological Institute (eKlima.no, 2016) for the Ny-Ålesund meteorological station (Svalbard) and Juvvasshøe meteorological station (mainland Norway) (Figs. A1 and A2). Both stations are located close to the three glaciers used for detailed analysis of the backscatter time series, namely Hellstugubreen (within ∼ 12 km and station is located 1894 m a.s.l.), Kongsvegen and Holtedahlfonna (within ∼ 20 km and station is located 8 m a.s.l.). We have used existing glacier outlines from both Svalbard (Nuth et al., 2013) and mainland Norway (Andreassen et al., 2008; Paul et al., 2011) as verification and in support of the analysis. Updated glacier outlines for the Norwegian glaciers are available (Andreassen et al., 2016). The discrepancies between new and old outlines are minor and do not influence the results presented in this paper.

The end of summer snow line (EOSS) is an approximation of the ELA (Østrem, 1975; Rabatel et al., 2013). In regions with a significant superimposed ice zone, the EOSS will often be above the ELA (Cogley et al., 2011). It can also be lower than the ELA in years when there was a low ELA in the previous year. In this application scenario, SAR backscatter data from Kongsvegen were used to map glacier surface melt and to track TSLs on Hellstugubreen in Norway (see Supplement). Optical satellite data, in situ observations of EOSS and ELA derived from SMB measurements have been used for validation. The EOSS can be used to reconstruct annual mass balance series regionally due to the strong relationship between the EOSS and ELA (e.g., Demuth and Pietroniro, 1999; Pelto, 2011). EOSS and TSLs have been used to determine glacier mass balance using modeling assessments (e.g., Rabatel et al., 2005; Huss et al., 2013; Hulth et al., 2013). The amount of usable images for reconstruction of annual mass balance based on EOSS is more predictable with SAR imagery compared to using optical satellite data due to a reliable repeat passes of the Sentinel-1 satellites and transparent clouds.

Wet snow absorbs most of the microwave signal, returning little energy back to the sensor (Stiles and Ulaby, 1980). The roughness of snow or ice and incidence angle of the SAR satellite sensor also affects the backscatter signal strength (Shi and Dozier, 1995, Fig. 4). In both the Sentinel-1A and RADARSAT-2 backscatter time series an abrupt seasonal change was found, from warm and wet conditions in the end of the melt season to the onset of the cold season with dry and cold conditions (corresponding to no.1 in Fig. 2). These backscatter differences are present due to a change in weather conditions between 26 August 2015 (DOY 238) and 7 September 2015 (DOY 250), causing an increase of ∼ -20 to ∼ -7 dB (Fig. 2). A temperature record from the closest weather station in Ny-Ålesund (8 m a.s.l.) shows lower temperature in the period between the images, indicating even colder temperatures on the glaciers since they are located on higher elevation (Fig. A2). In addition, an optical image from 9 September 2015 (DOY 252) over Kongsvegen shows significant precipitation as snow (Table A2), which also indicates conditions < 0 ∘C on the glacier. The snow line, where the onset of cold season starts, corresponds to number 8 in Fig. 2, representing the EOSS. The onset of surface melt is characterized by a rapid decrease in backscatter values (corresponding to no. 6 in Fig. 2; Stiles and Ulaby, 1980; Smith et al., 1997; Wolken et al., 2009), most likely introduced by warm and wet weather conditions (Rotschky et al., 2011). The lowest backscatter values in the ablation zone are found when wet snow covers the glacier (corresponding to no. 9 and the blue color in Fig. 2), which also reflects the length of the melt season. This has until now been mostly studied using QuikSCAT data (a Ku-band sensor) with daily temporal resolution, but low spatial resolution (e.g., Rotschky et al., 2011). It is difficult to define the melt season accurately in time from RADARSAT-2 images due to the low repeat time of 24 days (Fig. 5). Sentinel-1 provides 6-day repeat time with two satellite sensors and can be used to study the length of the melt season more accurately in terms of spatiotemporal resolution.

