Synthetic Aperture Radar Monitoring of Snow in a Reindeer-Grazing Landscape

Abstract

Snow cover and runoff play an important role in the Arctic environment, which is increasingly affected by climate change. Over the past 30 years, winter temperatures in northern Sweden have risen by 2 °C, accompanied by an increase in precipitation. This has led to a higher incidence of thaw–freeze and rain-on-snow events. Snow properties, such as the snow depth and longevity, and the timing of snowmelt in spring significantly impact the alpine tundra vegetation. The emergent vegetation at the edge of the snow patches during spring and summer constitutes an essential nutrient supply for reindeer. We have used Sentinel-1 synthetic aperture radar (SAR) to determine the onset of the surface melt and the end of the snow cover in the core reindeer grazing area of the Laevás Sámi reindeer-herding community in northern Sweden. Using SAR data from March to August during the period 2017 to 2021, the start of the surface melt is identified by detecting the season’s backscatter minimum. The end of the snow cover is determined using a threshold approach. A comparison between the results of the analysis of the end of the snow cover from Sentinel-1 and in situ measurements, for the years 2017 to 2020, derived from an automatic weather station located in Laevásvággi reveals a 2- to 10-day difference in the snow-free ground conditions, which indicates that the method can be used to investigate when the ground is free of snow. VH data are preferred to VV data due to the former’s lower sensitivity to temporary wetting events. The outcomes from the season backscatter minimum demonstrate a distinct 25-day difference in the start of the runoff between the 5 investigated years. The backscatter minimum and threshold-based method used here serves as a valuable complement to global snowmelt monitoring.

1. Introduction

Arctic environments are experiencing increasing temperatures at a rate that is about twice as fast as the average rate for the rest of the world [1]. The most pronounced increase has occurred during the winter and spring seasons [2,3,4]. In northern Sweden, the average winter temperatures exhibit a 2 °C increase over the past 30 years [5]. Warming scenarios also indicate a rise in winter snow accumulation, along with notable changes in the snow properties [6]. A changing winter climate has serious implications for reindeer husbandry in northern Scandinavia. Higher winter temperatures lead to more frequent occurrences of temperature fluctuations above and below zero, rain-on-snow events and thaw–freeze cycles [7,8]. Rain-on-snow involves melting and subsequent refreezing, creating hard layers at the base or within the snowpack, which make grazing difficult [8]. These types of events can also create conditions favourable for mould, resulting in toxic lichens [5] and adversely affecting the nutrient-holding capacity of Carex bigelowii [9]. Unfortunately, the frequency of such events is predicted to increase with further warming [3,8].

Plants adapted to low temperature and low nutrient supply, frost tolerance and photosynthetic efficiency characterise the alpine tundra vegetation. The most critical parameters affecting the length of the growing season in alpine tundra ecosystems are the snow accumulation during the winter [10,11], snowmelt timing [12], soil temperature beneath the snowpack, and air temperature before, during and after snowmelt [7,13,14]. Consequently, even a slight alteration in the length of the spring season by just a few weeks can have a substantial impact on these complex arctic systems [12,14,15]. Furthermore, snow-covered ground contains relatively higher soil nutrient content, and therefore, shorter snow seasons with reduced snow cover may adversely impact soil nutrients [10,16,17]. Previous studies have shown that the distribution of alpine tundra vegetation is strongly related to the snowmelt gradient [18,19]. The altered snow conditions resulting from rising temperatures and changing precipitation patterns [20] make it challenging for reindeer to smell, identify, and access the ground lichens they feed upon during winter [5].

Reindeer move over large distances during their annual cycle and use different habitats during winter, spring, summer and autumn grazing [5,21]. Grazing modifies the vegetation above and below ground [22]. A change in the grazing pattern can lead to a loss of plant biodiversity in the area [23] and lead to the ecosystem transitioning from moss-shrub-dominated tundra to graminoid-dominated tundra [24]. It has been suggested by Cohen et al. [25] that reindeer grazing may reduce ground heating because the grazing reduces the height of the vegetation, which, in turn, prolongs the snow cover and increases the albedo, thereby extending the spring snowmelt. In the alpine landscape, reindeer selectively forage in response to snowmelt gradients, often favouring snow-bed meadows with high plant diversity [26]. The prime period of nutritional abundance in vegetation aligns with the early stages of the growing season [27]. The improved nutrient quality during spring and early summer is crucial for the growth of both adult reindeer and calves. A later start to the growing season can substantially impact calf growth [5,28].

There is a significant variation in the types of forage consumed by reindeer throughout the year [29]. In spring, reindeer in Fennoscandia tundra ecosystems primarily consume 62% lichen, for example, Cladonia, 8% shrub, Betula nana, and 12% graminoids, including Deschampsia flexuosa and Carex Bigelowii [30]. During summer, the shrub and graminoid intake increases to 29% and 23%, respectively, with a decrease in lichen consumption [30]. One effect of higher summer temperatures is an increased number of insects, which can, in turn, cause stress in reindeer and make them migrate to higher altitudes, where food resources are more limited [5,31]. However, during warm summer days, reindeer prefer grazing on fresh vegetation appearing at the rim of the high-altitude perennial and seasonal snow patches. Due to the warming, this food resource is also diminishing [25,32]. The spring snowmelt season is an essential component of the arctic environments; remote sensing is an efficient way to monitor changes in high latitudes [33].

