Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 14-Day Trial for You or Your Team.

Learn More →

Dark zone of the Greenland Ice Sheet controlled by distributed biologically-active impurities

Dark zone of the Greenland Ice Sheet controlled by distributed biologically-active impurities ARTICLE DOI: 10.1038/s41467-018-03353-2 OPEN Dark zone of the Greenland Ice Sheet controlled by distributed biologically-active impurities 1,2,3 1,4 5,6 1 7 Jonathan C. Ryan , Alun Hubbard , Marek Stibal , Tristram D. Irvine-Fynn , Joseph Cook , 2 8 9 Laurence C. Smith , Karen Cameron & Jason Box Albedo—a primary control on surface melt—varies considerably across the Greenland Ice Sheet yet the specific surface types that comprise its dark zone remain unquantified. Here we use UAV imagery to attribute seven distinct surface types to observed albedo along a 25 km transect dissecting the western, ablating sector of the ice sheet. Our results demonstrate that distributed surface impurities—an admixture of dust, black carbon and pigmented algae— explain 73% of the observed spatial variability in albedo and are responsible for the dark zone itself. Crevassing and supraglacial water also drive albedo reduction but due to their limited extent, explain just 12 and 15% of the observed variability respectively. Cryoconite, con- centrated in large holes or fluvial deposits, is the darkest surface type but accounts for <1% of the area and has minimal impact. We propose that the ongoing emergence and dispersal of distributed impurities, amplified by enhanced ablation and biological activity, will drive future expansion of Greenland's dark zone. 1 2 Centre for Glaciology, Department of Geography and Earth Sciences, Aberystwyth University, Aberystwyth SY23 3DB, UK. Department of Geography, University of California, Los Angeles, Los Angeles, CA 90095, USA. Institute at Brown for Environment and Society, Brown University, Providence, RI 4 5 02906, USA. Centre for Arctic Gas Hydrate, Environment and Climate, Department of Geology, University of Tromsø, 9037 Tromsø, Norway. Department of Ecology, Faculty of Science, Charles University, 12844 Prague, Czech Republic. Department of Geochemistry, Geological Survey of Denmark and 7 8 Greenland, 1350 Copenhagen, Denmark. Department of Geography, University of Sheffield, Sheffield S10 2TN, UK. Institute of Biological, Environmental and Rural Sciences, Aberystwyth University, Aberystwyth SY23 3DB, UK. Department of Glaciology and Climate, Geological Survey of Denmark and Greenland, 1350 Copenhagen, Denmark. Correspondence and requests for materials should be addressed to A.H. (email: abh@aber.ac.uk) NATURE COMMUNICATIONS (2018) 9:1065 DOI: 10.1038/s41467-018-03353-2 www.nature.com/naturecommunications 1 | | | 1234567890():,; ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/s41467-018-03353-2 he Greenland Ice Sheet has become the largest cryospheric the specific ice surfaces across the dark zone and determine their contributor to global sea-level rise predominantly through impact on the mesoscale (1–10 km) albedo distribution during Tincreased surface melt and runoff, which accounts for over peak melt season, as represented by the MODIS albedo product, 1–4 half of its mass loss since 1991 . The dominant energy source MOD10A1. On 8 August 2014, a fixed-wing UAV equipped with for snow and ice melt is direct solar shortwave radiation, the a digital camera and upward and downward facing pyranometers absorption and reflection of which is predominantly modulated was deployed from a field camp based in the vicinity of the K- 5–7 by surface albedo . Accurately constraining spatiotemporal transect, S6 automated weather station (AWS) on a 25 km east- patterns of albedo across the ice sheet is hence fundamental to west transect dissecting the dark zone (Fig. 1). Seven distinct understanding and predicting surface melt and runoff along with surface types were visually identified on the ground by an expert their impact on ice sheet flow dynamics and sea-level rise. A and automatically classified based on their reflectance and conspicuous feature of Greenland’s ablation area is its dark zone, roughness properties. The survey transect was divided into sixty an area of bare ice with particularly low albedo that appears 500 × 500 m segments, co-located to the footprints of corre- across the west and southwest sectors of the ice sheet each sponding MODIS pixels, and the fractional area of each surface 8–10 summer . At the Arctic Circle, in the vicinity of the Kanger- type in each segment was determined using a supervised k- lussuaq (K-) sector, the dark zone extends between 20 and 75 km nearest neighbours (k-NN) classification (see Methods section: from the land-terminating margin where Moderate Resolution Surface Classification for more information) (Figs. 1, 2). Finally, Imaging Spectroradiometer (MODIS) data indicate a regional the mean albedo of each surface type was derived from the digital albedo minimum of ~0.34 . From 2000 to 2012, the spatial extent imagery and the relative contribution of different surface types to of the dark zone increased by 12% but also exhibited considerable mesoscale albedo variability (defined by MOD10A1 pixels) was 11,12 interannual variability . The extent of the dark zone is weakly calculated using principal component regression (PCR). positively correlated with air temperature and negatively corre- 11,12 lated with solar radiation during June, July and August (JJA) . Results This suggests that the ongoing albedo decrease observed during Surface type variation along the transect. Analysis of all UAV the melt season is not simply driven by melting of the winter imagery allowed us to visually identify and automatically classify snowpack to reveal the darker bare ice surface beneath, but, fol- seven distinct surface types across the survey transect: (i) clean lowing exposure, there are changes in the nature of the bare ice ice, (ii) ice containing uniformly distributed impurities, (iii) deep 11,12 itself . However, the surface characteristics of the dark zone water, (iv) shallow water, (v) cryoconite either in holes or fluvial remain unquantified because the spatial resolution of satellite deposits, (vi) crevasses and (vii) snow (Fig. 3). Distinction imagery is insufficient to fully resolve the specific surface types between clean ice and ice containing uniformly distributed that comprise it, and how these surfaces evolve through time, impurities was guided by qualitative assessment of 112 oblique distinguishing the dark zone from brighter ice surfaces adjacent and nadir photographs taken from the ground (<1 cm pixel to it. footprint) at specific study sites around the field camp (Fig. 1). Previous field-based, in situ observations indicate that western These images confirm that distributed impurities across the ice Greenland’s ablation zone is characterized by highly variable non- surface are responsible for bare ice albedo variability at the local 8,13–15 ice constituents and surface structures . These include fea- scale (1–10 m) (Fig. 4). In order to upscale and understand the 16,17 tures such as crevasses, fractures and foliations ; supraglacial impact of these surface impurities on the mesoscale albedo dis- hydrological features, including streams, rivers, ponds and tribution of the ablation zone, we divided bare ice into two 18,19 lakes ; snow patches and fracture cornices; cryoconite, con- categories: clean ice, with very low impurity concentrations, and 20,21 centrated in holes or in supraglacial fluvial deposits ; microbes ice containing some or an abundance of impurities. It is apparent 22–24 and their humic by-products ; mineral dust and aerosols that additional categories could be defined for bare ice given from outcropping or contemporary aeolian deposition including sufficiently high pixel resolution, but for the purpose of this study, 10,25,26 black carbon from wildfires and other aerosols. While the and considering the spectral limitations of the onboard camera, highest resolution optical satellite imagery currently available has we do not attempt to. We note that it would be a fruitful direction facilitated the examination of crevasse fields and surficial with multi- and hyper-spectral sensor payloads. Clean ice has hydrology , a quantitative assessment of the specific ice surface 57.2% aerial coverage in the lower, western half of the survey types that comprise the dark zone, and how they combine to yield transect between 0 and 17 km (Fig. 3). In the eastern half (17–27 observed albedo patterns across the ablation zone of the ice sheet, km), clean ice coverage is lower at 23.0%. Ice containing uni- has yet to be made. formly distributed impurities (Fig. 5a) varies inversely to clean Here, we utilize high-resolution (15 cm pixel size) imagery ice, with a higher fraction in the eastern half (74.5%) compared to acquired from an unmanned aerial vehicle (UAV) to characterize the western half of the transect (40.0%) (Fig. 3). Dark zone 50.0°W 49.5°W 49.0°W 67.1°N 5c Fig.5a 5b Russell Glacier 30 km 6c 6b Fig.6a 15 km 0 km 0 5 km UAV footprint Camp Fig. 1 Overview map showing location of UAV survey transect. The background is a Landsat 8 Operational Land Imager (OLI) true colour image of the Kangerlussuaq sector of the Greenland Ice Sheet from 6 August 2014. The transect was divided into sixty 0.25 km segments for comparison with the MODIS albedo product, MOD10A1. High-resolution aerial imagery and surface classification of six segments (coloured red) are shown in Figs. 5 and 6 2 NATURE COMMUNICATIONS (2018) 9:1065 DOI: 10.1038/s41467-018-03353-2 www.nature.com/naturecommunications | | | NATURE COMMUNICATIONS | DOI: 10.1038/s41467-018-03353-2 ARTICLE MOD10A1 pixel 500 m Fig. 2 Schematic summarizing the aims of the study. Aerial digital imagery are used to characterize the surface types that are found in the ablation zone and assess their impact on mesoscale spatial albedo patterns as represented by MODIS 6a 6b 6c 5a 5b 5c 0.50 0.45 MOD10A1 0.40 albedo 0.35 (unitless) 0.30 0.25 0.60 Clean ice fraction 0.40 (unitless) 0.20 0.00 Distributed 0.85 impurities 0.65 fraction 0.45 (unitless) 0.25 Surface 0.30 water 0.20 fraction 0.10 (unitless) 0.00 0.06 Crevasse 0.04 fraction 0.03 (unitless) 0.01 0.00 0.02 Cryoconite 0.01 fraction 0.01 (unitless) 0.00 0.00 0.02 Snow 0.01 fraction 0.01 (unitless) 0.00 0.00 510 15 20 25 30 W E Distance from most westerly point of transect (km) Fig. 3 Variation of albedo and the fractional area of each surface type across the UAV transect. The albedo and fractional areas derived from MOD10A1 and the UAV imagery, respectively, on 8 August 2014. The x axis is displayed in Fig. 1. The results of the classification for six segments, highlighted by the vertical grey bars, are shown in Figs. 5 and 6 Cryoconite is commonly found in holes but also in fluvial concentrated, rather than distributed, impurities. Cryoconite deposits near supraglacial streams and lakes (Fig. 6c). In this has a maximum aerial coverage of 1.6% at 8 km from the western study, cryoconite is distinguished from ice containing distributed end of the transect and a mean coverage of 0.6% across the entire impurities by its very low albedo, which is indicative of transect (Fig. 3). It is possible that we underestimate the fractional NATURE COMMUNICATIONS (2018) 9:1065 DOI: 10.1038/s41467-018-03353-2 www.nature.com/naturecommunications 3 | | | 500 m ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/s41467-018-03353-2 a b Fig. 4 Photograph showing close-up of bare ice found in the ablation zone. The photographs were taken near the field camp located close to the S6 automated weather station at ~1000 m a.s.l. (Fig. 1). a Ice containing distributed impurities on the surface and b clean ice with cryoconite holes a Original Albedo Classified MOD10A1 Albedo = 0.29 Ice containing Fraction distributed 0.037 impurities 0.951 0.011 200 m 0.001 b MOD10A1 Albedo = 0.27 Fraction 0.014 0.933 0.046 0.007 MOD10A1 Albedo = 0.36 Shallow surface Deep water Fraction