The TSL migration up-glacier is often correlated with temperature rise and topography (e.g., Hall et al., 2000; corresponding to no.7 in Fig. 2). It is also possible to track the wetness in the snow up-glacier with the surface dry-to-wet seasonal snow line (corresponding to no. 10 in Fig. 2), but here we have focused on the TSL. During the melt season, seasonal snow melts away exposing a rough ice or firn surface, thus creating a sharp contrast to the smooth and wet snow. Based on similar observations by Hall et al. (2000), we believe this can be observed from SAR backscatter due to different roughness lengths between the two targets causing higher scattering of the ice surface (Fig. 4). We have used this difference in roughness to track the TSL within the melt season. When tracking the TSLs within a melt season, the few optical images available can be gap-filled with SAR data (Fig. 3a, b). TSLs from optical and SAR images acquired almost on the same day were manually selected from the glaciers (Table A2). Heat plots, a DEM, SAR backscatter and optical satellite images were used for visual inspection of the TSL positions in the analysis (see also temperature plot in Fig. A1). Using mean backscatter in elevation zones of 25 m gives a location of the TSL representing the width of the glacier compared to using a centerline representing one point on the glacier. The same visual interpretation was used to retrieve the EOSS from SAR acquisitions on four of the glaciers in 2015 and 2016 (Hellstugubreen, Storbreen, Nigardsbreen and Austdalsbreen). On Gråsubreen we detected a melt regime which did not correlate with elevation, as the snowmelt was first apparent in the convex areas higher up on the glacier and not in the lower part where the snow accumulates (most likely due to the local topography and wind patterns). On such glaciers, satellite sensors other than Sentinel-1A and B and their according dense time series are not able to reveal the same level of temporal detail in melt patterns.

Strong correlation was found between TSLs from Sentinel-2 and Landsat 8 satellite images and TSLs from Sentinel-1 images (Fig. 3a, b, c). In the melt season with warm and wet conditions, the backscatter signal after snow events might be reduced on the glacier. They can cause, firstly, a smoother surface compared to the underlying rougher ice surface and, secondly, high absorption and forward scatter of the radar waves as the snow is wet (Rott, 1984). An example of this was when lower backscatter values were observed in the ablation area of Hellstugubreen on 3 October 2016 (DOY 277), most likely due to wet snow lowering the roughness and absorbing microwave energy (Fig. 3a). Snow was observed in an optical image on 6 October 2016 (DOY 280) on Hellstugubreen (not shown). In addition, another example from Kongsvegen showed a snow event causing disagreement between the Landsat 8 image and Sentinel-1 images (see outlier in Fig. 4c). This snow event must have happened between the compared acquisition dates of the optical and SAR satellite images (i.e., between 7 and 9 September 2015, DOY 250 to 252; Table A2).

The EOSS and ELA in Fig. 3d reflect the maritime–continental transect of glaciers in southern Norway, as the most continental glaciers have the highest ELA. EOSS positions from SAR backscatter data correlated well with ELA calculated from the SMB curves that were derived from the in situ measurements, in addition to direct in situ measurements of the EOSS (Fig. 3d). This indicates that EOSS derived from Sentinel-1A and B SAR data may be used as input when reconstructing annual mass balance series. We believe Austdalsbreen (A) is an outlier and that the signal we get in the Sentinel-1 image (∼ 1500 m a.s.l.) is related to the firn line from the previous year (ELA from SMB in 2015 was 1371 m a.s.l.), as little seasonal snow was left in 2016 and ELA was “above” the glacier (ELA from SMB in 2016 was > 1747). The maximum discrepancy between a field observation and a SAR acquisition was 21 days (SAR acquisition 3 October 2016 (DOY 277) and in situ on 12 September 2016 (DOY 256) on Hellstugubreen) (Fig. 3d).

