The surface albedo of the Greenland Ice Sheet's ablation area has been shown to play a crucial role in determining melt variability through the snow/ice albedo feedback effect (Box et al., 2012). The study of these albedo changes over the last decade and a half from modern satellite sensors such as MODIS has drawn considerable research attention (Tedesco et al., 2011; Stroeve et al., 2013; Alexander et al., 2014; Tedstone et al., 2017). The observed albedo decreases are closely tied to the increasing mass loss of the ice sheet through the enhancement of its surface melt (Enderlin et al., 2014; van den Broeke et al., 2016). The ice sheet also loses mass through the ice calving at outlet glaciers, or discharge, the magnitude of which has recently been shown to demonstrate significant spatial and seasonal variability (King et al., 2018). The mass loss of the Greenland Ice Sheet has been shown to currently be among the leading contributors to sea level rise (Box et al., 2017a).

Legacy satellite sensors such as the Advanced Very High Resolution Radiometer (AVHRR) offer us the possibility to study the Greenland Ice Sheet's albedo changes of over a period of more than 3 decades, doubling the temporal coverage of MODIS. The drawback is lower spectral precision and coverage. Yet, considerable efforts have recently been paid to the intercalibration of the AVHRR sensor family (Heidinger et al., 2010), greatly improving the sensors' capabilities to detect long-term trends in surface albedo. Here, we use the sensor-intercalibrated 34-year (1982–2015) surface albedo data record from the CLARA-A2 dataset family (Karlsson et al., 2017a) to study the albedo changes of the ice sheet at decadal scales.

Our analyses focus on the summer months between May and August (MJJA), comprising four topics. First, we calculate and present the trends in surface albedo (defined here as the directional–hemispherical reflectance of the surface, i.e. black-sky albedo or “inherent” albedo) over the ice sheet for each respective month. In addition to the trends for the full 1982–2015 period, we present temporally subsetted trends consistent with the MODIS observation era (2000–2015) and the pre-MODIS era (1982–1999), in part to compare our results with the trends identified from the latest editions of MODIS data (Casey et al., 2017) and also to offer a rare, if somewhat more uncertain, look into the ice sheet's surface changes in the 1980s and 1990s. Treating each month individually additionally allows us to delineate between albedo changes in the early and late melting season.

Second, we examine changes in melt intensity and speed through changes in the rate of albedo decrease and in the spatial distribution of albedo values indicative of bare ice and wet snow as the summer melt progresses. At this stage we also analyse atmospheric reanalysis data to discern if the observed changes may be explained by precipitation or air temperature anomalies.

Third, we conduct an observation-driven test for the dominance of surface runoff in the interannual variability of May–August surface mass balance (SMB) of the ice sheet. We calibrate discharge-corrected, gravimetrically observed SMB with the CLARA surface albedo to form a summer-aggregated proxy SMB. When compared against the SMB from a state-of-the-art regional climate model for the ice sheet (MAR, Fettweis et al., 2017), we expect to find good covariability only if surface runoff dominates SMB variability, as that is the sole term which can be considered well-correlated with albedo changes.

Finally, we downscale the analysis to six major drainage basins of the ice sheet. Again assuming that surface albedo changes are an acceptable proxy for surface runoff production, we examine the albedo–discharge relationships to ascertain from an observational basis if inferred surface melt anomalies correlate with the interannual variability in ice discharge.

The CLARA-A2 SAL surface albedo estimates represent the per-grid-cell mean solar illumination conditions of each 5 d or monthly time period and are not normalized to any particular solar zenith angle (SZA). We choose to use a non-normalized SZA because various in situ measurements have shown the symmetrical u-shape diurnal cycle assumed in common normalization algorithms (Briegleb et al., 1986; Gardner and Sharp, 2010) is not valid for snow during the melting period, when the u shape is often noon-asymmetrical (Dirmhirn and Eaton, 1975; Jonsell et al., 2003; Pirazzini et al., 2006; Meinander et al., 2013). We find, however, that albedo trends are not highly sensitive to the selection or exclusion of SZA normalization. Figure S1 in the Supplement reproduces Fig. 3 with the application of the SZA normalization proposed by Gardner and Sharp (2010) with a target SZA of 60∘, yielding consistent trends. Also, the separate treatment for each summer month and the use of a conservative SZA cutoff (of 70∘) in the albedo estimation will further ameliorate any impacts of excluding a normalization procedure.