Active microwave synthetic aperture radar (SAR) enables all-weather day and night imaging. Sentinel-1 (S-1) SAR offers a 6-day revisit time at the equator, with 2 satellites in a 12-day orbit and a spatial resolution of around 20 m before speckle filtering [34]. Previous studies have used the SAR backscatter coefficient to measure the snowmelt and to generate snow cover maps [35,36,37], monitor snow wetness [38] and assess the relationship between the spring snowmelt and tundra vegetation changes [17,39].

Dry snow conditions generally result in higher backscatter than wet snow conditions due to the lower absorption and higher scattering at the snow–ground interface. In spring, there occurs a period of alternating low and high backscatter as the snow begins to accommodate more liquid, and then, during the ripening phase, the backscattering values start to steadily decrease (Figure 1). The season minimum of the S-1 backscatter determines the start of the surface melt (SOSM_S-1). In this study, the SOSM_S-1 is employed to delineate the transition from dry snow to a snowpack with persistent liquid water during daytime. The S-1 end of the snow cover (EOS_S-1) is when the backscatter reaches its highest value after the SOSM_S-1 [36].

The objective of this study is to determine if the timing and spatial distribution of the snowmelt in the Laevás Sámi reindeer-herding area during the period 2017 to 2021 can be derived from S-1 SAR data. The choice of years is based on the availability of satellite data and in situ data from an automatic weather station (AWS) located in Laevásvággi [5,40]. In this study, we establish the beginning and end of the snowmelt period during the spring and summer season (March–August) using Sentinel-1 SAR data over the Laevás reindeer-herding area. We investigate the snowmelt timing and distribution linked to alpine vegetation important for reindeer grazing.

2. Study Area and Datasets

2.1. Study Area

The area of interest (Figure 2) is the western part of the Laevás Sámi reindeer-herding area (18.5E°, 68.5N°) where, traditionally, the spring and summer grazing occurs [5,41]. It is located in a sub-arctic environment in Kiruna municipality in northern Sweden. The area is located along a west–east oceanic–continental climate gradient with a rain shadow effect from the highest part of the Scandinavian mountain range [5]. Covering an expanse of 1038 km², this diverse area is alpine, with altitudes up to 2003 m.a.s.l dominated by heath and bare surfaces. Valleys are forested with Betula pubescens ssp. Czerepanovii [42]. Lush meadow vegetation covers the mountain slopes, and in the wind-exposed heath, dwarf shrubs and lichens grow sparsely on the gravelly substrate [43]. Laevásvággi, which is a core area because reindeer calve there in the spring [5], is located between Rávttasjávri and Visttasvággi and the vegetation there is dominated by extremely wind-exposed heath and grasslands with grasses and herbs including D. flexuosa and C. Bigelowii and dwarf shrubs, for example, B. nana [42].

2.2. Datasets

We used single-look complex (SLC) S-1 SAR data in interferometric wide-swath mode (IW) in the VV and VH polarisations. The acquisitions were from the 1st of March to the 30th of August from 2017, 2018, 2019, 2020 and 2021 in the ascending relative orbit 131, except for the 4th of July 2021, which has orbit 58, downloaded from the Alaska Satellite Facility [45] with a repeat cycle of six days, except for the 27th of June to 9th of July, where there was no acquisition on the 2nd of July over the area of interest. All the images were acquired around 4:15 pm UTC. In total, we used 31 scenes from 2017, 35 scenes from 2018, 33 scenes from 2019, 36 scenes from 2020, and 37 scenes from 2021 (Figure 3). Additionally, a digital elevation model (DEM) with a spatial resolution of 2 m, provided by Lantmäteriet [44], was used to pre-process and radiometrically normalise the S-1 scenes. Data from an automatic weather station (AWS) were provided by Rosqvist et al. [5] and the Swedish Infrastructure for Ecosystem Science [40] for the years 2017 to 2020. The AWS collected the average, minimum, and maximum daily air temperature, snow depth and rainfall at a location in Laevásvággi. The snowmelt patterns were compared with landcover data provided by Lantmäteriet [43].

3. Method

3.1. Pre-Processing of Sentinel-1 Data

A temporal stack of backscatter coefficient (γ₀) intensities in dB was generated for both polarisations (VV and VH) [36]. All the scenes were pre-processed using graph builder and bulk processing in the European Space Agency Sentinel Application Platform (SNAP) version 8.0.9 (Figure 4). Automation using Python 3.9 was performed in PyCharm 2022.1.1. The data were pre-processed with the latest orbit file containing the satellite position, velocity, and orientation information over time [36]. The image’s swaths were then split, deburst and merged between the sub-swaths IW2 and IW3 into a single image. Next, a subset of the area was extracted using the coordinates for the study area mentioned earlier (Section 2.1). The next step involved calibration of the backscatter coefficient or beta-nought (β⁰), which measures the radar signal reflected from the Earth’s surface back to the satellite. The data underwent terrain-flattening correction using the 2 m DEM from Lantmäteriet [44]. This involved estimating a reference surface [44] to correct for roughness [46,47] and rectify each pixel’s location, brightness, and radiometry [46], as well as to correct for atmospheric absorption [47]. The processed data were multilooked with 4 azimuth and 1 range samples; a 5 × 5 Lee Sigma speckle filtering with a sigma value of 0.9 was then applied [36,48]. Geometric terrain correction was applied to translate each input data product separately and convert the SAR to map geometry. Precise external orbital state vectors were utilized to interpolate each azimuth line in the application of range-Doppler geolocation [47]. The backscatter measurements were then resampled from the slant or ground range geometry of the input products into the map geometry of the national elevation model, which was provided Lantmäteriet [44]. All the scenes were resampled to 15 × 15 m, and radiometric layover and shadow pixels were excluded.