surface 0.249 water Clean ice 0.421 0.141 0.187 0 Albedo 1 0.002 Clean ice Distributed impurities Deep surface water Shallow surface water Cryoconite Fig. 5 RGB digital image, albedo map and classification of surface types in three MOD10A1 pixels. The albedo maps were derived from the digital images (Methods). The locations of the segments along the UAV transect are shown in Fig. 1. a Segment characterized by mostly ice containing uniformly distributed impurities. b Segment characterized by similar ice surface to a but with a larger fraction of channelized surface melt-water. c Segment dominated by a supraglacial lake with a previous shore consisting of clean ice 20,21 area of small cryoconite holes due to the limited, 15 cm pixel surrounding them . Coincident field measurements, made resolution of our UAV imagery. However, we note that smaller during UAV image acquisition, indicate that the cryoconite holes cryoconite holes (<15 cm) would also be hidden from virtually all observed in our study were well developed. The implication is aerial and satellite imagery obtained at low solar elevation angles. that once they have attained equilibrium depth, they cease to Furthermore, cryoconite hole depths tend to equilibrate as the absorb additional energy (which would make them deeper) melt season progresses, due to their low albedo and preferential compared to surrounding ice and hence are effectively neutralized radiative absorption in comparison to brighter ice surfaces from the effects of incoming solar radiation. For these reasons, we 4 NATURE COMMUNICATIONS (2018) 9:1065 DOI: 10.1038/s41467-018-03353-2 www.nature.com/naturecommunications | | | NATURE COMMUNICATIONS | DOI: 10.1038/s41467-018-03353-2 ARTICLE a Original Albedo Classified MOD10A1 Albedo = 0.38 Crevasse Fraction 0.558 0.381 0.002 0 Albedo 1 200 m 0.059 b MOD10A1 Albedo = 0.47 Clean ice Fraction 0.710 0.279 0.003 0.008 MOD10A1 Albedo = 0.37 Fraction 0.682 0.293 Cryoconite 0.016 0.009 Clean ice Distributed impurities Cryoconite Shallow surface water Crevasse Fig. 6 RGB digital image, albedo map and classification of surface types in three more MOD10A1 pixels. The albedo maps were derived from the digital images (Methods). The locations of the segments along the UAV transect are shown in Fig. 1. a Segment containing a high fraction of crevasses. b Segment characterized by a much lower relief surface and no crevasses. c Segment characterized by clean ice and numerous cryoconite holes argue that smaller cryoconite holes had a minimal net impact on area of ice containing uniformly distributed impurities explains MODIS-derived albedo compared to the ice surface surrounding 73% of the observed mesoscale albedo variability. Although not them. the darkest surface type observed, distributed impurities dom- At 23 km (all distances refer from the western start point of the inate the mesoscale albedo signal due to their extensive coverage UAV transect), a braided meltwater channel network with a and large variations in their fractional area across the survey fractional area of 5.3%, intersects the transect (Figs. 1 and 5b), but transect. Distributed impurities have been attributed to the out- otherwise, surface water comprises only 1.9% of the survey area cropping of aeolian dust deposited during the early Holo- 19 9,10,27 (Fig. 3). These results are consistent with Smith et al. who found cene and/or pigmented surface algal blooms and associated 2 22–24 that surface water accounted for 1.4% of a ~5000 km bare ice humic material . area in the K-sector of the ice sheet. At 28 km, a small, 0.83 km Locally, defined at the scale of a single MODIS pixel, supraglacial lake covers 32.8% of the segment (Fig. 5c). Crevasse supraglacial water, contained in both lakes and channels, has a density is highest on the western flank between 2 and 5 km and distinct impact on albedo due to its low albedo (α = 0.19–0.26). attains maximum coverage of 5.9% at 2 km (Figs. 3, 6a). Up- The segment of the transect that corresponds to a large braided glacier of 10 km, crevasses are almost entirely absent (Fig. 6b). channel network at 23 km has an albedo of 0.28 (Fig. 5b). This is Remnant snow patches, which persist within ice fractures and ~0.02 lower than the surrounding segments with <1% surface supraglacial channel incisions, attain a maximum coverage of water yet are otherwise composed of similar ice surfaces (Figs. 3, 1.7% with a mean aerial coverage of only 0.1%, at this time of 5a). However, in comparison to distributed impurities, supragla- year. cial water has a minor impact on the mesoscale albedo pattern and explains only 15% of the albedo variability across our survey transect. Surprisingly, the supraglacial lake located at 28 km is not associated with a significant reduction of MOD10A1 albedo. This Relationship between surface types and albedo patterns. The is because it is relatively narrow (260 m width) and covers just mesoscale albedo distribution, determined from MOD10A1 data, 33% of the segment, while 25% of the remaining segment consists exhibits considerable variability along the survey transect with of very bright clean ice faculae (mean α = 0.58): interpreted as a values between 0.27 and 0.47 (Figs. 1, 3). The dark zone, which is shoreline exposed when the lake level dropped (Fig. 5c). Hence, approximately located between 18 and 27 km along the survey the low albedo of the lake water (mean α = 0.19) itself is offset by transect, has a mean albedo of 0.29 and is characterized by dis- the brightness of the surrounding ice surface adjacent to it tinct and conspicuous banding that specifically relate to foliation yielding minimal change in the net albedo (MOD10A1) signature. structures apparent in Landsat 8 imagery (Fig. 1). Between 80 and Crevassing explains 12% of albedo variability and its impact is 95% of the dark zone is classified as ice containing uniformly distributed impurities (mean albedo (α) = 0.27), with the well illustrated at 3–4 km where a transition into a crevasse zone yields a significant