We find very similar spatio-temporal distributions of albedo trends and overall magnitude to those in Tedesco et al. (2016b), who performed an analysis of Greenland surface albedo changes during the summer JJA season from the GLASS dataset, based on coherently and consecutively processed AVHRR and MODIS observations. The JJA albedo trend for the whole ice sheet during 1996–2012 in CLARA-A2 is -0.020 per decade (p < 0.01), matching the result from GLASS (based on MODIS Terra and Aqua from 2000/2002 onward). They attribute the decreasing trend to increasing air temperatures and enhanced melting but also to exposure of older impurities resulting from increasing snow and ice melt. While our present analysis does not attempt to attribute changes to impurity deposition or exposure, we note that the melting seasons 2013–2015 brought about a clear amelioration in the overall albedo decrease. The JJA decadal albedo trend in CLARA-A2 for 1996–2015 is -0.01 (p < 0.05). This suggests that, if progressing impurity exposure indeed plays a role in the large-scale darkening of the ice sheet margins, cloudiness and circulation changes alongside regional precipitation increases are still the more powerful drivers.

We suggest that the recent inhibition of surface melt along the SE margin of the ice sheet is due to increased winter precipitation, but there are somewhat conflicting views in recent literature. Berdahl et al. (2018) examined observation precipitation records at Tasiilaq station, taken to be representative of the surrounding parts of the SE ice sheet, and found no significant December–February precipitation increases in the satellite era up to 2012. We note, however, that Tasiilaq station appears to lie at the no-change zone between the MERRA-2 bimodal peaks, where MERRA-2 winter precipitation anomalies during 2000–2015 increase in the southernmost parts of the SE margins but are reduced further east. While further analysis on this bimodal feature in precipitation anomalies is beyond the scope of this study, we suggest it might serve to explain the lack of observed precipitation increases by Berdahl et al. (2018). However, Koyama and Stroeve (2018) evaluated the Arctic System Reanalysis precipitation against in situ observation along the ice sheet margins, finding in situ observations indicating increasing precipitation during the 2010–2015 period near the southern tip of Greenland. These increases agree with the increasing winter precipitation signal we observe in MERRA-2 and are consistent with the lack of new albedo reductions or bare ice exposure in CLARA (Fig. 6). The suggested link between melt inhibition and increasing precipitation depends fully on the phase of the precipitation; recent evidence demonstrates that the proportion of rainfall to total precipitation is increasing, suggesting that our finding may be transient in nature (Oltmanns et al., 2019).

Wong et al. (2015) also reported consistent although regionally variable temperature and precipitation responses along the Greenland coastal regions to prevailing atmospheric circulation patterns, in particular to the warm and moist southerly airflows associated with a negative North Atlantic Oscillation (NAO) index. Their finding of increasing winter precipitation in the NW region of the ice sheet during 1990–2010 is consistent with the localized lack of systematic melt acceleration seen in CLARA over the same region (Fig. 4), even though the increasing precipitation is not seen by MERRA-2. However, the reversal of bare ice exposure in the NW when comparing 2013–2015 to 2000–2002 (Fig. 6) may also be related to more frequent cold northerly airflows post-2012 (Tedesco et al., 2014).