3.2. Extracting SOSM and EOS Using Python

Using Python, the SOSM and EOS were extracted from the stacks of pre-processed S-1 images from all five years, and both polarisations were extracted following Buchelt et al. [36]. Processing scripts were rewritten for this project to reduce the processing time and provide a better overview of the processing of the S-1 images. The NumPy 1.22.4 library was utilized in all the matrix calculations to speed up the heavy data processing. The Geospatial Data Abstraction Library (GDAL) 3.4.3 was used for the input and output operations, resampling, and accessing metadata.

First, the SOSM_S-1 in each pixel was identified and the day of year (DOY) recorded (Equation (1)). For each pixel in the image stack along the time axis, the function computes the minimum value of the backscatter intensity (γ₀), which corresponds to the earliest date when snowmelt occurred for that pixel. Pixels with missing data were assigned a “no data” value. The SOSM_S-1 phase detection was defined from Buchelt et al. [36]:

$${SOSM}_{S\text{-}1}\left(x, y\right)=\left\{\begin{array}{c}DOY,if {\gamma }_0\left(x, y, DOY\right)=\mathit{min}\left({\gamma }_0\right)\\ -1, otherwise.\end{array}\right.$$

(1)

After that, the EOS_S-1 was determined for the threshold (Equation (2)), while the DOY of the EOS_S-1 was determined when the backscatter had passed two criteria [36]. First, it had to exceed the chosen threshold in three consecutive images; the threshold was established based on the typical seasonal backscatter behaviour [49]. The main threshold in this study was 3 dB, but for comparison, a threshold of 4 dB was also implemented for all the images. In order to detect possible refreeze and re-melt, a second criterion, a re-melt threshold, was employed. The threshold needed to exceed the season minimum by more than the chosen threshold and to be continued in at least three images to pass the re-melt threshold. The second criterion was applied to prevent false backscatter detection caused by melt and refreeze. The threshold-based EOS_S-1 approach detected the point at which a pixel’s backscatter intensity (γ₀) exceeded the season minimum and the threshold (t) [36]. The EOS_S-1 were defined as [36]:

$${EOS}_{S\text{-}1}\left(x,y\right)=\left\{\begin{array}{l}first DOY, where {\gamma }_0(x, yDOY)\\ >\min \begin{array}{c}\left({\gamma }_0\right)+t\end{array}\\ (three consecutive times)\\ start-of-season snow free , otherwise.\end{array}\right.$$

(2)

Equation (2) identifies the minimum value for each pixel in the stack. Then, for each day afterwards in the stack, the t was applied to check if the snowmelt has stopped. If the snowmelt had stopped, the day of the year for EOS_S-1 was determined, and pixels where the threshold is not reached were assigned “no data” values [36].

For the year 2017 to 2020, the EOS from S-1 was compared to in situ data from Laevásvággi AWS [5,40] by comparing the value of the pixel nearest to the Laevásvággi AWS located at 18.96°E 68.04°N. The nearest pixels in the EOS_S-1 and Laevásvággi AWS for four years are shown in Section 4.3. Finally, for the extracted SOSM_S-1 and EOS_S-1, raster zonal statistics were calculated for mountain vegetation data provided by Lantmäteriet [43] using ArcGIS Pro 2.9.5.

4. Results

4.1. Backscatter Behaviour between the Different Polarisations and Years

There is a distinct difference between the years in terms of the backscatter behaviour between during the spring and summer season (March–August). The moistening and ripening phase started later in 2017 and 2020 than in 2018 and 2019, indicating a difference in the longevity of the snow cover. The average overall backscatter intensity at the seasonal minimum in 2017 and 2020 was lower than in 2018 and 2019 in both polarisations, though the difference in VH is close to the reported radiometric accuracy [50]. An exception is 2021, which shows a much lower backscatter on day 136 (16/5) compared to previous years in both polarisations. The S-1 backscatter trend in both polarisations over 5 years is characterized by a period of prolonged high backscatter intensity of between 30 and 36 days before it reaches the ripening phase (Figure 5). However, in 2017, the VV polarisation exhibited lower backscatter intensity between DOY 73 (14/3) and DOY 85 (26/3) before returning to the previous level. The VV polarisation has a higher intensity than the VH, but the VV intensity exhibits a larger range between the highest and lowest values.