In both study regions, we found a discrepancy between the datasets in that the TSLs from SAR generally show a mean of 21 m lower altitude compared to optically derived TSLs (not taking the outlier into consideration; Fig. 3c). On Kongsvegen, as the snow line retreats up into the superimposed ice or firn area, the roughness difference is less compared with glacier ice, making it more difficult to retrieve a correct TSL (Figs. 5 and 6a). The same applies to optical satellite images as it is challenging to derive the TSL when superimposed ice is present (e.g., Winther, 1993; Kundu and Chakraborty, 2015).

The glacier geometry, elevation, size, sun exposure and local snow accumulation influence the melt pattern of seasonal snow on glaciers, and thus the EOSS. If the glacier spans a small elevation range, the TSL could be more difficult to trace and may be either above or below the elevation range that the glacier covers. In addition, interaction between snow lines and crevassed areas, especially in ice falls, can be a challenge on larger glaciers (e.g., Chinn, 1995). Extracting TSLs and EOSS using SAR backscatter data can be valuable for

refining spatial variations in the melt pattern of well-studied glaciers with already existing surface mass balance programs;

SAR backscatter imagery can be used to identify distinct zones of consistent backscatter that correspond to glacier facies (Fahnestock et al., 1993; Brown et al., 1999; König et al., 2002; Jaenicke et al., 2006; Langley et al., 2008). This is because the SAR backscatter is influenced by physical properties of ice and snow, as well as weather conditions and surface texture (e.g., Smith et al., 1997; Rau et al., 2000). Nonetheless, it is challenging to directly connect glacier facies to SAR classified glacier zones, as backscatter is a complex composite signal reflected from a surface volume, the material properties of which often vary temporally with external atmospheric forcings, e.g., with winter rain events (Sect. 5.5). Glacier facies are defined as properties on a glacier dividing one part of the glacier from others, often connected to mass balance processes (i.e., ablation and accumulation area), and are synonymous with the term glacier zones (Cogley et al., 2011). Here we use the term glacier facies for snow and ice ground properties and the term SAR glacier zones for the interpretations and classifications from SAR satellite imagery. SAR glacier zones often have an annual frequency but also vary seasonally as backscatter changes from being sensitive to surface properties in the melt season to volume properties in the cold season (Fig. 2). We define the following SAR glacier zones (corresponding to no. 5 in Fig. 2): (1) percolation zone (firn zone), (2) the superimposed ice (SI zone), (3) ice zone and (4) wet-snow zone (e.g., Cuffey and Paterson, 2010). The firn and SI zones are part of the accumulation area, and the ice zone is part of the ablation area. The wet-snow zone represents wet snow and firn in the melt season (Sect. 5.1), as well as rain events during the cold season (Sect. 5.5), and occurs in both ablation and accumulation areas. Note that our definition of the wet-snow zone differs from traditional definitions (e.g., Cogley et al., 2011; Cuffey and Paterson, 2010), as ours is based upon a SAR observational perspective.