While our observation-based test confirms the dominance of surface melt in the surface mass loss of the ice sheet, the general agreement between modelled and proxy SMB shown here contains substantial (though mutually cancelling) disagreements at the monthly level. Most notably, the ERA-Interim-forced MAR v.3.5.2 SMB generally sees the largest annual mass loss occurring in July, whereas the corresponding GRACE SMB estimates assign the largest mass loss to the month of August. Such differences have emerged in prior studies (Alexander et al., 2016), although based on different GRACE mass balance solutions and a MAR version of different forcing and spatial resolution, invalidating direct comparisons between results. The reasonably good covariability between MAR SMB and albedo-based proxy SMB is likely due, in part, to the improved agreement between GRACE and MAR mass balance at elevations below 2000 m a.s.l. (Alexander et al., 2016), where most of the ice sheet albedo changes occur. Also, in principle snowfall acts to increase both SMB and albedo, although the correlation is short-lived, diminishing once the fresh snow layer becomes optically semi-infinite.

The ESA Greenland Ice Sheet CCI distributes mass balance solutions from two algorithms (TU Dresden and DTU). For completeness, we repeated the SMB analysis also for the DTU GRACE data. The results (Fig. S2) show a less favourable fit between CLARA and discharge-corrected GRACE SMB with a larger slope, leading to generally much less negative surface mass balance in the CLARA–GRACE proxy relative to MAR. The MOD10A1 is similarly affected but to a lesser degree. It is notable that both discharge-corrected GRACE SMB time series from ESA Greenland Ice Sheet CCI produce systematically less negative SMB during the summers 2004–2010 relative to MAR.

Geolocation errors have been reported as a source of difference between the MODIS and AVHRR albedo estimates by Box et al. (2006). To further investigate the chance that geolocation and cloud masking uncertainties have a substantial impact on the obtained results, we compared the CLARA record to the latest MOD10A1 (Box et al., 2017b) in terms of decadal albedo trends and grid-cell-specific mean differences per summer month during 2000–2015 (Fig. S3). The CLARA-MOD10A1 mean differences display a substantial latitudinal dependency in May and August, in line with results shown by Alexander et al. (2014). However, the decadal trends agree very well between the datasets, thus reinforcing the robustness of the observed trends in CLARA albedo. Discerning the cause of the differences would require a rigorous study accounting for differences in spatial resolution, diurnal sampling, and geolocation uncertainties and is as such beyond the scope of the present work.

A variety of drivers of the post-2000 increases in the ice sheet's surface melt (and thus decreases in surface albedo) have been proposed. First, atmospheric circulation pattern changes (ridging over Greenland) have led to warm and moist air advection over the ice sheet (e.g. Fettweis et al., 2013; Mioduszewski et al., 2016), leading to both increases in SAT (Reeves Eyre and Zeng, 2018) and increases in clear-sky downwelling longwave radiation (Mioduszewski et al., 2016; Välisuo et al., 2018). Changes in the surface radiative energy budget caused by cloud coverage and composition have also been proposed as a primary driver. Hofer et al. (2017) argued, using the CLARA-A1 dataset, the predecessor of CLARA-A2 used here, that decreasing cloudiness over the ablation zone in the 2000s, which increases both downwelling shortwave and clear-sky downwelling longwave radiation in the presence of advected warm air masses, has caused a significant increase in insolation and surface melt. Bennartz et al. (2013) provided a different mechanism for increased melt, arguing that, over the high-elevation centre of the ice sheet, intrusions of optically thin, low-level, and liquid-bearing clouds enhance the downward longwave flux while still allowing shortwave flux penetration to the surface, promoting extensive surface melt.

Here, we have focused on analysing the surface albedo changes using reanalysis SAT and winter precipitation as supporting evidence. We calculated the mean MJJA cloud fractional coverage over the ice sheet from CLARA-A2 (1982–2015) for comparison against the corresponding MAR cloud cover field. Figure 10 shows a recovery of cloud cover after 2007 in CLARA, from updated algorithms and AVHRR radiance calibration. MAR cloud cover, however, continued to decline until 2011. This highlights the uncertainties in cloud observations from legacy AVHRR data, even though the detection performance is improved in CLARA-A2 (Karlsson et al., 2017a). A limited study over the central part of the accumulation zone during 2010 showed good agreement between CLARA and MODIS cloud cover (Riihelä et al., 2017), implying that the latest years of CLARA with the highest number of concurrently serving AVHRR instruments should be the most reliable.