4.2. Seasonal Variations in SOSM and EOS from Sentinel-1 (2017–2021)

The analysis of the S-1 data shows that the snowmelt in the study area differs between the years 2017 and 2021 (Figure 5). The ripening phase, when the backscatter curve has a downward trend before reaching the minimum seasonal backscatter coefficient, differs across all the years, with a span of approximately 25 to 45 days. The date of the minimum backscatter (SOSM_S-1) is also different between years. The minimum backscatter occurs at day 151 (31/5) in 2017, day 154 (2/6) in 2020 and day 136 (16/5) in 2021, and about 25 days earlier in 2018 and 2019, at day 122 (2/5) and day 123 (3/5). This indicates that the spring snow-melting period, after the backscatter has reached its minimum values, started about 25 days earlier in 2018 and 2019 than in 2017 and 2020.

In 2017, the greatest number of pixels identified as the SOSM_S-1 was observed at day 79 (20 March) in the VV (56% of the area), which was not found in the VH or other years (

Table 1. An overview of the SOSM_S-1 data during the period from March to July across the years 2017 to 2021. The presented percentages correspond to the SOSM_S-1 measurements obtained for both the VV and VH polarisations. Notably, records for August are omitted due to the minimal percentage of SOSM occurrences during this month.

SOSM
2017			2018			2019			2020			2021
Date	VV	VH	Date	VV	VH	Date	VV	VH	Date	VV	VH	Date	VV	VH
2-Mar	0%	4%	3-Mar	1%	1%	4-Mar	3%	5%	4-Mar	1%	1%	5-Mar	0%	2%
8-Mar	0%	2%	9-Mar	1%	4%	10-Mar	2%	2%	10-Mar	1%	2%	11-Mar	0%	2%
14-Mar	0%	2%	15-Mar	1%	1%	16-Mar	3%	5%	16-Mar	0%	0%	17-Mar	0%	3%
20-Mar	56%	1%	21-Mar	1%	2%	22-Mar	1%	1%	22-Mar	1%	2%	23-Mar	0%	1%
26-Mar	0%	2%	27-Mar	1%	1%	28-Mar	4%	3%	28-Mar	0%	0%	29-Mar	0%	1%
1-Apr	0%	1%	2-Apr	1%	2%	3-Apr	1%	1%	3-Apr	1%	2%	4-Apr	0%	1%
7-Apr	0%	1%	8-Apr	1%	0%	9-Apr	2%	3%	9-Apr	1%	0%	10-Apr	0%	1%
13-Apr	0%	1%	14-Apr	4%	5%	15-Apr	3%	2%	15-Apr	1%	1%	16-Apr	2%	6%
19-Apr	0%	1%	20-Apr	21%	11%	21-Apr	19%	25%	21-Apr	4%	5%	22-Apr	0%	4%
25-Apr	0%	2%	26-Apr	19%	21%	27-Apr	14%	5%	27-Apr	1%	3%	28-Apr	0%	0%
1-May	0%	2%	2-May	6%	2%	3-May	4%	6%	3-May	5%	3%	4-May	0%	1%
7-May	0%	4%	8-May	6%	14%	9-May	5%	3%	9-May	2%	3%	10-May	1%	3%
13-May	0%	7%	14-May	21%	10%	15-May	9%	9%	15-May	0%	0%	16-May	56%	78%
19-May	18%	19%	20-May	3%	11%	21-May	10%	5%	21-May	19%	21%	22-May	2%	3%
25-May	6%	14%	26-May	1%	0%	27-May	10%	12%	27-May	38%	23%	28-May	3%	11%
31-May	1%	2%	1-Jun	1%	1%	2-Jun	3%	3%	2-Jun	10%	21%	3-Jun	3%	1%
6-Jun	10%	22%	7-Jun	1%	1%	8-Jun	1%	2%	8-Jun	6%	3%	9-Jun	1%	1%
12-Jun	1%	1%	13-Jun	2%	4%	14-Jun	0%	0%	14-Jun	1%	3%	15-Jun	0%	0%
18-Jun	2%	3%	19-Jun	0%	0%	20-Jun	1%	1%	20-Jun	2%	0%	21-Jun	0%	1%
24-Jun	1%	0%	25-Jun	3%	6%	26-Jun	0%	0%	26-Jun	0%	1%	27-Jun	0%	0%
30-Jun	2%	1%	1-Jul	1%	0%	2-Jul	2%	3%	2-Jul	1%	1%	4-Jul	0%	0%
6-Jul	0%	0%	7-Jul	1%	1%	7-Jul	0%	0%	8-Jul	1%	1%	9-Jul	0%	0%
12-Jul	0%	1%	13-Jul	0%	0%	14-Jul	1%	0%	14-Jul	0%	0%	15-Jul	53%	0%
18-Jul	0%	1%	19-Jul	0%	0%	20-Jul	0%	0%	20-Jul	0%	1%	21-Jul	0%	1%
24-Jul	1%	0%	25-Jul	0%	0%	26-Jul	0%	0%	26-Jul	0%	0%	27-Jul	0%	0%

