Accurate measurements and simulations of Greenland Ice Sheet (GrIS) surface
albedo are essential, given the role of surface albedo in modulating the
amount of absorbed solar radiation and meltwater production. In this study,
we assess the spatio-temporal variability of GrIS albedo during June, July,
and August (JJA) for the period 2000–2013. We use two remote sensing products
derived from data collected by the Moderate Resolution Imaging
Spectroradiometer (MODIS), as well as outputs from the Modèle
Atmosphérique Régionale (MAR) regional climate model (RCM) and data
from in situ automatic weather stations. Our results point to an overall
consistency in spatio-temporal variability between remote sensing and RCM
albedo, but reveal a difference in mean albedo of up to
Over the past decade, the Greenland Ice Sheet (GrIS) has simultaneously experienced accelerating mass loss (van den Broeke et al., 2009; Rignot et al., 2011) and records for the extent and duration of melting (Tedesco et al., 2008, 2011, 2013; Nghiem et al., 2012). Increased melt over Greenland has been associated with both changes in temperature and an amplifying ice-albedo feedback: increased melting and bare ice exposure reduce surface albedo, thereby increasing the amount of absorbed solar radiation and, in turn, further amplifying melting (Box et al., 2012; Tedesco et al., 2011). Recent studies (van den Broeke et al., 2011; Vernon et al., 2013) also indicate that albedo plays an essential role in the GrIS surface energy balance, and consequently, the surface mass balance (SMB) of regions where considerable melting occurs. Due to the impact of albedo on the surface energy balance, it is crucial to assess the performance of models that simulate albedo over the GrIS and the quality of albedo estimates from remote sensing or in situ observations. These assessments are pivotal for improving our understanding of the physical processes leading to accelerating mass loss, and for improving future projections.
Several studies investigating GrIS albedo trends and variability have primarily relied on satellite measurements, particularly those collected by the Moderate Resolution Imaging Spectroradiometer (MODIS; Stroeve et al., 2005, 2006, 2013; Box et al., 2012). Remote sensing measurements can continuously capture changes at large spatial scales and for long periods, with the exception of cases when the surface is obscured by clouds. Previous studies have found MODIS albedo products to agree reasonably well with in situ data, especially with regards to capturing the seasonal albedo cycle and mean seasonal values in regions where variability is small (Stroeve et al., 2005, 2006, 2013), but lower accuracy at high solar zenith angles has been identified (Stroeve et al., 2005, 2006), limiting the periods and locations for which these data can be used. Nevertheless, given their relatively high temporal and spatial resolution, these products are useful for evaluating albedo derived from regional climate models (RCMs). RCMs are an important tool for estimating both current and future changes in the GrIS SMB (Box and Rinke, 2002; Box et al., 2006; Ettema et al., 2009; Fettweis et al., 2007, 2011; Rae et al., 2012; Tedesco and Fettweis, 2012), and the surface albedo schemes employed by these models have a substantial impact on their simulation of the SMB (Rae et al., 2012; van Angelen et al., 2012; Lefebre et al., 2005; Franco et al., 2012).
In this paper, we report the results of an assessment of GrIS albedo spatio-temporal variability and trends for the period 2000–2013. To our knowledge, this is the first time a multi-tool integrated assessment of albedo over Greenland is presented. We use (1) data from two remote sensing products from MODIS, the MOD10A1 daily albedo product (Hall et al., 2012) and MCD43A3 16-day albedo product (Schaaf et al., 2002), (2) in situ albedo data from the Greenland Climate Network (GC-Net, Steffen et al., 1996) and Kangerlussuaq-Transect (K-Transect; van de Wal et al., 2005), and (3) outputs from two versions (v2.0 and v3.2) of the Modèle Atmosphérique Régionale (MAR; Fettweis et al., 2013a, b). In order to carry out comparisons between products, MODIS data have been re-gridded to the MAR model grid and, where necessary, daily data have been averaged over 16-day periods. The role of potential errors associated with differences in spectral range between satellite and in situ data and cloud cover have been considered and corrected for where possible.