While the overall cloudiness decline from the mid-1990s to 2007 is still in line with the melt mechanisms proposed by Hofer et al. (2017), the post-2007 cloudiness recovery in CLARA-A2 is then inconsistent with the continuing large mass loss from the CLARA and MOD10A1 proxy SMB time series, as well as MAR, up till 2012. Alongside the comparison to the results from Tedesco et al. (2016b), this suggests that the relationship between cloudiness, atmospheric circulation patterns, and surface melt may be still more complicated than has been shown to date. Further, cloud radiative properties can also significantly impact bright snow surface radiative budgets, but these parameters (e.g. cloud optical thickness (COT) and effective particle radius, or liquid/ice water path (LWP/IWP)) are challenging to retrieve from AVHRR (Riihelä et al., 2017).

Another means of linking large-scale atmospheric circulation to the surface albedo changes is via the Greenland Blocking Index (GBI; Hanna et al., 2013, 2016). Positive anomalies in GBI represent atmospheric conditions which block northerly zonal airflow over the ice sheet, allow for warm southerly airflow particularly across the west margin, and promote clear-sky conditions over the ice sheet, enhancing surface melt (Hanna et al., 2014). The basin-scale comparisons between surface albedo anomalies and GBI anomalies (both calculated with respect to the 1982–2015 mean) are shown as time series in Fig. 11. Generally, there is a negative correlation between GBI and albedo (see Fig. S4), although with variability between basins. For example, in agreement with Tedesco et al. (2016a), the 2015 positive GBI anomaly was linked to an atmospheric ridge which enhanced melt in northern Greenland (basins 1 and 6) while preventing it in the south, a feature reproduced in the corresponding albedo anomalies. Along with the concurrent recovery in ice sheet mean cloudiness (Fig. 10), the significant covariability between GBI and albedo implies that the apparent post-2012 slowing in further albedo decreases is linked to a shift in large-scale atmospheric circulation, allowing for more cold northerly airflow (Tedesco et al., 2014) and increasing cloudiness (during summers; Fig. 10).

McLeod and Mote (2016) also connect increasing surface melt since the early 1990s with the increasing frequency, and partly duration, of very large GBI (Greenland blocking episodes, GBEs), in line with results from our observation-driven analysis. Rajewicz and Marshall (2014) report that a substantial part of summer melt variability is directly explainable by increasing anticyclonic circulation over Greenland, isolating that signal from the overall near-surface air temperature increases and contesting that any abrupt change in atmospheric regime is behind the change. They also note that the anticyclonic circulation began to impact South Greenland during the 1990s, with the effect spreading to other parts of the ice sheet after 2001. Further, they note that while the NAO index is often used to explain melt anomalies, NAO is not a causal mechanism for the melt but is simply covariant with melt-inducing ridging conditions over the ice sheet. This reasoning is consistent with the gradually decreasing albedo seen in CLARA during the pre-MODIS era on the western margin but does not explain the substantial late-summer albedo decrease in the northeast in the same era. They are also consistent with a stronger overall correlation between CLARA JJA albedo and GBI (r=-0.64) than between JJA albedo and NAO (r=0.47) using data from Hurrell and Deser (2009).