reduction in mesoscale albedo compared to the remaining 5–20% consisting of predominately clean ice (mean α = 0.55) (Fig. 5a). Application of PCR reveals that the fractional adjacent, flatter surface (Fig. 6a, b). Crevasses enhance shortwave NATURE COMMUNICATIONS (2018) 9:1065 DOI: 10.1038/s41467-018-03353-2 www.nature.com/naturecommunications 5 | | | ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/s41467-018-03353-2 radiation absorption; radiative transfer modelling indicates that summer ablation alone . Instead, positive correlations between the presence of crevasses can double the downward energy the dark zone extent and proxies for the availability of liquid absorbed relative to a homogeneous, flat ice surface and reduce water and nutrients are interpreted as evidence that blooms of 28,29 albedo by between 0.10 and 0.25 . The amount of radiation surface ice algae control bare ice albedo across the dark zone. Any absorbed by crevasses is determined by their size, orientation, increase in temperature and/or liquid water production in the density and whether they are water-filled. One of the most presence of dust promotes further colonization of surface algae densely crevassed areas across the transect (5.9% fractional area) yielding an increase in pigmented biomass and net albedo 24,36 (Fig. 6a), with crevasse widths up to ~10 m and depths in excess reduction . The 12% expansion of the dark zone between 2000 of 8 m, yields an albedo reduction of ~0.06 in comparison to the and 2014 in western Greenland, corresponding with an increase segment in Fig. 6b, which has no crevassing but similar fractions in mean summer air temperature of 0.13 °C per year over the of other surface types (Figs. 3, 6b). Elsewhere, at 6–7 km, smaller same period , provides further support for these ongoing crevasses, with mean widths of 5 m, have a reduced impact on processes. mesoscale albedo, lowering it by only 0.02 in comparison to the While our results attest that the variation in the fractional area control segment in Fig. 6b. This observation is at odds with of supraglacial lakes, streams, crevasses and cryoconite do not radiative transfer modelling results because the modelled significantly affect mesoscale albedo, they still play a secondary crevasses were compared against a flat, clean ice surface, which role in determining interannual and seasonal albedo variability. 28 29 is not the case here (Fig. 6b) . Cathles et al. modelled crevasses For example, supraglacial water may act to consolidate or dis- with width to depth ratios similar to the crevasses we observed tribute sediment and impurities across the ice sheet surface . and found that they have a melt enhancement factor of 1.14–1.20 Moreover, a relatively small expansion in the spatial extent of at solar zenith angles of 45°, which is in broad agreement with our surface water would have a disproportionate impact on mesoscale findings. albedo and further amplify ablation. Melt rates at the base of Increases in the fractional area of cryoconite, either in large supraglacial lakes and water bodies are double that of bare ice 38,39 holes or fluvial deposits, are not particularly associated with surfaces due to enhanced shortwave radiation absorption . mesoscale albedo reduction across the survey transect, and Atmospheric warming has been shown to increase the spatial 40,41 surprisingly, the lowest concentration of cryoconite is actually extent and duration of ponded supraglacial water . During observed within the dark zone itself (Fig. 5a). Cryoconite only years with higher summer temperatures, such as 2007, 2010 and occupies a very small fraction of the total coverage (1.6% 2012, supraglacial lakes formed earlier in the season and occupied maximum and 0.6% mean), which can be explained by the nature a 40% larger area than in cooler summers . It follows that of the cryoconite material itself. The thread-like, filamentous increased storage of water in supraglacial lakes will play an structure of cyanobacteria enables them to entangle debris and important role in net albedo reduction across an expanding bare facilitate the formation of granules. These granules absorb more ice area in future. solar radiation and melt down into the ice until they are in Our analysis also demonstrates that crevasses reduce local 30–32 radiative and thermodynamic equilibrium . Although this (0.1–1 km) albedo, and hence any increase in crevasse extent will mechanism means that the cryoconite hole has a very low albedo impact on mesoscale albedo patterns. Crevasses form due to value (mean α = 0.10) when observed from directly above, the localized concentration of tensile stresses which, due to highly hole occupies a relatively small area and is effectively hidden at variable subglacial conditions and longitudinal stress coupling, non-zenith solar illumination resulting in an increase in are spatially and temporally variable across the Greenland Ice 20 16,42,43 mesoscale albedo . Furthermore, cryoconite holes are often . In response to increased surface melt, GPS obser- Sheet covered by an ice lid, caused by the refreezing of water that has vations by van de Wal et al. report reduced net flow over the 31 45 filled the hole during negative net radiation conditions . While marginal zone of the K-transect, whereas Doyle et al. report thin frozen lids may undergo partial or complete ablation during persistent ice flow acceleration above the equilibrium line up to the day, their higher albedo acts to further moderate the impact of 140 km from the ice sheet margin in this same sector. Recent 46 46,47 cryoconite that they cover and render the holes indistinguishable modelling and observations of new crevasses forming over from the adjacent ice surface (much to the dismay of many a field 160 km from the western margin reveals that the ice sheet interior campaigner with sodden feet). is also more dynamically sensitive to transient stress perturba- tions originating from downstream than a previous steady-state model suggests . Regardless of ice dynamics, inland migration of Discussion the equilibrium line caused by atmospheric warming will drive The analysis presented here demonstrates that the dark zone has increased bare ice extent, further exposing existing crevasses that low fractional areas of surface water (<1.0%), cryoconite holes were formerly snow and firn covered. Hence, surface crevasse (<0.5%) and crevasses (<0.2%). Instead, it appears that ice con- extent will likely expand in future, resulting in mesoscale albedo taining uniformly distributed impurities draped over a relatively reduction and enhanced surface absorption of incoming energy flat surface are the primary agent responsible for the low available for melt. (MOD10A1) albedo values observed during the melt season Finally, future spatial expansion of cryoconite does have the (Fig. 5a). Near the S6 AWS, from where the UAV was launched, potential to significantly impact surface albedo. Hodson et al. Stibal et al. report that samples of distributed impurities consist showed that 53% of plot-scale (0.01–0.5 m) variation in albedo of an abundance of ice algae (Fig. 4a), which are characterized by was correlated with the growth of cryoconite holes, and Chandler a grey/brown hue due to the brown-to-purple coloured pigments et al. report that the gradual seasonal reduction in albedo also 24,33,34 surrounding the algae chloroplasts . Correlations between correlates with an increase in cryoconite hole size and number. dust content and the abundance of microbes suggest that the An increase in the extent of cryoconite holes may be caused by melt-out of particulates may provide nutrients for surface ice longer and warmer ablation seasons, which would increase the 22,35 12 algae to grow and indirectly control the extent of dark zone . heat energy to the walls and base of the hole, leading to further Further support for this hypothesis is provided by Tedstone melting and hole expansion . On the other hand, an increase of et al. who argue that the large interannual variability in the meltwater may promote aggregation of distributed impurities and extent of the dark zone, and its significant reduction in 2013 and could have a surface cleaning effect, potentially raising the albedo 20,31 2015, demonstrates that bare ice albedo is not a consequence of of the surrounding bare ice . The growth and development of 6 NATURE COMMUNICATIONS (2018) 9:1065 DOI: 10.1038/s41467-018-03353-2 www.nature.com/naturecommunications | | | NATURE COMMUNICATIONS | DOI: 10.1038/s41467-018-03353-2 ARTICLE cryoconite holes on mesoscale albedo is hence complex and still surface particulates and nutrients fertilizes pigmented ice surface somewhat ambiguous. algae, which drives albedo reduction over the duration of the melt 12,22 In this study, we characterized the spatial variability of surface season . Further research though is required to determine how types across the western ablating margin of the ice sheet towards these factors combine to increase the spatial extent and con- the end of the melt season when bare ice surfaces are most centration of pigmented surface algae, and their interaction with apparent. However, for much of the year (September to May), the the availability of in situ and aeolian-derived nutrients, changing ice sheet is snow-covered and it is likely that snow grain size and atmospheric forcing and enhanced ice melt and runoff. impurity concentration govern mesoscale albedo patterns across the K-transect and elsewhere during this period. Likewise, early Methods melt-season albedo patterns are primarily governed by the rela- MODIS albedo. Albedo patterns were determined from a MOD10A1 C6 broad- tive proportions of snow and ice extent. Accurately determining band (spectral range of 300–3000 nm) albedo product from the 8 August 2014 and available from the National Snow and Ice Data Center (NSIDC) . MOD10A1 is snow melt and the timing of bare ice exposure has therefore been gridded in a sinusoidal map projection and has a resolution of ~500 × 500 m or a priority for surface mass and energy balance models and the 0.25 km . The value of each pixel represents the best single albedo observation in theoretical determinants of snow albedo and melt are relatively 59 the day based on cloud cover and viewing and illumination angles . We estimated 5,50 well established . In contrast, few studies have investigated the that MOD10A1 has a root mean square difference (RMSD) of 7.0% in comparison to albedo measured by CNR1 or CNR4 thermopile pyranometers at the PRO- albedo of the bare ice surface types that characterize the ablation MICE/GAP automatic weather stations KAN-L, KAN-M and KAN-U between zone and they have commonly been treated as temporal and 60 59 2009 to 2014 . This compares well to Stroeve et al. who estimated an RMSD of 51–53 spatial constants in surface melt models . 6.7%. The observed spatiotemporal variability in albedo across the 54,55 ablation zone has motivated a new generation of surface UAV platform. Aerial imagery was acquired by a fixed-wing UAV identical to that energy balance models that assimilate spatial patterns of albedo 14,61,62 used by Ryan et al. The UAV has a 2.1 m wingspan and is powered by a 10 4,56,57 derived from MODIS data . Across our 25 km survey Ah, 16.8 V LiPo battery pack which, with a total weight of 4 kg, yields a 1 h endurance and 60 km range. The autonomous control system is based around an transect, the MODIS-derived surface albedo pattern is dominated Arduino navigation and flight computer updated in real-time by a 10 Hz data by variations in the extent of uniformly distributed impurities, a stream comprising of a GPS, magnetometer, barometer and accelerometer. These result that contradicts previous research attributing it to an data are logged along with a timestamp for each activation of the digital camera increased occurrence of supraglacial water . The source, pro- shutter which automatically triggers when a horizontal displacement threshold is cesses and drivers of distributed