MAR v3.2 mean September 2000–August 2013 SMB (mWE yr
The MAR model (Gallée and Schayes, 1994;
Gallée, 1997; Lefebre et al., 2003) is a coupled
land-atmosphere RCM featuring the atmospheric model
described by Gallée and Shayes (1994) and the Soil Ice Snow Vegetation
Atmosphere Transfer scheme (SISVAT) surface model. SISVAT incorporates the
multilayer snow model
In contrast with MAR v2.0, MAR v3.2 sub-grid scale parameterizations make it possible to have fractions of different land cover types within a single grid box. Quantities were computed for the sectors within each grid box and a weighted average of these quantities was used to represent the average value for a grid box.
For convenience, mean SMB for September 2000–August 2013 from MAR v3.2 is shown in Fig. 1, along with the equilibrium line dividing positive and negative SMB, together with the locations of the weather stations used in this study. In this study, areas below the mean 2000–2013 equilibrium line as defined by MAR are collectively referred to as the “ablation area”, while areas above this line are referred to as the “accumulation area”.
The basis for the MAR albedo scheme is described in detail by Brun et al. (1992) and
Lefebre et al. (2003). MAR snow albedo (
The minimum albedo of snow is set to 0.65. In MAR v2.0, bare ice albedo is
simply assigned a fixed value. In MAR v3.2 (the version primarily used
here), bare ice albedo is a function of accumulated surface water following
the parameterizations of Lefebre et al. (2003), described below. In the case
of bare ice, which occurs in MAR when the surface snow density is greater
than 920 kg m
Range of possible albedo values for different surface types in MAR v2.0 and v3.2
Additionally, to ensure temporal continuity in simulated albedo, values of
albedo between the maximum bare ice and minimum snow albedo are possible
when the surface (or near-surface) snow density lies between 830 and 920 kg m
In cases where there is a snowpack with a thickness of < 10 cm
overlaying ice or firn (with a density greater than 830 kg m
We used the daily MODIS albedo product (MOD10A1, Version-5) distributed by
the National Snow and Ice Data Center (Hall et al., 2012; available at
The MOD10A1 Version-5 product contains daily albedo (0.3–3
GC-Net and K-Transect weather stations used in this study and years of coverage.
The MCD43A3 Version-5 product makes use of all atmospherically-corrected
MODIS reflectance measurements over 16-day periods to provide an integrated
albedo measurement every eight days. A semi-empirical bidirectional reflectance
function (BRDF) model is used to compute bi-hemispherical reflectance as a
function of these reflectance measurements (Schaaf et al., 2002). The
MCD43A3 product contains, in addition to albedo values for each MODIS
instrument band, “short-wave” (SW) albedo values calculated over a wavelength
interval of 0.3–5.0
Both MODIS products provide quality flags indicating “good quality” vs. “other quality” data. In the case of MCD43A3, “other quality” data were produced using a backup algorithm. When few observations were available, the backup algorithm was used to scale an archetypal BRDF based on past observations (Schaaf et al., 2002). In order to understand the influence of data quality on our results, we present results for both “all quality” as well as “good quality” data.
