**4. Results**

#### *4.1. Albedo Before and After the Melt Season*

The change in albedo of Northern Hemisphere land areas between 40◦N and 80◦N before the melt season shows large spatial variations (Figure 4). The changes are concentrated in large areas with homogeneous change characteristics. These areas are listed in Table 2. The areas are shown on a map together with place names in Figure 5. For the observations with clear trends (*R*<sup>2</sup> > 0.5, 30% of all observations) the pre-melt albedo decreased by 2.4 absolute albedo percentage units on average over the whole study area over the 34 years of record. The statistically reliable trends in pre-melt season albedo are concentrated in the region of the boreal forest zone. This can be seen in Figure 4b, which shows the land-use classes for the resolution units with albedo trends with *R*<sup>2</sup> > 0.5. The trends in the boreal forest zone show decreasing pre-melt albedo in the southern half of the Central Siberian Plain and Scandinavia, and increasing pre-melt albedo in the north of Mongolia and China. In North America the direction of changes varies over short distances, whereas in Eurasia there are larger areas with similar direction of change in albedo.

**Figure 4.** (**a**) The rate of change in albedo before melt season (absolute albedo % per year) between 1982 and 2015 using 5 year rolling mean albedo (showing cases for which *R*<sup>2</sup> of the fitting was larger than 0.5). The positive rates of change mean higher albedo and negative lower albedo values. (**b**) The GlobCover land-use class for the areas with clear trends in pre-melt albedo. (**c**) The coefficient of determination (*R*2) value for trends in albedo before the onset of melt. (**d**) *R*<sup>2</sup> value for multiple variable linear correlation between albedo before melt season and 6 climatic parameters for 14 days prior onset of melt. The climatic parameters used are the mean air temperature, number of days with maximum temperature above 0 ◦C, −4 ◦C and −10 ◦C, accumulated precipitation and mean wind speed.


#### *Remote Sens.* **2018**, *10*, 1619

*Remote Sens.* **2018**, *10*, 1619

**Figure 5.** The areas listed in Table 2.

Most of the tundra areas show no significant change in albedo prior to melt. This can be seen in Figure 4c, which shows the *R*<sup>2</sup> values for all albedo trend retrievals. For all the areas with an *R*<sup>2</sup> value, the retrieval of melt season parameters was successful, but the data show no reliable trends (having *R*<sup>2</sup> values lower than 0.5). The slope values of these excluded trends are typically close to 0. In many areas the annual variability of the pre-melt albedo values is so large that even 34 years is not long enough to determine a small trend in the albedo value. In southern Eurasian tundra the pre-melt albedo shows weak negative trends, but no trends in the higher latitudes. The Kola Peninsula and northern Finnish Lapland show strong negative trends (−0.29 albedo percentage units per year).

The role of vegetation in the observed pre-melt albedo changes can be estimated by looking at the albedo levels right after the snow has melted, before the vegetation has started greening. Figure 6 shows the trends for albedo after melt as well as the corresponding *R*<sup>2</sup> values for the trend fitting. The trends in the level of albedo after melt are much weaker than those in the pre-melt season albedo before melt season, which can be expected since the differences in the albedo between different biomes are much smaller than the changes in the snow cover. The post-melt albedo of the northern Eurasian tundra decreased over the study period. In more southerly areas of tundra, however, there are no clear trends. In the boreal forest zone the trends are towards lower albedo values in most of the area, except for the area west of the River Ob in Russia. One potential reason for the darkening of the boreal forest zone after the melt season could be the increased size of trees and denser forests, causing more shadowing of the surface and increased multiple scattering. The darkening of the southern tundra prior to melt could be explained by the reported shrubification of tundra [49].

The role of climate change in altering the pre-melt season albedo was studied using the linear fitting between the melt season data and the ERA-Interim reanalysis data [45] on air temperature, precipitation, wind speed and the number of days with maximum temperature above 0 ◦C, −4 ◦C and −10 ◦C for 14 days prior to onset of melt. Figure 7 shows the 34-year trends in air temperature, accumulated precipitation and wind speed. Changes in all these climatic parameters contributed to the changes in the pre-melt albedo (mean *R*<sup>2</sup> for the whole area being 0.64 and the 80th percentile being 0.79) (Figure 4d). In the area around the borders of China, Mongolia and Russia (Figure 5) the climatic parameters explained almost all of the albedo change. The mean air temperature was the dominant influence (mean *R*<sup>2</sup> = 0.51 for the whole area). It was the largest explanatory factor in particular in Yablonovyy and Verhoyansk Mountain Ranges, Northern West Siberian Plain, Kola Peninsula, Baffin Island and Central Siberian Plain.

**Figure 6.** (**a**) The rates of change for the albedo after melt season (absolute albedo % per year) between 1982 and 2015 using 5-year rolling mean albedo (showing cases for which *R*<sup>2</sup> of the fit was larger than 0.5). (**b**) The coefficient of determination (*R*2) value for trends in albedo after the end of melt.

(**a**) **Figure 7.** *Cont*.

**Figure 7.** ERA-Interim trends for 1982–2015 for (**a**) mean air temperature (**b**) accumulated precipitation (**c**) snowfall (units are given as amount of snow converted into liquid water) (**d**) wind speed for 14 days prior to melt onset. The maps show trends for which the coefficient of determination was larger than 0.5.