impurities are yet to be exceeded. The UAV was hand launched on 8 August 2014 from a base camp at 67.08°N, 49.40°W, located at the site of the Institute for Marine and Atmospheric unequivocally established, with some studies indicating a wind- Research (IMAU), University of Utrecht S6 automatic weather station. It was pre- blown origin, others revealing that they are derived from melt-out programmed to carry out a 25 km survey across the Kangerlussuaq sector of the 9–11 of englacial dust . Recent research has promoted the concept western Greenland Ice Sheet (Fig. 1). The Greenland Ice Mapping Project (GIMP) of bioalbedo, which argues that the melt-out and release of digital elevation model (DEM) was used during the selection of three- 250 250 Snow 200 200 Shallow water Clean ice Deep water 150 150 100 100 Distributed impurities 50 50 Cryoconite 0 0 0 50 100 150 200 250 0 50 100 150 200 250 Green DN Red DN 250 250 200 200 150 150 Crevasse 100 100 50 50 0 0 0 50 100 150 200 250 0 1 Green DN Crevasse Fig. 7 Scatter-plots showing the key attributes of the seven surface types identified in this study. DN is digital number of the Sony NEX-5N calibrated RAW image. Deep and shallow supraglacial water is easily distinguishable because it has low reflectivity in the red band and forms a unique cluster in the feature space. Snow, clean ice, ice containing distributed impurities and cryoconite form another cluster and are distinguishable because they reflect different proportions of RGB visible wavelengths NATURE COMMUNICATIONS (2018) 9:1065 DOI: 10.1038/s41467-018-03353-2 www.nature.com/naturecommunications 7 | | | Red DN Blue DN Green DN Blue DN ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/s41467-018-03353-2 dimensional waypoints to ensure the UAV maintained a constant altitude of 350 m vignette of all images acquired at nadir during the survey period. Barrel distortion above the surface during the autonomous sorties. On return from the sortie, the was corrected using ImageMagick (http://www.imagemagick.org/), which utilized UAV was manually landed into a 10 × 5 m net. the coefficients stored in the image’s ancillary metadata also known as exchange- able image file format data. We then corrected the images for changing illumination conditions during the Digital imagery. Digital imagery was acquired by a Sony NEX-5N digital camera survey using downward irradiance measured by a ground-based upward facing vertically mounted inside the front of the airframe. The camera has a 16 mm fixed Apogee SP-110 pyranometer. To do this, images of a 25 × 25 cm Teflon white focus lens (53.1 by 73.7° field of view) yielding an image footprint of ~525 × 350 m reference target were acquired every 10 min using the UAV digital camera from the during the autonomous sortie. The width of each image is approximately similar to ground. The relationship between the mean RGB DNs of the white reference target the pixel footprint of MODIS. The camera was preset with a fixed shutter speed of and the downward irradiance recorded by the upward facing pyranometer were 1/1000 s, ISO 100 and F-stop of 8, and triggered every 35 m to provide a 90% used to construct a calibration curve using a linear least squares regression (R = forward image overlap. The relatively fast shutter speed minimizes image blur 0.96). The ratio of reflected radiation recorded by the camera and the downward while the low ISO and F-number ensures maximum image quality where even the radiation estimated from the calibration curve enabled the illumination-corrected brightest surfaces do not saturate the image. The camera was set to record the reflectance of each pixel to be defined. images in RAW format, an image format that contains minimally processed data Since snow and ice are non-Lambertian surfaces, a nadir measurement of from the camera’s sensor. During the survey, ~2000 RAW images were acquired at reflectance underestimates albedo by between 1 and 5% in the visible band . The the camera’s maximum (4912 × 3264 pixels) resolution which, once the images illumination-corrected images were therefore calibrated again by multiplying the were corrected for barrel, or geometric, distortion, equates to a ground sampling image pixel numbers by a factor calculated by dividing the mean pixel value of the distance of ~11 cm. illumination-corrected image by the albedo recorded by upward and downward facing Apogee SP-110 pyranometers mounted on the UAV. Ryan et al. found that albedo determined using this method has an accuracy of ±5% over ice sheet Orthomosaic and DEM generation. The R(ed)/G(reen)/B(lue) images were used surfaces typically found in the ablation zone. to produce an orthomosaic and DEM using Agisoft PhotoScan Pro (http://www. agisoft.com/) following the processing sequence described by Ryan et al. The images were georeferenced by providing latitude, longitude and altitude data Data availability. The MODIS (MOD10A1) albedo data are available from the recorded by the flight controller. The orthomosaic was produced in the software’s National Snow and Ice Data Center (NSIDC) at http://nsidc.org/data/MOD10A1. ‘mosaic’ mode, meaning that pixels in the centre of the images were preferentially The UAV images are archived in the PANGAEA repository: https://doi.pangaea. used to provide the output pixel value. The orthomosaic and DEM were nearest de/10.1594/PANGAEA.885798. neighbour resampled to a ground resolution of 15 cm and 50 cm, respectively. We divided the orthomosaic into sixty 0.25 km segments and each segment was assigned a MOD10A1 value. Received: 21 June 2017 Accepted: 7 February 2018 Surface classification. The fractional area of each surface type was calculated by dividing the number of pixels of each surface type by the total number of pixels in each orthomosaic segment. The number of pixels of each surface type was esti- mated using a supervised k-NN classification from the scikit-learn Python mod- References ule . The pixels were classified using a majority vote based on the Euclidean distance to