We used automatic weather station (AWS) data from two sources,
GC-Net (Steffen et al., 1996) and the
K-Transect (van de Wal et al., 2005). The locations of the weather
stations are shown in Fig. 1, and a list of the weather stations used and
their period of coverage is provided in Table 2. We used all available GC-Net
and K-transect June–July–August (JJA) data within the period 2000–2012 for
comparison with MODIS and MAR albedo. GC-Net data for the summer of 2013
were not available when data analysis for this study was conducted. We
followed a procedure similar to that used by Stroeve et al. (2005) to
generate an albedo time series from GC-Net and K-Transect data. Mean daily
albedo was computed as the sum of daily incident SW radiation
divided by the sum of daily outgoing SW radiative flux. Instances where
hourly upward SW radiative flux exceeded downward SW radiative flux were
excluded. Upward and downward hourly radiative fluxes were excluded when
downward fluxes were smaller than 250 W m
Snow albedo is generally higher during cloudy conditions due to the masking
of a portion of the incoming solar spectrum by clouds (Greuell and
Konzelman, 1994). Both MAR v3.2 and MAR v2.0 account for this by
applying a correction to
For the purpose of comparing model results and satellite data, MODIS albedo products were re-gridded to the MAR 25 km resolution grid from the original 463 m spatial resolution at which they are distributed. Re-gridded values contain the median value of all the MODIS values falling within a MAR grid box. When comparing satellite data against model results, our analysis was restricted to the GrIS. For all comparisons including MAR v3.2 results, areas where the MAR sub-grid level ice cover percentage was less than 100 % were excluded. For all comparisons including MAR v2.0 results, the same mask from MAR v3.2 was used, except pixels classified as 100 % ice covered in MAR v3.2, but classified as non-ice-covered in MAR 2.0 were also excluded from the analysis.
Comparisons at in situ stations were conducted between weather station data
and data or outputs from the MODIS or MAR grid box that encompassed the in
situ station. In this case, we used the original (
As in the case of the original MAR albedo outputs, in situ measurements also included measurements made during cloudy conditions while MODIS albedo data did not. Given a lack of available measurements, we did not explicitly correct in situ data for the presence of clouds in this study, but only considered data where coincident satellite and in situ measurements were available. Stroeve et al. (2013) applied a correction to GC-Net data using a radiative transfer model, but found that the correction did not significantly impact results.
The GC-Net LI-COR sensors are sensitive within the 0.4–1.1
In the following analysis, we focus on the JJA period because MODIS data are less reliable during other months, when solar zenith angles are high, as discussed by Box et al. (2012), and because this is the period when surface albedo has the largest impact on SMB.
In order to compare spatial variations in albedo we calculated the mean
2000–2013 JJA MOD10A1, MCD43A3 SW BSA albedo, and MAR clear-sky
albedo using all available measurements or model outputs over the specified
period, excluding cases where greater than 25 % of data were missing for a
given pixel. When differences between data sets or between satellite data and
model results were calculated, we only used measurements or results
overlapping in time and space, to avoid the possibility of bias introduced by
missing data. The mean difference between two samples for a given grid box
was deemed to be statistically significant if the
In some cases, observational data or model results have been spatially averaged or aggregated within the ablation and accumulation areas defined using MAR v3.2 or MAR v2.0. The ablation (accumulation) area is defined as the area that experienced a net loss (gain) of mass over the 2000–2013 period, as simulated by either version of the model.
For analyses of temporal variability, we considered daily variability, for
which MOD10A1 data, in situ values, and MAR outputs were available, as well as
variability over 16-day MCD43A3 periods. In the case of the analysis of
16-day data, MOD10A1, MAR, and in situ daily data were averaged to produce a
value for each overlapping MCD43A3 16-day period. The correlation
between daily satellite data and between satellite data and model results was examined
using Pearson's coefficient of determination (
To compare the distribution of ablation area albedo for satellite data and MAR model outputs, we produced frequency histograms for ablation area albedo using a bin width of 0.0099. Parameters for the best fit of a bimodal distribution to the histograms was obtained using the maximum likelihood estimation function in MATLAB, assuming a bimodal normal distribution for the fit.
Box et al. (2012) investigated changes in GrIS albedo using the MOD10A1
albedo product, and found that between 2000 and 2012, surface albedo decreased
over almost the entire ice sheet. Here, we built on the analysis of Box et
al. (2012) and extended our analysis to include MCD43A3, MAR v3.2, and in situ
JJA data for the period 2000–2013. Trends in albedo have been obtained by
performing linear regression on 16-day albedo values for satellite products,
in situ data, and model outputs, excluding albedo values outside of the JJA
period. We have also computed trends for annual JJA average values. A trend
was determined to be statistically different from 0 if the
Mean 2000–2013 JJA albedo (unitless)
for
Mean 2000–2012 JJA GrIS albedo, for MOD10A1, MCD43A3 BSA SW, and MAR clear-sky albedo, averaged within the mass balance areas shown in Fig. 1 and Table 2. Only good quality MODIS data are used here. All data have been averaged over the same 16-day period of the MCD43A3 product. Only periods when coincident data for all data sets were available have been included.