In this application scenario, we present a time series of SAR glacier zones from 2009 to 2016 including RADARSAT-2 (24-day repeat) and Sentinel-1A (12-day repeat) images in northwest Svalbard. Previous studies have correlated distinct glacier facies with SAR glacier zones on Kongsvegen (Engeset et al., 2002; König et al., 2004; Brandt et al., 2008; Langley et al., 2008), and these facies also correspond to previous interpretations in literature (Benson, 1962; Rau et al., 2000; Cuffey and Paterson, 2010). Dry-snow glacier facies is absent on both Kongsvegen and Holtedahlfonna as melting occurs over the entire surface (Engeset et al., 2002; Langley et al., 2007), which agrees with our interpretation of the SAR backscatter time series (Fig. 2). The firn line does not vary much from year to year, but several years of negative mass balance will eventually migrate the firn line up-glacier, and vice versa with positive mass balance years (e.g., König et al., 2004; Brown, 2012). Wet snow typically has the lowest backscatter values, followed by wet and dry ice (here, these show similar values; Fig. 2), dry superimposed ice and dry snow and firn (Fahnestock et al., 1993). A dry-snow pack has low dielectric contrast, and SAR microwaves are volume scattered. In the firn area, ice lenses, pipes and layers act as randomly oriented dielectric cylinders, responsible for the high scattering of microwave signal back to the SAR sensor. We consider snow and firn to be Rayleigh scatterers when the particle size is lower than 10 % of the wavelength (λ=5.5 cm) or 0.55 cm and Mie scatterers up to 10 times the wavelength, or 55 cm (Woodhouse, 2006). The backscatter response of superimposed ice is dependent on air bubble content and size, where a high frequency of bubbles typically causes higher backscatter values (e.g., König et al., 2002; Langley et al., 2009). Thus, during the cold season, the superimposed ice zone has typically lower backscatter compared to firn, but higher than glacier ice (e.g., Langley et al., 2009). During the melt season, high backscatter on ice surfaces is caused by surface roughness rather than volume scattering (Shi and Dozier, 1995; Hall et al., 2000).

Stabilization of the SAR glacier zones between years in the cold season indicates a steady glacier facies area through time (Figs. 5 and 6a). In Fig. 2 we show three distinct SAR-zones on Kongsvegen in the cold season (7 September 2015 to 4 May 2016, DOY 250 to 125), located at elevation bins 0–500, 500–700 and 700–800 m. These SAR glacier zones mirror the glacier facies, ice, and SI and firn area, respectively (Langley et al., 2009). Thus, the RADARSAT-2 time series from 2009 to 2015 on Kongsvegen showed a relatively stable firn line altitude and a superimposed ice altitude (SIA), even though a retreat of SIA can be observed in 2011 and 2012 (Fig. 5). The winter SAR images are useful for identifying the superimposed ice zone since with optical satellite imagery it can be challenging to separate this zone from the bare ice zone (e.g., Kundu and Chakraborty, 2015). Surface snow, lacking large scatterers and dielectric contrasts, promotes low backscatter; glacier ice below the surface snow cover promotes surface scattering at the snow–ice interface and transmission within the ice volume. Together these scattering processes result in lower backscatter intensity (light blue to yellow colors in Fig. 6a) in the ablation area. We suggest that backscatter amplitude from ice surfaces changes seasonally due to surface roughness variations, as melting creates a rougher ice surface due to changes in local topography on the ice surface, e.g., microwater evacuation streams (Shi and Dozier, 1995; Hall et al., 2000) (Fig. 6a, see yellow to orange colors on ice in the melt season compared to the light blue color in the cold season). According to the modified Rayleigh criterion a surface is considered rough in the C-band SAR (using an incidence angle of 38∘) when the root-mean-square (RMS) surface height variation is more than 1.3 cm and considered a smooth surface when the RMS height variation is less than 0.2 cm (Lillesand et al., 2004). Drying of meltwater channels does not cause the backscatter to be higher in the cold season compared to the melt season because they cover only a very small percentage of the SAR pixels; therefore we believe the ablation effect (caused by melting and water on the ice surface) is responsible for a rougher surface and higher backscatter.

SAR glacier zones are less clear on Holtedahlfonna compared with Kongsvegen. Backscatter values are gradually increasing between 570 and 775 m, but no clear separation between zones was found (Fig. 7a). The lower part of Holtedahlfonna, Kronebreen, had several stripes of high backscatter (below 570 m in the heat plot in Fig. 7a), indicating highly crevassed areas due to double-bounce scattering of the physical corner reflector (Woodhouse, 2016). Although such corner reflectors probably blurred the seasonal signal, there was still seasonal variability with low backscatter values (blue colors) in early spring 2016 (around 28 May, DOY 149), indicating the onset of melt season (Fig. 7a).