or Figure 6a). Instead, the VH for the same year showed large SOSM_S-1 areas in May (25%) and June (15%) rather than in March (Figure 7b). In 2018 and 2019 (Figure 6b,c, and Figure 7b,c), both polarisations showed similar patterns of the SOSM_S-1, with some variations in different slopes and elevations. The analysis identified the highest amount of SOSM_S-1 in 2018 in April (46% in VV and 39% in VH) and May (37% VV and VH). In 2019, the largest number of pixels identified as the SOSM_S-1 was in April (39% in VV and 35% in VH) and May (37% VV and 36% VH). In 2020 (Figure 6d and Figure 7d), the VV showed the largest SOSM_S-1 in May (65%) and June (20%), while the VH showed a more even SOSM_S-1 throughout May (50%) and June (28%). In 2021, both the VV and VH exhibited an SOSM_S-1 in May, with the VV at 62% and VH at 96%. However, while the VV showed a significant snowmelt in July (53%), the VH exhibited its peak only in May (Figure 6e and Figure 7e). There is a similarity in the SOSM_S-1 patterns in 2017 (VH) and 2020 (VV and VH) and between the years 2018 (VV and VH) and 2019 (VV and VH). The results show the delayed onset of the spring surface melt in 2017 and 2020 compared to 2018 and 2019. After the backscatter minimum, the runoff phase differs in length between the polarisations in 2019 by 18 days between the VV and VH but is similar in both polarisations in the other years, with a range of 49 to 107 days. The EOS occurs at day 211 (30/7) in 2017 and day 212 (31/7) in 2018, at day 207 (26/7) in the VH and day 237 (25/8) in the VV in 2019, day 214 (1/8) in 2020 and day 185 (4/7) in 2021. In 2021, the runoff period was the shortest compared to previous years, lasting only 49 days.

From 2017 to 2021, the Laevás reindeer-grazing region experienced significant changes in the EOS_S-1, according to the data (Figure 8 and Figure 9). The EOS_S-1 was identified for all the years, and in 2017 (Figure 8a and Figure 9a), there was a clear difference between the polarisations. The VV had a higher number of pixels identified as the EOS_S-1 in March (40%), while a larger EOS_S-1 was observed in June for the VH (49%) but also for the VV (38%). The highest amount of the area that was identified as the EOS_S-1 for 2018 (Figure 8b and Figure 9b) was found in May (54%), and for 2019 (Figure 8c,e and Figure 9c), the EOS_S-1 was similar in both polarisations, with the highest number of pixels in May (30% in VV and 28% in VH) and June (34% in VV and 31% in VV). There was an equivalency in the EOS_S-1 in 2020 (Figure 8d and Figure 9d) for both polarisations, with the highest number in June (66% for VV and 56% for VH). In 2021, both the VV and VH exhibited an EOS in May (VV at 24% and VH at 48%). However, while the VV showed a significant snowmelt in July (77%), the VH exhibited an EOS in June (53%).

4.3. Sentinel-1 and In Situ Data from Laevásvággi AWS

The comparison of the EOS_S-1 and ground data collected by the Laevásvággi AWS [5,40] revealed a difference of 2 to 10 days between the timing of the EOS_S-1 and the in situ measurements (

Table 2. The EOS_S-1 at the location of the Laevásvággi AWS and the discrepancies between the EOS_S-1 and ground truth data. A generally good agreement was noted between the EOS_S-1 and ground truth measurements, with only a small-time range of 2 to 10 days observed. However, anomalies were detected in 2019, where the EOS_S-1 data indicated a delay in the snow-free day beyond the expected range, occurring on 2 May and 7 and 13 August. In all the years, except for 2019, the day the ground becomes snow-free is consistent across all the polarisations. Data for year 2021 are not available from the Laevásvággi AWS and are therefore not included in this table.

Year	Date	AWS Snow Free Date	Polarisation
2017	12 June	15 June	VV
	12 June and 18 June		VH
2018	20 May	22 May	VV
	26 May and 1 June		VH
2019	7 August	22 May	VV
	2 May and 13 August		VH
2020	9 June	19 June	VV
	21 June		VH

). However, anomalies were detected in 2019, where the EOS_S-1 indicated a delay to 7 and 13 August, which is beyond the expected range. In all the years except for 2019, the day the ground becomes snow-free is consistent across both polarisations. It should be noted that the S-1 method, requiring 3 consecutive acquisitions, introduces a potential lag of up to 12 days in the estimates of the SOSM and EOS. Comparisons with in situ data should take this into consideration.

4.4. Differences between Landcover Classes

The zonal statistics were extracted from the SOSM_S-1 and EOS_S-1 datasets (VV and VH and all years) for different landcover classes. The resulting statistics, including the mean DOY values and the standard deviation (STD) per class, represent the degree of variation or spread around the mean DOY value in the different types of vegetation. These statistics are presented in

Table 3. Zonal statistics for the SOSM_S-1 raster dataset and mountain vegetation data [43]. The mean day of year (DOY) values from the SOSM_S-1 and the standard deviation (STD) discounts. The STD represents the degree of variation around the mean DOY from the SPSM_S-1 within the specific vegetation zone. The area is in square kilometres.