MAR v3.2 and the two MODIS data sets show coherent spatial patterns of JJA
mean 2000–2013 albedo (Fig. 2) that are consistent with previous studies
(e.g. Box et al., 2012), with low-elevation areas in the ablation area
dominated by lower albedo values (< 0.7 on average, Table 3) due to
the presence of meltwater and bare ice, and high elevation areas by
relatively higher albedo (> 0.74). The most obvious discrepancy
between the satellite products occurs north of 70
Mean difference in JJA albedo (unitless) for the 2000–2013
period:
Same as Table 3, but for the average of 16-day data for all in situ stations within each region.
2000–2012 mean JJA albedo (unitless) for the MAR accumulation zone vs. latitude, for MOD10A1, MCD43A3 BSA SW, MAR v3.2 clear-sky, and GC-Net station data (black circles) for stations with a record spanning at least seven years of the 2000–2012 period. Only MODIS data flagged as “good quality” were used here. The error bars for GC-Net stations indicate the range of corrections to GC-Net data (between 0.04 and 0.09) employed by Stroeve et al. (2005).
The pattern of differences between MAR v3.2 and the two satellite products
(Fig. 3b and Fig. 3c) appears to vary with both elevation and latitude, while the
difference between the two satellite products varies primarily with latitude
(Fig. 3a). Because any systematic biases in the satellite products are
likely to be relatively consistent across space (at least as a function of
longitude), it is likely that MAR v3.2 biases contribute to some of the
elevational differences seen in Fig. 3b and Fig. 3c. Within the accumulation area
south of 70
MAR v3.2, MOD10A1 and MCD43A3 mean 2000–2013 JJA albedo values show a
similar logarithmic dependence of albedo with elevation (Fig. 4a); below
2000 m, albedo increases relatively rapidly with elevation (both MAR and the
MODIS products show a statistically significant albedo increase of
For in situ stations in the ablation area (Table 4), in situ mean albedo
(0.56
MCD43A3 BSA visible (0.3–0.7
Mean 2000–2012 JJA albedo values for ablation area GC-Net stations with a
record of at least seven years do not appear to exhibit a clear variation with
latitude when compared with satellite data and model results (Fig. 5).
GC-Net albedo at stations north of 70
It appears possible from Fig. 5 that the bias at GC-Net sites (between 0.04
and 0.09 according to Stroeve et al., 2005) could increase with latitude,
rendering corrected GC-Net mean 2000–2013 albedo comparable to MCD43A3
albedo. In order to indicate how the GC-Net albedo bias is likely to vary
spatially, the mean difference between MCD43A3 visible BSA (for the interval
0.3–0.7
The spatial variability of the difference appears to be associated with the differences in spectral albedo between different materials. Because ice does not exhibit the spectral dependence of albedo that snow does (Hall and Martinec, 1985), the difference between MCD43A3 visible and SW albedo is lower in the ablation area where bare ice is exposed during summer. In locations where melting occurs, snow grains tend to be larger because of constructive metamorphism, reducing reflectance mostly in the near infrared band (Wiscombe and Warren, 1980), resulting in a larger difference between visible and near infrared reflectance. This suggests that in situ albedo values do not exhibit the decrease of albedo with latitude indicated by MCD43A3.
