3.1.3. Field Measurements

We used the data from previous studies [39,40] to evaluate the VSIA performance under various sea-ice conditions in the Arctic, such as the thin first year sea ice and the mixed scenario of sea ice and open water.

Istomina et al. [39] measured the spectral albedo at six ice stations in the central Arctic using ASD FieldSpec Pro 3 during the POLARSTERN cruise ARK-XXVII/3 (IceArc) in 2012. At each station, they obtained surface albedo measurements every 10 m along 200 m transects at 1 m height. A variety of land cover types were observed, including ice, snow, pond, and their mixtures. The spectral albedo was converted to broadband albedo using the solar spectral radiation distribution observed by Hudson et al. [41]. The solar irradiance was corrected to the local measurement time according to the solar zenith angle. All samples at each station were averaged to represent the overall albedo at the pixel scale. The temporally averaged VIIRS albedo during the measuring period was calculated to compare with the average broadband albedo.

Dou et al. [40] measured the spectral snow albedo on the frozen gulf nearby Barrow, Alaska, in 2015. Their measurements were carried out from April to June. The covered land surface types included snow, melting snow, and refrozen snow crust on ice. Spectral albedo was sampled every 5 m along a 100 m line using the ASD FieldSpec Pro 3. Four days of daily mean broadband albedo values were reported, which were used to validate the daily averaged satellite albedo. The dates were 26 April, 18 May, 22 May, and 25 May, respectively.

#### *3.2. Comparison between PROMICE and VSIA Match-Ups*

The PROMICE-VSIA match-ups span more than five years. The overall scatter plot is illustrated in Figure 3. Comparisons of albedo measurements at each PROMICE station are presented in Figure 4 to show further detail. Additionally, we present the box plot whiskers to show the distribution of the residuals at each station in Figure 5.

Figure 3 demonstrates reasonable agreemen<sup>t</sup> between VSIA albedo and the ground reference. Statistical measures that are used to describe these comparisons include the correlation coefficient, the mean difference error (bias), the standard deviation error (precision), and the root mean square error (RMSE) between the two. 1) The correlation coefficient R is 0.947, which indicates an obvious consistency between the two datasets. 2) The bias *e* is 0.028 (about 4.56% relative bias) and the precision *std*(*e*) is 0.066. Here, the bias is calculated as the mean difference between the VSIA and the PROMICE albedo; the precision is calculated as the standard deviation of the difference, showing the spread of the error. The small positive bias shows that VSIA provides slightly higher albedo estimates than the PROMICE measurements. 3) The overall RMSE (root mean square error) is 0.072 and the relative RMSE (Se/Sy) is 0.344. Se/Sy, i.e., the ratio of RMSE to the unbiased standard deviation of the in situ albedo. The goodness of fit indicates a high accuracy of the albedo retrievals using the rule of thumb that Se/Sy < 0.5 represents good accuracy.

The albedo values span from 0.2 to 1, which means that the dataset might contain observations of ice/snow, water ponds, bare ground, or a mixture of those, indicating the extensive representativeness of the sample. The goodness of fit is comparable with the previous validation result of MOD10A1, a regional albedo product over Greenland, Iceland and the Canadian Arctic Region, by Ryan et al. [37]. After post-processing including de-noising and calibration of the MOD10A1 albedo (Collection 6), its RMSE with PROMICE ranges from 0.017~0.1 over three KAN sites between 2009 and 2016. In summary, the validation of VSIA shortwave albedo against ground measurements showed promise at the PROMICE sites.

At many sites, an overestimation of VSIA albedo is exhibited in high albedo regions, as shown in Figures 4 and 5, such as SCO-L, TAS-L, QAS-L, QAS-U, KAN-U. These sites contain both the south (TAS, QAS, KAN) and north (SCO) stations, covering a broad latitude range. In combination with the box plot of error between VSIA and in situ albedo, as shown in Figure 5, it is shown that most stations overestimate albedo by about 0.05. An exception of underestimation happens at NUK\_L and KAN\_L as a large fraction of their observations lie in the lower value regions. These two sites are located in the southern Greenland ice sheet at 500~700 m elevation, having more obvious characteristics of the ablation zone. The land cover type transitions from dry snow to melting snow to glacier ice at melting season.

**Figure 3.** Comparison between VSIA and PROMICE clear-sky in situ albedo over 18 automatic weather stations. The match-up from each site is assigned as one specific color. RMSE: root mean square error.

**Figure 4.** Comparison between VSIA and PROMICE-measured albedo at each station. Each sub-scatterplot corresponds to one station (station name as the title). The horizontal axis represents the PROMICE-measured albedo, and the vertical axis corresponds to the VIIRS-retrieved albedo. Labels: n-sample size, A-accuracy (bias), P-precision (the standard deviation of the difference between retrieved albedo and the corresponding measured albedo), U-uncertainty (the root mean square error).

The boxplot also suggests that 14 of the 19 stations contain the zero bias within μ ± σ data range, demonstrating that the differences between the VSIA albedo and the PROMICE station data are in general indistinguishable from zero. The sites with a larger spread of albedo error suffer from the smaller sample size, making it difficult to reach solid conclusions on the product's performance at these sites. It is also illustrated that the mean values of most stations are larger than the median value, suggesting the distribution of residuals are skewed and the higher albedo points are dominant.

**Figure 5.** Distribution of residuals between VIIRS-retrieved albedo and PROMICE-measured albedo at each site. The box plots denote the distribution of the difference between VIIRS LSA and PROMICE LSA at each station. It shows lower quartiles, medians, and upper quartiles in the central boxes. Whiskers extend from each central box to show the standard deviation spread of the difference. The width of the box is proportional to the sample size at each site

According to the site-specific validation results, VSIA albedo demonstrates strong sensitivity to in situ surface albedo fluctuation and anomaly. (1) VSIA shows lower albedo value at NUK\_L and KAN\_L, which are the lower stations on the southwest Greenland coast, while demonstrating higher albedo values at other sites. NUK and KAN are two typical regions reported of albedo anomaly due to darker-than-average ice caused by a stronger warming trend, higher melt, and less winter accumulation [42–44]. Their bias derives from many factors. First, it is mainly attributable to the scale difference. The in situ measurements have a much smaller footprint than the satellite retrievals. The higher impurity concentration around these two sites leads to stronger surface spatial variability, especially in melt seasons [45,46]. Second, the bias might be ascribed to the absence of some dark zone constituents in the VSIA algorithm, including algae, crevassing, supraglacial water, and cryoconite distributed in the Greenland ice sheet [45]. The third attribute is the topography's effect. Some of the PROMICE stations are located on slightly sloping terrain. Sloping terrain alters the incident radiation composition and solar/view zenith angles, which would introduce wavelength dependent uncertainties in satellite albedo retrievals [47]. (2) For most regions, the upper station has lower bias than the lower station, such as SCO\_U and SCO\_L, TAS\_U and TAS\_L, UPE\_U and UPE\_L, and THU\_U and THU\_L. Lower sites have earlier melt onset and suffer from severe fluctuations. The scale difference between satellite retrieval and ground measurements is amplified. (3) The largest RMSE appears at the north tip site and the smallest RMSE corresponds to the south tip site. RMSE shows a slightly negative relationship with the latitude belt since southern Greenland has a much warmer summer and earlier melt than the northern regions.