SOSM
		2017						2018
		VV			VH			VV			VH
Vegetation	Area **	Mean DOY	STD *		Mean DOY	STD *		Mean DOY	STD *		Mean DOY	STD *
Dry heath (e.g., B. nana)	131.3	98	±	30	125	±	34	115	±	19	114	±	20
Alpine low grass meadow (e.g., C. Bigelowii)	132.0	103	±	31	134	±	29	117	±	18	118	±	19
Low-growing shrubs	37.2	101	±	31	128	±	29	116	±	13	115	±	14
Grassland (e.g., D. flexuosa)	104.7	110	±	35	144	±	28	124	±	20	128	±	23
Open marsh vegetation	14.2	90	±	29	118	±	36	118	±	19	113	±	21
Heath (e.g., heather)	80.0	93	±	28	122	±	33	113	±	19	111	±	20
Extreme dry heath (e.g., B. nana)	21.6	93	±	26	123	±	36	116	±	17	118	±	17
Alpine tall grass meadow	9.2	101	±	31	132	±	27	114	±	20	108	±	19
Moist–wet heath (e.g., Sweetgale)	3.9	121	±	31	139	±	19	117	±	13	116	±	15
		2019						2020
		VV			VH			VV			VH
Vegetation	Area **	Mean DOY	STD *		Mean DOY	STD *		Mean DOY	STD *		Mean DOY	STD *
Dry heath (e.g., B. nana)	131.3	112	±	26	108	±	25	140	±	21	136	±	25
Alpine low grass meadow (e.g., C. Bigelowii)	132.0	119	±	23	115	±	24	144	±	17	140	±	23
Low-growing shrubs	37.2	113	±	20	109	±	19	142	±	15	139	±	18
Grassland (e.g., D. flexuosa)	104.7	125	±	26	128	±	27	145	±	19	146	±	23
Open marsh vegetation	14.2	110	±	35	99	±	31	136	±	25	129	±	24
Heath (e.g., heather)	80.0	110	±	27	105	±	24	137	±	23	132	±	26
Extreme dry heath (e.g., B. nana)	21.6	110	±	27	107	±	26	137	±	23	132	±	28
Alpine tall grass meadow	9.2	119	±	26	111	±	24	140	±	22	137	±	19
Moist–wet heath (e.g., Sweetgale)	3.9	122	±	18	117	±	18	147	±	8	145	±	13
		2021
		VV			VH
Vegetation	Area **	Mean DOY	STD *		Mean DOY	STD *
Dry heath (e.g., B. nana)	131.3	161	±	30	128	±	21
Alpine low grass meadow (e.g., C. Bigelowii)	132.0	158	±	29	134	±	16
Low-growing shrubs	37.2	151	±	27	131	±	16
Grassland (e.g., D. flexuosa)	104.7	157	±	27	137	±	13
Open marsh vegetation	14.2	157	±	37	116	±	29
Heath (e.g., heather)	80.0	163	±	31	126	±	22
Extreme dry heath (e.g., B. nana)	21.6	166	±	30	128	±	21
Alpine tall grass meadow	9.2	170	±	31	124	±	21
Moist–wet heath (e.g., Sweetgale)	3.9	144	±	21	134	±	15

* STD = Standard deviation ** Square kilometres.

for SOSM_S-1 and

Table 4. Zonal statistics were based on the EOS_S-1 raster dataset and mountain vegetation data from Lantmäteriet (2022b). The analysis calculated the mean day of year (DOY) values derived from the EOS_S-1 alongside the corresponding standard deviation (STD) measurements. The STD indicates the level of variation around the mean DOY obtained from the EOS_S-1 within each specific vegetation zone. The area is in square kilometres.

EOS
		2017						2018
		VV			VH			VV			VH
Vegetation	Area **	Mean DOY		STD *	Mean DOY		STD *	Mean DOY		STD *	Mean DOY		STD *
Dry heath (e.g., B. nana)	131.3	117	±	40	126	±	34	125	±	60	127	±	60
Alpine low grass meadow (e.g., C. Bigelowii)	132.0	126	±	38	130	±	29	127	±	56	130	±	63
Low-growing shrubs	37.2	121	±	39	137	±	29	128	±	41	129	±	38
Grassland (e.g., D. flexuosa)	104.7	133	±	42	114	±	28	124	±	72	119	±	91
Open marsh vegetation	14.2	97	±	49	108	±	36	121	±	55	120	±	62
Heath (e.g., heather)	80.0	112	±	39	128	±	33	123	±	59	126	±	53
Extreme dry heath (e.g., B. nana)	21.6	110	±	37	104	±	36	118	±	71	121	±	75
Alpine tall grass meadow	9.2	120	±	39	123	±	27	120	±	67	122	±	62
Moist−wet heath (e.g., Sweetgale)	3.9	142	±	36	150	±	19	131	±	30	129	±	34
		2019						2020
		VV			VH			VV			VH
Vegetation	Area **	Mean DOY		STD *	Mean DOY		STD *	Mean DOY		STD *	Mean DOY		STD *
Dry heath (e.g., B. nana)	131.3	125	±	79	133	±	58	144	±	61	142	±	70
Alpine low grass meadow (e.g., C. Bigelowii)	132.0	130	±	77	138	±	63	150	±	51	145	±	71
Low-growing shrubs	37.2	123	±	63	126	±	45	147	±	48	148	±	48
Grassland (e.g., D. flexuosa)	104.7	125	±	90	123	±	97	145	±	68	133	±	93
Open marsh vegetation	14.2	109	±	87	115	±	70	134	±	70	131	±	77
Heath (e.g., heather)	80.0	120	±	82	130	±	52	142	±	62	144	±	59
Extreme dry heath (e.g., B. nana)	21.6	118	±	91	129	±	71	137	±	73	130	±	88
Alpine tall grass meadow	9.2	121	±	86	130	±	68	137	±	71	145	±	58
Moist−wet heath (e.g., Sweetgale)	3.9	132	±	42	131	±	42	155	±	26	152	±	41
		2021
		VV			VH
Vegetation	Area **	Mean DOY		STD *	Mean DOY		STD *
Dry heath (e.g., B. nana)	131.3	182	±	27	151	±	13
Alpine low grass meadow (e.g., C. Bigelowii)	132.0	182	±	27	154	±	13
Low-growing shrubs	37.2	170	±	29	147	±	10
Grassland (e.g., D. flexuosa)	104.7	180	±	26	156	±	13
Open marsh vegetation	14.2	184	±	29	143	±	18
Heath (e.g., heather)	80.0	184	±	27	148	±	13
Extreme dry heath (e.g., B. nana)	21.6	186	±	26	152	±	14
Alpine tall grass meadow	9.2	191	±	23	150	±	16
Moist−wet heath (e.g., Sweetgale)	3.9	163	±	26	148	±	8