Previously, one satellite image was used when mapping SAR glacier zones from SAR imagery. Here, we examined glacier zones on Kongsvegen and Holtedahlfonna thoroughly through time, using consecutive and dense time series of SAR imagery. This time series shows in particular that SAR glacier zones vary through time representing the changing surface conditions that ultimately control backscatter return. In this example, we show the potential to detect and study changes in transient glacier facies. Here, a stabilization of the SAR glacier zones in the cold season improves certainty about the designation of glacier facies.

For the first time, we present a comparison of Sentinel-1A backscatter time series with results from a SMB and firn pack evolution model. A coupled energy balance–multilayer firn model was used to simulate the evolution of temperature, density and water content in snow and firn. The model has previously been used to simulate the long-term (1961–2012) mass balance and firn evolution of Kongsvegen and Holtedahlfonna, as described in Van Pelt and Kohler (2015). For the experiment in this paper the model is forced with weather station data from the Ny-Ålesund weather station (provided by the Norwegian Meteorological Institute), which provided time series at sea level for temperature, precipitation, relative humidity, cloud cover and air pressure. Elevation lapse rates for temperature and precipitation were optimized to remove biases between modeled and observed seasonal mass balance. The precipitation lapse rate was optimized against winter balance data, and the temperature lapse rate was optimized against summer balance data. Lapse rates were determined individually for Kongsvegen and Holtedahlfonna, which is mainly relevant for precipitation, which is known to increase much faster with elevation on Kongsvegen than on Holtedahlfonna (Nuth et al., 2012). Validation of modeled subsurface density against shallow firn core observations on Kongsvegen and Holtedahlfonna has been discussed in Van Pelt and Kohler (2015), showing good agreement. There is no real validation for subsurface temperature and water content for these glaciers. However, the model performed well in simulating vertical temperatures at the top of Lomonosovfonna, as shown in Van Pelt et al. (2014). Model output contains daily depth-dependent fields with a 300 m horizontal spacing along centerline profiles on the glaciers. The sub-surface grid contains a total of 100 layers with layer thickness increasing with depth and ranging between 10 and 40 cm. Here, firn air content and the subsurface water content were compared with the SAR backscatter time series. Firn air content (in meters) represents the amount of empty pore space in a vertical column (in this study we used a 2 m column). Since ice has zero pore space, the firn air content is a measure of snow and firn depth and density (Fig. 6b). The subsurface water content (kg m-3) represents irreducible water at a specified depth. From the subsurface water content variable in the snow and firn, we have calculated depth (in meter) of the transition between wet (water content > 0) and dry snow and firn (water content = 0), indicating which depth has a water content above zero (Fig. 7b). This result was correlated with the SAR backscatter data.

In the start of the cold season, the ice surface was covered by seasonal snow had low firn air content (e.g., at 7 September 2015, DOY 250 in Fig. 6b). During the cold season the firn air content increased through time and moving up-glacier, as fresh snow accumulated (e.g., at 23 April 2016, DOY 113 in Fig. 6b). We observed high backscatter values (Fig. 6a) when firn air content was high (Fig. 6b), as the dry and cold conditions were found to enhance volume scattering. For each Sentinel-1A acquisition, we found a positive significant correlation between firn air content and backscatter in the cold season and a negative significant correlation in the melt season (Fig. 6c). To avoid assumptions of a linear relationship between firn air content and backscatter, we used the nonparametric Spearman's rank correlation coefficient. The reason for the negative correlation is the change of surface properties affecting the backscatter signal, as the volume scattering in the cold season is exchanged with the SAR sensitivity to wet snow and higher surface roughness in the melt season. The point in Fig. 6c on 1 September 2016 (DOY 245) showed no correlation, because it is located in the transition zone between the melt season and cold season, where multiple melt end freeze events happen within short time spans. The snow line was identified in the modeled data (Fig. 6b) and in the observed backscatter data (Sect. 5.1).