five equally weighted nearest neighbours (Fig. 7). The k-NN was 1. Box, J. E. et al. Changes to Arctic land ice. In Snow, Water, Ice and Permafrost in the Arctic (SWIPA) 137–168 (Arctic Monitoring and Assessment manually trained with seven distinct and visually identified surfaces found in the orthomosaic: (i) clean ice, (ii) ice containing uniformly distributed impurities, Programme (AMAP), Oslo, Norway, 2017). (iii) deep water, (iv) shallow water, (v) cryoconite either in holes or fluvial deposits, 2. Hanna, E. et al. Ice-sheet mass balance and climate change. Nature 498,51–59 (vi) crevasses and (vii) snow. The training samples of the surface types were (2013). manually digitized from 10 orthomosaic segments based on RGB brightness and a 3. Enderlin, E. M. et al. An improved mass budget for the Greenland ice sheet. layer that specified whether or not the pixel was situated within a crevasse or Geophys. Res. Lett. 41, 866–872 (2014). fracture. This roughness layer was determined by calculating the residual between 4. van den Broeke, M. R. et al. On the recent contribution of the Greenland ice the original 50 cm DEM subtracted from a 30 m Gaussian-smoothed DEM. sheet to sea level change. Cryosphere 10, 1933–1946 (2016). Negative anomalies with a vertical displacement >1 m were identified as crevasses. 5. Gardner, A. S. & Sharp, M. J. A review of snow and ice albedo and the Small cracks and fractures were detected on the basis of sharp RGB contrast, and development of a new physically based broadband albedo parameterization. J. were discriminated using an edge detector algorithm . Pixels within 2 m of a linear Geophys. Res. Earth Surf. 115,1–15 (2010). feature were also identified as crevasses. 6. van den Broeke, M. R. et al. Partitioning of melt energy and meltwater fluxes The efficacy of the k-NN classifier was evaluated by comparison with in the ablation zone of the west Greenland ice sheet. Cryosphere 2, 179–189 independently digitized surface types in three orthomosaic segments, at the centre (2008). and extreme ends of the transect. We found that 92% of the pixels were classified 7. Mikkelsen, A. B. et al. Extraordinary runoff from the Greenland ice sheet in accurately. The k-NN classified crevasses with an accuracy of 88% and performed 2012 amplified by hypsometry and depleted firn retention. Cryosphere 10, better for deep and shallow water (96%) than for cryoconite, ice containing 1147–1159 (2016). distributed impurities, clean ice and snow (90%). The classification of supraglacial 8. Knap, W. H. & Oerlemans, J. The surface albedo of the Greenland ice sheet: water is relatively accurate because water has low reflectance in the red band and satellite-derived and in situ measurements in the Søndre Strømfjord area forms a unique cluster in the feature space (Fig. 7). The performance of the k-NN during the 1991 melt season. J. Glaciol. 42, 364–374 (1996). for clean ice and ice containing distributed impurities is less accurate (90%) 9. Wientjes, I. G. M. & Oerlemans, J. An explanation for the dark region in the because these surfaces reflect the RGB visible bands in similar relative proportions western melt zone of the Greenland ice sheet. Cryosphere 4, 261–268 (2010). and the surface classes overlap in the feature space (Fig. 7). Misclassification could 10. Wientjes, I. et al. Carbonaceous particles reveal that Late Holocene dust causes also be caused by shadows, especially in segments with steep topography (eg, the dark region in the western ablation zone of the Greenland ice sheet. J. Fig. 6b). Shadows increase the proportions of the surface classified as ice containing Glaciol. 58, 787–794 (2012). distributed impurities and/or cryoconite. Accounting for shadows would reduce 11. Shimada, R., Takeuchi, N. & Aoki, T. Inter-annual and geographical variations the proportion of ice containing distributed impurities in the crevassed zone in the extent of bare ice and dark ice on the Greenland Ice Sheet derived from (Fig. 6b) and subsequently increase their variation across the transect. This would MODIS satellite images. Front. Earth Sci. 4,1–10 (2016). make the impact of distributed impurities on the mesoscale albedo variability, and 12. Tedstone, A. J. et al. Dark ice dynamics of the south-west Greenland Ice Sheet. the conclusions of this study, more significant. Finally, we used PCR to explore the Cryosphere 11, 2491–2506 (2017). dominant modes of surface type variation along the transect and assess which 13. Moustafa, S. E. et al. Multi-modal albedo distributions in the ablation area of surface type, or set of surface types, best correlate with mesoscale albedo variability, the southwestern Greenland Ice Sheet. Cryosphere 9,1–11 (2015). as represented by MOD10A1. Linear regression was used to yield the correlation 14. Ryan, J. C. et al. Attribution of Greenland’s ablating ice surfaces on ice coefficients between the principle component scores and MOD10A1 albedo. sheet albedo using unmanned aerial systems. The Cryosphere Discuss.1–23. https://doi.org/10.5194/tc-2016-204 (2016). Estimating albedo of digital image pixels. An estimate for the albedo (α) of each 15. Ryan, J. C. et al. Derivation of high spatial resolution albedo from UAV digital surface type was obtained from the UAV digital imagery. An explicit description of imagery: application over the Greenland Ice Sheet. Front. Earth Sci. this method can be found in Ryan et al. , but we briefly summarize it here. Firstly, 5, 40 (2017). the RAW digital numbers of each proprietary Sony RAW image were preserved by 16. Colgan, W. et al. An increase in crevasse extent, West Greenland: Hydrologic converting to a 16-bit TIFF image using dcraw (http://cybercom.net/~dcoffin/ implications. Geophys. Res. Lett. 38, L18502 (2011). dcraw/). A vignette correction mask was universally applied to compensate for 17. Colgan, W. et al. Glacier crevasses: observations, models and mass balance image and lens distortion due to edge effects which were as high as 17.6% at the implications. Rev. Geophys. 