* STD = Standard deviation ** Square kilometres.

for EOS_S-1.

The results reveal notable variations across the vegetation classes, polarisations, and years. As mentioned earlier, the VV data from 2017 show an anomaly compared to the VH data from the same year and other investigated years. For example, in 2017, dry heath and alpine low grass meadow exhibited the SOSM_S-1 around day 98 and 103 (VV), respectively, compared to approximately day 125 and 134, respectively, for the VH polarisation (Table 3). In 2020, dry heath and alpine low grass meadow showed the SOSM_S-1 at day 112 and 119 (VV), respectively, contrasting with approximately day 108 and 115, respectively, for the VH polarisation. In 2021, the mean DOY for the SOSM_S-1 in alpine low grass meadows was observed at day 158 for the VV and day 134 for the VH. In dry heat, the mean DOY was day 161 for the VV and day 128 for the VH. Notably, there was a 26-day difference observed within specific landcover classes in the VV and a 21-day difference in the VH.

Furthermore, between the VV data from 2018 and 2019, there is a 1- to 12-day difference in the timing of the SOSM_S-1 within specific landcover class. Similarly, the difference between the VH data from 2018 and 2019 ranges from 3 to 14 days.

The study area primarily consists of two dominant landcover classes: dry heath and alpine low-grass meadow. The range of variation in the start of the SOSM_S-1 each year across the different landcover class extends from a minimum of 10 days in 2020 to a maximum of 29 days in 2019. Laevásvággi exhibits three main landcover classes: dry heath, extreme dry heath, and alpine low-grass meadow. The time difference in the mean DOY between 2018 and 2019 is only seven days in both polarisations for all three landcover classes. However, in 2020, and specifically in 2021, the SOSM_S-1 occurred later in all three landcover classes compared to previous years.

The most significant variation is observed in the mean DOY in the EOS_S-1 for grassland vegetation between 2017 and the subsequent years for both the VV and VH polarisations (Table 4). Specifically, in 2017, the mean EOS_S-1 date for the VV polarisation was approximately day 133, contrasting with approximately day 114 for the VH polarisation, while in 2020, these dates shifted to around day 145 (VV) and day 133 (VH), respectively. In 2021, there was a later overall mean DOY for the EOS_S-1. The mean DOY across different vegetation classes and polarisations ranged from day 136 to day 191, indicating a delayed EOS_S-1 compared to previous years.

5. Discussion

The goal of this investigation was to determine if information about the timing and spatial distribution of snowmelt in the Laevás Sámi reindeer-herding area can be derived from Sentinel-1 SAR data between the years 2017 and 2021. The results from Sentinel-1 showed that there was high variability between the years in the timing of the start of the snowmelt, which began 25 days earlier in 2018 and 2019 compared to 2017 and 2020.

5.1. Variation in Backscattering and Spring Snowmelt

There was similar backscatter behaviour over time for each of the five years (2017−2021). Also, the three melting phases defining the snow melting process, moistening, ripening, and runoff, can be distinguished using multi-temporal C-band SAR measurements in the study area. These results compare well with the findings of Buchelt et al., Marin et al. and Lund et al. [36,49,51].

It is notable that there are significant differences in elevation across the study area, which may explain the early SOSM across the valleys in 2017. The difference in the VV and VH responses in March 2017 is difficult to explain. The VV backscatter recovered after dipping to −11 dB, which may indicate that VH data provide a more stable response to short-term weather events (Figure 5). The onset of melt, particularly at lower elevations, may also indicate warm air intrusion in the valleys that caused some wetting of the snow surface to which the co-polarised channel was more sensitive, perhaps due to volume scattering from shrubs and trees in the cross-polarised data (Figure 6, Figure 7, Figure 8 and Figure 9). Spring 2017 otherwise experienced deep snow late into the spring, according to data from the Swedish Meteorological and Hydrological Institute [52]. The results from 2021 show that the VV signal is confused by the presence of vegetation in the forested lowlands. The sharp contrast in the timing of the SOSM_S-1 in the VV and VH strongly argues for the use of VH data where possible. Failing that, we recommend masking out forests and dense shrublands.