In the beginning of the cold season when the surface refreezes, backscatter values are sensitive to subsurface melt in the snow and firn pack (Ashcraft and Long, 2006; Rotschky et al., 2011). A spatiotemporal refreeze signal from the Sentinel-1A time series was observed in the firn area of both Kongsvegen and Holtedahlfonna (corresponding to no. 2 in Fig. 2). This was also observed as a weaker signal in the RADARSAT-2 time series just after the melt season each year (Figs. 5 and A3). Christianson et al. (2015) found a perennial firn aquifer containing liquid water in the upper accumulation area on Holtedahlfonna, and they argued that the firn aquifer had a depth below the surface of approximately 3.5–15 m. We found stabilizing backscatter intensity values in time during the winter period (Figs. 6a and 7a) and speculate that C-band SAR backscatter data do not receive backscatter responses from firn that deep as presented by Christianson et al. (2015). An aquifer, or wet conditions in the firn, causes absorption, limiting backscatter response from the depth of the firn volume (e.g., Ashcraft and Long, 2006). In the firn zone of the glaciers (> 650 m on Kongsvegen and > 775 m on Holtedahlfonna), we found a gradual increasing backscatter signal during the cold season from 7 September to 12 December 2015 (DOY 250 to 346) (Figs. 6a and 7a, ∼ -10 to ∼ -4 dB). This is most likely due to multiple scattering events, from volume scattering in the snow and firn, in addition to wetness in the firn that gradually refreezes due to the winter cold wave. This can reflect stored and percolated water in the firn from melt and rain events that refreeze over time, also releasing latent heat and possibly slowing down the process. Backscatter eventually stabilized (12 December 2015, DOY 346, in Fig. 7a) and showed similar high values throughout the rest of the cold season. The backscatter measured at the sensor antenna is the sum of multiple scattering occurrences throughout the firn volume. The penetration depth of radar waves varies over time and contributes to the backscatter, and the backscatter intensity is therefore not constant in time and represents the integral of depth through time. Using the modeled data, we were able to explore the penetration of SAR backscatter in snow and firn. It is likely that absorption with limited volume scattering of radar waves in wet snow and firn can give an indication of how deep the Sentinel-1A C-band SAR can penetrate due to a strong sensitivity to wet conditions. Using the model results we were able to estimate the depths of the intersection between the dry and wet zone in the firn pack (Fig. 7b) and compare this to the backscatter data (Fig. 7c–e). Results indicate increased penetration depths over time and then stabilization once the transition exceeds a certain depth where the radar waves do not reach it anymore. Similar trends between the transient modeled dry-to-wet transition depth in the firn and the backscatter time series for the uppermost part of the glacier were found (Fig. 7c). The deeper the dry-to-wet transition zone, the higher the SAR backscatter values (Fig. 7e). This correlation can thus be explained by a mix of volume scattering returns and radar waves sensitivity to wet conditions in snow and firn through time until January 2016.

To investigate the penetration depth further, values from black points in Fig. 7a were selected where the backscatter values stabilized in the firn. Values from the same points of dry-to-wet transition depths (Fig. 7b) gave an indication of the depth of the dry-to-wet zone in the backscatter data. The mean transition depth of the point values from Fig. 7b is 1.7 m when all nine points were included. When only including the upper six points with significant correlation > 920 m (showed in Fig. 7c and e), a mean transition depth of 2.0 m was found. We speculate that backscatter intensity cannot reflect small changes in snow depth during the dry wintertime due to high volume scattering, thus suggesting deeper and temporally constant penetration of the radar waves.

In this example, modeled data were used to help interpret time series of backscatter intensity data. This might be inverted in the future, as SAR backscatter data will be further understood and can potentially be used as refined modeling input. Such information is valuable in remote regions in the high Arctic that are lacking meteorological stations and where it is costly to do field observations.