The standard deviation of an albedo time series provides information on the magnitude of its temporal variability. Within the low elevation ablation area of the ice sheet, both MAR and the MODIS products exhibit a relatively high standard deviation for the 2000–2013 period (0.07 on average for 16-day periods; Fig. 7, Table 3). At high elevations, variability is smaller (0.02 to 0.03 on average for 16-day periods). The MCD43A3 and MOD10A1 products show similar spatial patterns of standard deviation when the daily product is averaged over 16-day MODIS periods (Fig. 7a and Fig. 7c). Table 3 suggests that MAR v3.2 ablation area temporal variability is identical to MODIS variability on average, but Fig. 7 shows there are locations, particularly within the west coast ablation area, where MODIS variability is considerably higher. MAR v3.2 albedo variability in low elevation areas reaches a maximum of 0.09, while MODIS variability for the same region is 0.15 at maximum.
Standard deviation of JJA albedo (unitless) (2000–2013) for
Coefficients of determination (
At a daily temporal resolution, MOD10A1 daily variability in the ablation area (0.17 maximum, 0.07 on average) is considerably larger than the variability of MAR v3.2 albedo (0.12 maximum, 0.04 on average). As will be discussed in Sect. 4.2, this may be the result of a positive bias in bare ice albedo from MAR, but may also be associated with errors introduced by cloud artifacts in the MOD10A1 product. For the accumulation area, the standard deviation of albedo for MAR and MODIS generally falls within the 16-day uncertainty of 0.04 for MCD43A3 high-quality albedo and daily uncertainty of 0.067 for MOD10A1 albedo estimated by Stroeve et al. (2005, 2006). This limits the comparison among MAR and the MODIS products for high elevations.
Scatter plots for 2000–2013 JJA albedo for
For areas south of 70
Further insights into the consistency of spatio-temporal variations in albedo
between MODIS products and between MAR and MODIS products can be drawn from
scatter plots for all MCD43A3 vs. MOD10A1 2000–2013 JJA albedo values (Fig. 9a)
and MAR vs. MODIS values (Fig. 9b–d). Figure 9a indicates that MCD43A3
albedo is lower (by 0.03 on average) compared to MOD10A1 albedo, consistent
with the significant difference between the products at high latitudes seen
in Fig. 3a. There is a fairly good correlation between MCD43A3 and MOD10A1
(
When MAR is compared with MCD43A3 and MOD10A1 over 16-day periods (Fig. 9b
and Fig. 9c), the correlation between MAR and satellite data is as good or better
than the correlation between MOD10A1 and MCD43A3 (
Scatter plots of 2000–2012 JJA mean albedo [unitless] vs.
automatic weather station (GC-Net and K-Transect) albedo:
Scatter plots of 2000–2012 JJA albedo values for both satellite products and
MAR v3.2 vs. all weather station measurements (Fig. 10) indicate a strong
correlation between in situ data and the two satellite products over 16-day
periods (Fig. 10a and Fig. 10b;
On a daily basis, MOD10A1 albedo exhibits a nearly 1
In situ and satellite data and MAR v3.2 outputs all indicate that
spatio-temporal variability of albedo is higher in the ablation area (where
the standard deviation of albedo is
In order to further examine some of the discrepancies between MAR and observations, it was useful to examine differences between MAR v3.2 and MAR v2.0. MAR v2.0 has been validated against satellite and in situ data (e.g. Fettweis et al., 2005, 2011) and used for making future projections (Fettweis et al., 2013b; Tedesco and Fettweis, 2012). A major difference between MAR v3.2 and MAR v2.0 is in the scheme for calculating the albedo of bare ice; MAR v2.0 bare ice albedo is set to 0.45, while in MAR v3.2 it ranges between 0.45 and 0.55 as a function of surface melt (Table 1).
Scatter plots for MAR vs. MODIS 2000–2012 JJA albedo in the ablation area,
along with frequency histograms and best fit curves of the distribution
(Fig. 11), suggest there is a bimodal distribution of ablation area
albedo, which we attribute to the presence of two main surface types: ice
(and firn) and snow. Pixels classified by MAR as having bare ice (or firn,
surface density > 830 kg m
Mean and standard deviation for the best fit to the distributions of ablation area albedo shown in Fig. 11 (assuming that the appropriate distribution is a combination of two normal distributions).