## *4.2. Melt Season Timing*

In addition to the changes in albedo, the timing of the melt season has also changed (Figures 8–10 and Table 2). The changes are, as with albedo, significant and spatially consistent but they also vary within the study area. The observations with high values for coefficient of determination (*R*<sup>2</sup> > 0.5) are concentrated in large distinct areas. The changes were in general towards longer melt seasons and earlier onset of melt. The mean start date of melt season in the pixels for which melt season data are available for the whole 34 years, became 6.1 days earlier over the 34 years. Similarly, the melt ended on average 5.2 days earlier and the melt season, therefore, became 1 day longer on average. The majority of the observations showed no clear reliable trends (Figure 11), but in many areas the changes were significant (Figures 8–10). In many areas the inter-annual variation in the start and/or end dates of the melt season were so large that it was not possible to detect a statistically significant trend. In Eurasia

all the parameters showed changes over large homogenous areas, but in North America the trends are typically more localized and variable.

**Figure 8.** (**a**) The rates of change for the start day of melt between 1982 and 2015 using 5-year rolling mean albedo (showing cases for which *R*<sup>2</sup> of the fit was larger than 0.5). The negative rates of change mean earlier onset of melt and the positive rates of change mean later dates of onset of melt. (**b**) The coefficient of determination (*R*2) values for the trend fitting for start day of melt. (**c**) The multiple variable correlation (*R*2) between ERA-Interim climate data and start day of melt. *R*<sup>2</sup> value for linear correlation between the time that melt season started (day of year) and 3 climatic parameters for 14 days prior to the onset of melt. The climatic parameters used are the mean air temperature, accumulated precipitation and mean wind speed.

**Figure 9.** (**a**) The rates of change for the end date of melt between 1982 and 2015 using 5-year rolling mean (showing cases for which *R*<sup>2</sup> of the fit was larger than 0.5). The negative rates of change mean earlier end of melt and the positive rates of change mean later dates of end of melt. (**b**) The coefficient of determination (*R*2) values for the trend fitting for end day of melt.

**Figure 10.** (**a**) The rates of change for the length of melt season between 1982 and 2015 using 5-year rolling mean (showing cases for which *R*<sup>2</sup> of the fit was larger than 0.5). The negative rates of change mean shorter melt seasons and the positive rates of change mean longer melt season. (**b**) The coefficient of determination (*R*2) values for the trend fitting for length of melt season.

**Figure 11.** The rate of change observations for (**a**) start and (**b**) end day of melt. The dashed line shows the mean value of the observations for which *R*<sup>2</sup> is larger than 0.5.

The distribution of all the computed trends for start and end day of melt can be seen in Figure 11. The majority of the trends had *R*<sup>2</sup> values lower than 0.5. and the rate of change of these were typically close to zero but slightly towards earlier onset and end of melt.

The trends in melt season timing showed no significant dependency on the land use. In the Central Siberian Plain the melt started and ended earlier (Table 2), resulting in earlier melt seasons across the whole region regardless of the vegetation type (both boreal and tundra). The decreasing trends for mean air temperature before the melt onset in the Central Siberian Plain (Figure 7a, Table 2) show similar spatial patterns as the melt season timing parameters. This can be explained by the fact that the air temperatures prior to melt are not derived from the same time of year, but change together with the start date of melt. With earlier onset of melt, the air temperatures are also derived from an earlier period. In the mid-winter the air temperature is more heavily influenced by the lack of heating from the Sun, whereas later in the spring other climatic factors start to affect the air temperature more significantly. Earlier onset of melt can be associated with colder air temperatures prior to melt onset and more rapid change in the air temperature from cold mid-winter values to melting conditions.

In the area around the borders of Russia, Mongolia and China the melt starts earlier and ends later (Table 2), resulting in longer melt seasons. This is also the case for the Canadian Rocky Mountains. In North America, the northern parts of Labrador Peninsula, which are tundra, also show trends towards a shorter melt season and earlier end of melt.

Using the climatic data from ERA-Interim, three parameters (mean air temperature, mean wind speed and accumulated precipitation) are required to explain the changes in the start date of melt, giving a mean *R*<sup>2</sup> value of 0.65 for the whole study area (Figure 8). In some regions, the mean wind speed and accumulated precipitation (for 14 days prior to melt) are strongly correlated with the starting time of the melt, while the mean air temperature is not (Figure 12), thus supporting the multivariate explanation. For example, wind speed affects the start of melt more than air temperature in the Southern West Siberian Plain and the Southern Byrranga Mountains. Changes in wind conditions affect snow surface scattering by affecting sublimation, mechanical metamorphism of the surface crystals, and distribution of the snow, thus affecting the albedo and depth of snow, and the length of the melt season. The distribution of impurities, such as litter from vegetation on the snow cover, can change due to wind conditions. Impurities increase both the absorption of solar energy into the snow pack and the melt rate. Precipitation (together with air temperature) correlates with the start of melt particularly in the Yablonovyy Range and south of Taymur Peninsula. The trends in the 5-year rolling mean of climatic parameters in individual grid cells show similar spatial patterns as the melt season parameters. The trends in these areas are summarized in Table 2.

**Figure 12.** The correlation (*R*2) between start day of melt and (**a**) mean air temperature (**b**) total precipitation and (**c**) wind speed for 14 days prior to the onset of melt.