High Arctic regions like Svalbard have encountered substantial warming the last decades, especially in the winter season with a temperature increase of 3.8 ∘C from the measured period 2001–2015 to the reference period 1971–2000 (Isaksen et al., 2016). Frequent warm winter weather events are occurring especially related to increasing numbers of melt and precipitation days in the mid-winter (Vikhamar-Schuler et al., 2016). Winter rain events are important to map in a glaciological context as it can be a substantial component to internal accumulation as it refreezes in the snow pack (e.g., Jansson et al., 2003; Van Pelt and Kohler, 2015; Van Pelt et al., 2016). The SAR backscatter contrast of ice and snow is directly dependent on the meteorological conditions before and at the time of acquisition of the satellite image. Rain on an ice surface gives stable backscatter values similar to dry ice, since glacier ice is still rough even when wet. Consequently, roughness is considered as a highly important scattering variable (e.g., Hall et al., 2000). Wet snow absorbs SAR waves, and little energy is transmitted back to the SAR sensor in comparison to dry snow and firn surfaces (Rott, 1984). During dry-snow conditions, the radar signals depend on the underlying firn or ice conditions (e.g., Brown et al., 2005). In this application scenario, we have examined a rain event on Kongsvegen and Holtedahlfonna from dense Sentinel-1A images during the cold season (corresponding to no. 3 in Fig. 2; see time series animation of Kongsvegen in Supplement). RADARSAT-2 ScanSAR satellite images were used to investigate the extent of the rain event.

On 5 January 2016 (DOY 5), Sentinel-1A acquired an image over the Kongsfjorden region indicating wet conditions on the upper part of Kongsvegen and its surroundings (corresponding to no. 3 in Fig. 2). The temperature and precipitation data from Ny-Ålesund showed a warm and wet weather event from 29 December 2015 (DOY 363) to 4 January 2016 (DOY 4) (Fig. A2 and Table A1). Figure 8a and b show backscatter images before and after the rain event, and Fig. 8c, which shows the difference between the two, indicates an area of wet conditions (blue color). Presumably, the extent of the rain event covered the entire surface of Kongsvegen. However, on the lower part on the glacier (the ice facies) this was hard to observe due to little snow and rough surface. A clearer signal was found further up-glacier with present seasonal snow on firn (white circle in Fig. 8b).

Wet conditions in a SAR image (indicated by low backscatter values in Fig. 8b) do not necessary mirror the weather conditions at the time of acquisition of the satellite image but may instead reflect the previous consecutive days, since rainwater percolates through the snow and firn pack creating prolonged wet conditions. In RADARSAT-2 ScanSAR data (Fig. 9) and meteorological data (Fig. A2 and Table A1), colder conditions were measured on 4 January 2016 (DOY 4, 1 day before the Sentinel-1A acquisition). As stable low backscatter values were observed on 4–5 January 2016 (DOY 4 to 5) (Fig. 9 and Table A1), we suggest that capillary water in the snow and firn, which was not yet frozen, was detected by the Sentinel-1 image on 5 January 2016 (DOY 5). This could be related to the isolation capability of snow and release of latent heat when water refreezes.

The same wet-snow conditions were not found on Holtedahlfonna (Figs. 7a and 8c). We observed a rain event on Holtedahlfonna on 30 December 2015 (white arrow on DOY 364 in Fig. 9), corresponding to 26.5 mm rain in Ny-Ålesund (Table A1). In the following days, Holtedahlfonna had gradually higher SAR backscatter values, while Kongsvegen remained dark (low backscatter). Either precipitation fell as snow on Holtedahlfonna between 3 and 5 January 2016 (DOY 3 to 5) or conditions were cold with no snowfall. Thus, the RADARSAT-2 ScanSAR images revealed different local weather conditions in the Kongsfjorden region. Colder conditions are in general present on Holtedahlfonna as it has a more continental climate compared to Kongsvegen, indicating that the rainwater in the snow and firn pack froze faster than on Kongsvegen.