Scatter plots and histograms for JJA 2000–2012 albedo
[unitless] within the MAR v3.2-defined GrIS ablation zone, for
However, there are differences in the observed distributions. MAR v2.0
exhibits a clustering of albedo values above 0.65 and below 0.55 (Fig. 11a).
MCD43A3 exhibits an overlap in the distribution of the two modes, and there
is a wider range of low albedo values (
We compared MAR v2.0 mean 2000–2012 clear-sky JJA albedo with albedo from MAR
v3.2 and MODIS in Fig. 12a–c. MAR v3.2 albedo is significantly larger
in the ablation area compared with MAR v2.0 (Fig. 12a). Rather than being
positively biased relative to MODIS (as is the case for MAR v3.2; Fig. 3), MAR v2.0 albedo is either negatively biased or is not significantly
different from MODIS data (Fig. 12b and Fig. 12c). The difference in albedo scheme
is the major difference between MAR v3.2 and MAR v2.0, and it results in a
significant difference in SMB (Fig. 12d). The average ablation area JJA SMB
for MAR v3.2 is higher by 0.53 mWE yr
MAR v3.2, MCD43A3, and MOD10A1 consistently agree that there has been a
significant decrease in albedo within the ablation area over 2000–2013, and
that the largest decreases in albedo have occurred below 2000 m a.s.l.
(Fig. 13 and Fig. 14). MCD43A3 shows a decrease of up to
JJA mean albedo trends (2000–2013) in units of fraction per
decade for
Within the accumulation area, MAR v3.2 disagrees with the two MODIS products
as to the direction and magnitude of trends. MCD43A3 shows a decrease of
Mean annual JJA ice sheet albedo (solid lines) simulated by
MAR v3.2 (clear-sky; blue), MOD10A1 (black) and MCD43A3 BSA SW
(orange) for 2000–2013 and best linear fit (dashed lines) for
For locations within the GrIS ablation area, trends at GC-Net stations with
a record of at least nine years are consistent with significant decreases in
albedo, indicated by MODIS and MAR for the periods covered (2000–2012 or
2004–2012; Table 6). The magnitude of the trends varies between MAR v3.2,
MODIS, and in situ data at individual stations. These differences can be
attributed in part to the high spatio-temporal variability of albedo within
the ablation area. This can potentially lead to trends at a weather station
that are substantially different from trends within a 500 m MODIS grid box
containing the location of that weather station. At higher elevations, this
factor is less important as there is less spatio-temporal variability in
albedo (Fig. 9 and Fig. 10). Within the accumulation area, trends at weather
stations are generally within
The results presented above highlight certain features of GrIS
albedo variability that are common to in situ, satellite data, and
model results. MAR, MODIS, and in situ data capture general spatial patterns
of low albedo in the ablation area, which increases with increasing
elevation below
As noted in Sect. 3.5, MAR, MODIS, and in situ data agree there has been a significant decline in ablation area albedo between 2000 and 2013. These trends in surface albedo are associated with increased melting and bare ice exposure resulting in a decline in ablation area SMB, captured by models (Fettweis et al., 2011; Ettema et al., 2009) and in situ observations (van de Wal et al., 2012). Increased melting has been linked to higher regional atmospheric air temperatures, associated with atmospheric circulation changes (Fettweis et al., 2013a; Häkkinen et al., 2014).
Results from Sect. 3.1 indicate that above 70
Theoretically, snow albedo is expected to increase with increasing solar
zenith angle, particularly for high solar zenith angles (Wiscombe and
Warren, 1980), and, therefore, will increase slightly at high
latitudes, as long as other factors do not contribute to lower albedo
values. Wang and Zender (2009) compared 16-day MCD43C3 albedo with GC-Net
measurements and suggest that the MCD43C3 product is unrealistic at higher
latitudes, in particular for solar zenith angles > 55
Trends (and 95 % confidence intervals) in JJA albedo (fraction per decade) at GC-Net and K-Transect weather stations and the nearest MOD10A1, MCD43A3, and MAR pixels. In this case, MODIS data flagged as “other quality” have been included. Only 16 day periods when coincident estimates are available for all data sets have been used. Values in bold indicate trends significant at the 95 % confidence level.