All the S-1 SAR data that were used for this study were acquired in the afternoon and can therefore provide an earlier SOSM_S-1 because of the natural 24 h cycle of the snowpack, with greater potential for daytime surface melt and refreeze during the night [36]. In Visttasvággi, the vegetation includes taller vegetation such as birches. When the radar encounters elements like tree trunks, leaves, and branches within a forest canopy, it experiences double bounce and volume scattering, typically resulting in higher backscatter [39]. The method used in this paper may be particularly effective for studying the snowmelt in areas without significant tree height variations due to the distinctive values obtained in such an environment. The elevation and terrain significantly influence the environmental conditions that are favourable for the different vegetation habitats’ possibility of establishment within the region. Consequently, the snowmelt pattern varies depending on the dominant vegetation class. Both dry and wet terrain, as well as higher and lower vegetation types, play a decisive role in the rate of snowmelt [53]. Moist–wet heath may impact the S-1 observations due to the greater surface and ground water content affecting the backscatter.

5.2. Impact of Variation in Spring Snowmelt on Vegetation and Reindeer

In the five years of the SOSM_S-1 observations, two patterns emerge. Here, 2017 and 2020 show late melting at higher elevations, whereas 2021 shows rapid melting early in the summer. The earlier snowmelt in 2018 might be attributed to an extended period of warm temperatures commencing in May, coupled with lower-than-average precipitation, as indicated by data from the Swedish Meteorological and Hydrological Institute [54]. This high interannual variability will no doubt impact the vegetation productivity [13], with effects on the wildlife [27,29]. Previous investigations indicate that longer growing seasons and the early onset of snowmelt can result in tundra vegetation species being outperformed by species from more temperate biomes [18,55]. Satellite data from 1987 to 2004 show a decline in the number of days with snow cover and an earlier snowmelt in the Northern Hemisphere, with an estimation of 10 to 20% less total snow cover [56] and 4−7 days earlier start of the snowmelt during the time period [56,57]. And since 1990, Sweden has observed a reduction of up to 16 days in the snow-covered season [58]. Prolonged snow-free periods may have detrimental impacts on the environment in these regions, potentially resulting in changed vegetation, for example, shrubification [59], and result in reduced nitrogen levels in deciduous shrub species [55]. Between 1995 and 2011, there was a noticeable increase in the average cover per square meter of both evergreen and deciduous shrubs in the Swedish mountains [60]. Evergreen shrubs, for instance, have risen from 26% to 49% in heath sites, while deciduous shrubs, including B. nana, have increased from 5% to 10%. The expansion of shrubs negatively affects the growth of moss and lichens [60,61]. Longer-term changes with higher air temperature [62] and changes in seasonal snow significantly affect sensitive alpine vegetation [63,64], ultimately affecting grazing opportunities for reindeer.

Changes in the timing of the start of the surface melt impact reindeer as they may miss feeding opportunities, which are crucial for nutrient intake during critical spring and early summer [5,27]. Even small shifts in the snowmelt timing can affect species with an earlier niche, influencing mortality, phenology, and reproduction [65,66,67,68,69]. We detected as much as a 25-day temporal difference between start of the surface melt, which must have impacted the plant communities in the study area significantly [64,70,71].

The five-year study period is too short to provide insights into the long-term impact of changes in the snow cover and surface melt timing on vegetation. However, our results provide information about the large interannual variability of the timing of the snow melt and the length of the snow cover period that is crucial for assessing the impact on plant communities in a changing climate.

As temperatures are predicted to continue to increase in the Arctic, the snow properties and the amount of runoff during spring will be altered in the future. As a result, tundra mountain vegetation will significantly change and the total area covered by tundra will shrink, and the conditions for Laevás reindeer will also change. The method presented here serves as a valuable tool for, for example, decision-makers, applicable in different reindeer husbandry areas, which enables analysis of the impact of the snowmelt on reindeer husbandry.

Therefore, utilization of existing satellite data can provide essential insights into how vegetation is impacted by the reduction of snow patches and the onset of surface melt in northern Sweden. Previous research [36,49,51] underscores the efficacy of backscatter behaviour from synthetic aperture radar (SAR) data as a practical approach for monitoring the Earth’s surface changes. The accessibility of Sentinel-1 data further enhances the feasibility of monitoring the initiation of the surface melt on a national scale.

6. Conclusions

Here, Sentinel-1 SAR data were used to investigate the start of the surface melt and end of the snowmelt from 2017 to 2021 in the spring and summer grazing area of the Laevás reindeer-herding community in northern Sweden. The comparison between the end of the snow melt timing and the snow ground truth data showed promising results, with only a 2- to 10-day time shift, implying that the method is suitable for investigating snowmelt. The results show a clear backscatter behaviour trend for all five investigated years. The result indicates that there was a 25-day difference in the start of the surface melt between the years 2017 and 2020. The snowmelt began earlier in 2018 and 2019 compared to 2017 and 2020, which must have impacted the phenology and hence also reindeer grazing. The unusually early snowmelt detected in the VV data in 2017 suggests that VH data provide a more reliable overview of the SOSM and EOS than VV data. The end of the snow cover occurred at approximately the same time during all five years, which means that the spring and summer seasons were of different lengths between the investigated years. Monitoring the seasonal pattern and changes in snow in the region is essential for assessments of how future climate change will affect reindeers’ ability to find the vital spring and summer food, and it can serve as a valuable supplement for potential climate adaptations.