The influence of oceanic heat advection is another possible cause for some of the observed albedo decreases. Both CLARA and MOD10A1 detect localized negative albedo trends around Helheim and Kangerdlussuaq glaciers in the southeast margin. Straneo et al. (2011) documented the presence of warm subtropical waters in the Helheim Glacier fjord, Häkkinen and Rhines (2009) showed that such waters have begun to penetrate the seas around SE Greenland with increasing intensity, and Christoffersen et al. (2011) connected an increase in oceanic heat fluxes with subsequent Kangerdlussuaq Glacier retreat. It could thus be postulated that the release of heat from these warmer waters could act as a localized additional source of energy for the surface snow and ice in the area surrounding these two glaciers served by relatively large fjords. However, outlet glacier thinning accelerated by oceanic heat (Murray et al., 2010) and exposure of impurities from the ice with increasing melt (Tedesco et al., 2016b) could also produce this effect.

Finally, we note that while the CLARA proxy does provide observation-based confirmation on the dominance of surface melt in the variability of the summed MJJA surface mass balance, the results show clear physical constraints on the validity of mechanisms linking enhanced albedo reductions as a proxy for SMB decreases. The question of increasing biological activity as a driver of albedo reduction along the ice sheet margins remains open for want of long-term records to compare with the present albedo estimates from CLARA. A large-scale spread of algae colonies on the ice surface would certainly provide a mechanism for albedo decrease unaccounted for in present models and satellite-based retrieval algorithms.

We have investigated the decadal trends in the surface albedo of the Greenland Ice Sheet between 1982 and 2015 and their connection to melt rate, bare ice exposure, and surface mass balance and ice discharge rate. The main results of this study may be summarized as follows.

Over the 34-year investigation period, the late-summer surface albedo of the ice sheet margins shows a statistically significant decrease at the 95 % confidence level. The 1982–2015 decadal albedo trends reach a maximum of approximately -0.05 over the ablation region in the Kangerlussuaq sector in July, although some grid cells in the area display decadal trends twice as large during 2000–2015. The trends are consistent with MODIS data for the overlap period.

The albedo decrease along the west coast mainly occurred between 2000 and 2012, concurrently and consistently with a previously reported change in atmospheric circulation favouring the advection of warm and moist southerly air masses along the west coast. The albedo decrease of the northeastern and eastern margins was initiated during the 1982–1999 period. While the darkening there is statistically significant, we cannot at present identify a clear driver for it.

The melt seasons along the western and northern ice sheet margins intensified primarily after 2000, as seen by an acceleration of the albedo decrease rate during the melting season. Also, regions where surface melt causes the albedo to drop to values typical of bare ice were much more widespread in 2012–2015 than in the early 1980s and expand to elevations 50–100 m higher on the ice sheet.

A notable exception to the widespread albedo decrease was observed along the southern margins, consistently and concurrently with increased winter snowfall in the MERRA-2 reanalysis. The future persistence of this effect will undoubtedly depend on the future magnitude and phase of precipitation along the ice sheet's southern margin.

When calibrated against surface mass balance estimates based on GRACE gravimetric measurements and ice discharge estimates, the CLARA surface albedo estimates form a proxy for summer-aggregated SMB reaching back to 1982. This proxy shows reasonably good agreement and a high degree of covariability with MAR regional climate model (v3.5.2) estimates for the full investigation period. This confirms the dominant role of surface runoff in the ice sheet's SMB variability during the satellite era.

We find no observational evidence for a connection between annual surface albedo changes (as a proxy for runoff production anomalies) and anomalies in annual rates of ice discharge at drainage basin scales.

Our 3-decade analysis of homogenized AVHRR imagery places the ice sheet's significant albedo changes after the turn of the millennium in a broader context. Most changes are clearly attributable to concurrently prevailing patterns in atmospheric circulation and precipitation. The robustness of the first half of the AVHRR record may only be assessed indirectly for want of reliable reference data. However, we see no significant instability in time in the CLARA surface albedo record prior to 2000. During the same period, the reasonably good covariability between MAR and the CLARA-derived proxy provides further reinforcement to the plausibility of the early satellite record. Our results are consistent with a broad range of prior studies on the changing surface state of the ice sheet.