It should also be noted that the MOD10A1 product may be
positively biased above 70
The major difference between MAR v3.2 albedo and observed albedo is an
overall positive bias in the ablation area. This bias can be seen most
clearly as a difference of
Scatter plots of ablation area albedo appear to confirm this: when MAR v3.2 is compared with both MODIS data and in situ measurements (Figs. 9b, 9c and 10c) the result is a best fit line with a slope smaller than one. Additionally in the same area where MAR v3.2 appears positively biased in the west coast ablation area, MODIS exhibits relatively high variability compared with MAR v3.2 (as discussed in Sect. 3.2; Fig. 7).
Biases in MAR ablation area albedo are related to its ability to capture the observed bimodal distribution in ablation zone albedo (Fig. 11) associated with two main surface types, ice and snow. The positive bias from MAR v3.2, as well as the relatively low modelled variability in the ablation area is the result of the albedo values set for bare ice in MAR v3.2 (Table 1) that may be too high on average. MAR v2.0 albedo, by contrast, which has a fixed bare ice albedo of 0.45, generally exhibits a negative bias in most portions of the ablation zone. A bare ice albedo that is too high will also lead to a smaller difference between the albedo values of melting snow and bare ice, reducing temporal variations in ablation area albedo, resulting in the relatively low variability from MAR v3.2 (Fig. 7).
An examination of Fig. 11 indicates that the low albedo peak for MAR v3.2 is
closer to being normally distributed compared with the peak for MAR v2.0,
and is, therefore, a better match to the distribution from MCD43A3. However, MAR
v3.2 overestimates bare ice albedo, as already
discussed, and still does not fully capture variability in the low
albedo peak for MODIS albedo (
MAR albedo is only a function of accumulated meltwater and does not explicitly take into account the presence of dust, surface lakes and surface streams, including the West Greenland “dark zone” (van de Wal and Oerlemans, 1994; Wientjes and Oerlemans, 2010), which reduces bare ice albedo and likely introduces increased ablation area albedo variability. Assigning a wider range of MAR albedo values for bare ice (which has been implemented in the most recent release of MAR, v3.4) may improve its representation of the distribution of bare ice albedo, but may not necessarily improve its ability to capture the spatial distribution of ablation area albedo. This could potentially be achieved through the inclusion of an explicit representation of dust and sub-grid-scale hydrology in the model.
As noted in Sect. 3.5, there is a discrepancy between the satellite
products, in situ data, and model results regarding albedo trends in the
accumulation area of the ice sheet. MOD10A1 and MCD43A3 show significant
decreases in accumulation area albedo (
A possible explanation for this discrepancy is that MODIS trends are negatively biased as a result of declining instrument sensitivity of the MODIS sensors (Wang et al., 2012). In particular, a larger degradation has been observed for the MODIS Terra satellite (Wang et al., 2012). The MCD43A3 product uses data from both the Terra and Aqua satellites, while MOD10A1 only uses data from Terra. This could potentially explain the larger trends for MOD10A1 relative to MCD43A3 (Table 6, Fig. 13). Box et al. (2012) conclude that declining instrument sensitivity does not substantially affect GrIS albedo trends, as larger trends are found in GC-Net data relative to MOD10A1 for 70 % of cases where trends are deemed to be significant. In contrast to the findings of Box et al. (2012), we did not find JJA GC-Net trends larger than those of MODIS, except in some instances within the ablation area, where there is high local variability in surface properties. The analysis performed here differs from that employed by Box et al. (2012). Differences in trends found in this study may have resulted from a focus on trends for the entire JJA period rather than on monthly trends, and calculated trends for 16-day albedo values rather than calculated monthly albedo from integrated fluxes over a 1-month period, as was done by Box et al. (2012). We also investigated the possibility that the smaller spectral interval of GC-Net data influences trends by comparing MCD43A3 visible vs. SW albedo trends, but did not find the trends to be significantly different from each other. We are not able to confirm that the larger trends from MODIS are associated with declining instrument sensitivity, as this analysis is outside the scope of this study. However, the findings of this study seem consistent with this possibility and it is suggested as a topic for future research.