Directly after the winter rain event, we observed a changing backscatter pattern in the ablation zone and lower part of the superimposed ice zone, indicating change in snow and ice properties on Kongsvegen (corresponding to no. 4 in Fig. 2). A change was also apparent from the modeled firn air content in Fig. 6b and reflects an increase in air content in the snow and firn pack after a rain event. Backscatter values increased > 2 dB after the winter rain event between 300 and 600 m, and the border between the superimposed ice and ablation zone became unclear (Fig. 6a). We know from in situ observations on Kongsvegen that glacier ice in the ablation area was covered with little snow in the spring 2016. The lower zone of the ablation area might have had little snow cover already before the rain event since the backscatter was continuously low during winter (indicated by light blue color in Fig. 6a; present around the 200 m elevation line from 17 January to 4 May 2016, DOY 17 to DOY 125). In the upper zone of the ablation area and before the rain event, more snow was present compared to lower elevations. Ice lenses and pipes might thus have been created from this penetration of water in the snow after the rain event, resulting in snow saturation giving stronger permittivity contrast as shown by increasing backscatter values (indicated by yellow color in Fig. 6a; present around the 400 m elevation line from 17 January to 4 May 2016, DOY 17 to DOY 125).

Detection of winter rain events from SAR time series might be used to refine modeling inputs, especially in regions where meteorological data are scarce.

It is well known that glacier ice is mapped more efficiently with optical satellite imagery than with SAR satellite data (e.g., Shi et al., 1994). However, when using the multispectral band ratio method on a single optical image for deriving glacier outlines (e.g., Paul and Kääb, 2005), it can be difficult to separate seasonal snow from glacier and perennial snow patches (e.g., Andreassen et al., 2012; Winsvold et al., 2016). Local differences in weather conditions within a single optical satellite scene are common in high mountain regions. Mapping conditions in a single optical image might not be ideal due to cloud cover, seasonal snow around the glacier perimeter and thin layers of newly fallen snow. Interferometric coherence images have been used for deriving glacier outlines (e.g., Atwood et al., 2010; Falk et al., 2016). In maritime regions such as southwestern Norway it can be challenging to use SAR coherence images due to rapid coherence loss between the time of acquisitions often caused by precipitation, melt and wind. SAR backscatter information can be a valuable replacement. Wet snow and ice surfaces have lower backscatter values than a snow-free surface outside the glacier (Rott, 1984; Strozzi et al., 1997). In this application scenario, we investigate how averaged summer SAR images from glaciers in southern Norway can be used to assist in glacier outline mapping (e.g., corresponding to no. 11 in Fig. 2).

In 2015, seasonal snow remained throughout the ablation season in our study area (Fig. 10d, e and f). A SAR composite consisting of the mean of six summer SAR images might assist the mapping process of glacier outlines and perennial snow patches (Fig. 10a, b and c). This is possible as the seasonal snow was less visible in the summer SAR images compared with the optical image (Fig. 10b and e, and c and f). Most likely, some SAR wave penetration in the seasonal snow is possible despite the potentially wet conditions. Thin snow that contaminates the optical mapping scene might become largely transparent in the SAR stack. In addition, we speculate that water content in the seasonal snow might have been low in some of the SAR acquisitions. The Sentinel-1 scenes used here are taken in the afternoon at time 15:45. Finally, water might drain easier from snow patches than on the glacier surface due to often higher slopes, which results in a clearer signal from on-glacier pixels since the signal is consistently low.

For our example, it is possible to retrieve a good estimate of the glacier outlines when summer SAR images are averaged. Although the backscatter images were affected by some radar distortions, we found good agreement with the existing glacier outlines (Andreassen et al., 2008) and the Landsat 8 image (Fig. 10a and d). The potential of the stacked SAR image can be further improved with double revisit time of Sentinel-1A and B. Clearly, the SAR image stack used cannot fully replace optical scenes for glacier mapping but can be of help as an additional layer to discriminate glacier areas from seasonal snow.