We have examined spatio-temporal variability and trends in GrIS albedo using in situ measurements, satellite products obtained from MODIS data, and outputs of two versions of the MAR RCM. The results presented here reveal areas of agreement as well as discrepancies between observational and model estimates of GrIS albedo spatio-temporal variability. Examining local measurements, satellite data, and model results concurrently reveals information about the GrIS albedo and potential biases that would not be revealed by examining observational data sets or model results individually.
The results presented here show that albedo varies spatially as a function primarily of surface properties, in particular melting and bare ice exposure in the ablation area. These factors are also associated with temporal variations in albedo, resulting in high variability in low elevation regions. The differences in variations with latitude indicated by satellite products appear likely to be a function of inaccuracies associated with the products themselves, rather than a record of actual variations in surface albedo, particularly as the two products are derived from the same MODIS sensors.
Both satellite products and MAR model outputs (for v2.0 and v3.2) suggest a bimodal distribution of surface albedo within the ablation area of the ice sheet. Based on model results, we infer that this distribution is associated with the presence of two primary surface types within the ablation area, snow and bare ice. The model's inability to capture the full range of low elevation albedo leads to inaccuracies in the representation of spatio-temporal variations in albedo, which can substantially impact the representation of SMB. The MAR version examined here (v3.2) appears to better represent the full range of bare ice albedo in the ablation area relative to a previous version (v2.0), but a lower minimum bare ice albedo value (as is implemented in the next version of MAR, v3.4), may produce results that are more consistent with observations. Even so, it may be necessary to account for the presence of impurities and sub-grid- scale hydrology in order to fully capture spatial variations in albedo.
The analysis performed here indicates a statistically significant decrease in ablation area albedo over the period 2000–2013 and is consistent with previous studies (Box et al., 2012; Tedesco et al., 2011, 2013; Stroeve et al., 2013). This decrease is consistent with a coincident decline in ablation area SMB recorded by both models and observations (e.g. Fettweis et al., 2011; Ettema et al., 2009). Our results are inconclusive regarding high elevation trends in albedo; we observe inconsistencies between satellite-derived trends and trends obtained from in situ measurements and MAR v3.2 results. We are therefore unable to confirm previously reported decreases in surface albedo at high elevations.
Future research should be directed towards understanding the reasons for discrepancies between satellite products, in situ data and model results, in order to better understand changes in GrIS albedo. This includes resolving discrepancies regarding high-elevation trends, and discrepancies in mean satellite-derived surface albedo at high latitudes. Models such as MAR appear to be effective at capturing surface albedo, but refinements are necessary for representation of surface albedo in low elevation areas. In particular, the representation of bare ice albedo is critical. Sensitivity studies, such as those performed by van Angelen et al. (2012) of the impact of surface albedo on SMB variability, may help to quantify the accuracy with which surface albedo must be modelled for a given region. Analysis of spatio-temporal variations in albedo across different spatial scales (including at a higher spatial resolution than has been examined here) may also become increasingly important as models operate at higher spatial resolutions, and as we seek to understand the GrIS surface mass and energy budget in greater detail. Given the strong relationship between surface albedo and SMB, future studies are crucial for efforts aimed at estimating and predicting the impact of current and future climate change on GrIS SMB.
This research is supported by NSF Grant 0909388. In situ data along the K-transect are financed over time by Utrecht University, and several grants from the Polar Program of the Netherlands Organization for Scientific Research (NWO) including the Spinoza Program, and by the Royal Netherlands Academy of Sciences (KNAW). The authors would like to thank Rajashree Datta, Jason Box, and two anonymous reviewers for comments and suggestions that improved the manuscript. Part of this work was performed while Marco Tedesco was serving as Program Director at the National Science Foundation. Edited by: A. Klein