54, 119–161 (2016). corners of some images. The correction mask was calculated from the mean 8 NATURE COMMUNICATIONS (2018) 9:1065 DOI: 10.1038/s41467-018-03353-2 www.nature.com/naturecommunications | | | NATURE COMMUNICATIONS | DOI: 10.1038/s41467-018-03353-2 ARTICLE 18. Greuell, W. Melt-water accumulation on the surface of the Greenland Ice 47. Scheick, J. Investigating Greenland Crevasses at Raven Camp. Field Notes, Sheet: effect on albedo and mass balance. Geogr. Ann. A Phys. Geogr. 82, Polar Field Services Newsletter, http://www.polarfield.com/blog/blog/ 489–498 (2000). investigating-greenland-crevasses-raven-camp. Accessed 25 May 2014 (2012). 19. Smith, L. C. et al. Efficient meltwater drainage through supraglacial streams 48. Poinar, K. et al. Limits to future expansion of surface-melt-enhanced ice flow and rivers on the southwest Greenland ice sheet. Proc. Natl Acad. Sci. USA into the interior of western Greenland. Geophys. Res. Lett. 42, 1800–1807 112, 1001–1006 (2015). (2015). 20. Bøggild, C. E., Brandt, R. E., Brown, K. J. & Warren, S. G. The ablation zone in 49. Hodson, A. et al. A glacier respires: quantifying the distribution and northeast Greenland: ice types, albedos and impurities. J. Glaciol. 56, 101–113 respiration CO flux of cryoconite across an entire Arctic supraglacial (2010). ecosystem. J. Geophys. Res. 112, G04S36 (2007). 21. Chandler, D. M., Alcock, J. D., Wadham, J. L., Mackie, S. L. & Telling, J. 50. Wiscombe, W. J. & Warren, S. G. A model for the spectral albedo of snow. I: Seasonal changes of ice surface characteristics and productivity in the ablation Pure snow. J. Atmos. Sci. 37, 2712–2733 (1980). zone of the Greenland Ice Sheet. Cryosphere 9, 487–504 (2015). 51. Zuo, Z. & Oerlemans, J. Modelling albedo and specific balance of the 22. Stibal, M. et al. Algae drive enhanced darkening of bare ice on the Greenland Greenland ice sheet: calculations for the Sondre Strømfjord transect. J. Glaciol. ice sheet. Geophys. Res. Lett. 44, 11463–11471 (2017). 42, 305–316 (1996). 23. Stibal, M. et al. Organic matter content and quality in supraglacial debris 52. Hock, R. Glacier melt: a review of processes and their modelling. Prog. Phys. across the ablation zone of the Greenland ice sheet. Ann. Glaciol. 51,1–8 Geogr. 29, 362–391 (2005). (2010). 53. Alexander, P. M. et al. Assessing spatio-temporal variability and trends in 24. Yallop, M. L. et al. Photophysiology and albedo-changing potential of the ice modelled and measured Greenland Ice Sheet albedo (2000-2013). Cryosphere algal community on the surface of the Greenland ice sheet. ISME J. 6, 8, 2293–2312 (2014). 2302–2313 (2012). 54. Tedesco, M. et al. The role of albedo and accumulation in the 2010 melting 25. Dumont, M. et al. Contribution of light-absorbing impurities in snow to record in Greenland. Environ. Res. Lett. 6, 014005 (2011). Greenland’s darkening since 2009. Nat. Geosci. 7, 509–512 (2014). 55. Box, J. E. et al. Greenland ice sheet albedo feedback: thermodynamics and 26. Keegan, K. M., Albert, M. R., McConnell, J. R. & Baker, I. Climate change and atmospheric drivers. Cryosphere 6, 821–839 (2012). forest fires synergistically drive widespread melt events of the Greenland Ice 56. van Angelen, J. H. et al. Sensitivity of Greenland Ice Sheet surface mass Sheet. Proc. Natl Acad. Sci. USA 111, 7964–7967 (2014). balance to surface albedo parameterization: a study with a regional climate 27. Wientjes, I. G. M., Van de Wal, R. S. W., Reichart, G. J., Sluijs, A. & model. Cryosphere 6, 1175–1186 (2012). Oerlemans, J. Dust from the dark region in the western ablation zone of the 57. Langen, P. L. et al. Liquid water flow and retention on the Greenland Ice Sheet Greenland ice sheet. Cryosphere 5, 589–601 (2011). in the regional climate model HIRHAM5: local and large-scale impacts. Front. 28. Pfeffer, W. T. & Bretherton, C. S. The effect of crevasses on the solar heating Earth Sci. 4, 110 (2017). of a glacier surface. In Proc. of the Vancouver Symposium, August 1987, IAHS 58. Hall, D. K. & Riggs, G. A. MODIS/Terra Snow Cover Daily L3 Global 500m Publ. no. 170, 191–205 (IAHS Publications, Wallingford, UK, 1987). Grid, Version 6 (NASA National Snow and Ice Data Center Distributed Active 29. Cathles, L. M., Abbot, D. S., Bassis, J. N. & MacAyeal, D. R. Modeling surface- Archive Center, Boulder, 2016). roughness/solar-ablation feedback: application to small-scale surface 59. Stroeve, J. C., Box, J. E. & Haran, T. Evaluation of the MODIS (MOD10A1) channels and crevasses of the Greenland Ice Sheet. Ann. Glaciol. 52,99–108 daily snow albedo product over the Greenland ice sheet. Remote Sens. Environ. (2011). 105, 155–171 (2006). 30. Langford, H., Hodson, A., Banwart, S. & Bøggild, C. The microstructure 60. van As, D. et al. Large surface meltwater discharge from the Kangerlussuaq and biogeochemistry of Arctic cryoconite granules. Ann. Glaciol. 51,87–94 sector of the Greenland ice sheet during the record-warm year 2010 (2010). explained by detailed energy balance observations. Cryosphere 6, 199–209 31. Cook, J., Edwards, A., Takeuchi, N. & Irvine-Fynn, T. Cryoconite: the dark (2012). biological secret of the cryosphere. Prog. Phys. Geogr. 40,66–111 (2015). 61. Ryan, J. C. et al. UAV photogrammetry and structure from motion to assess 32. Takeuchi, N. Optical characteristics of cryoconite (surface dust) on glaciers: calving dynamics at Store Glacier, a large outlet draining the Greenland ice the relationships between light absorbency and the property of organic matter sheet. Cryosphere 9,1–11 (2015). contained in the cryoconite. Ann. Glaciol. 34, 409–414 (2002). 62. Ryan, J. C. et al. How robust are in situ observations for validating satellite- 33. Remias, D. et al. Characterization of an UV- and VIS-absorbing, derived albedo over the dark zone of the Greenland Ice Sheet? Geophys. Res. purpurogallin-derived secondary pigment new to algae and highly abundant Lett. 44, 6218–6225 (2017). in Mesotaenium berggrenii (Zygnematophyceae, Chlorophyta), an 63. Howat, I. M., Negrete, A. & Smith, B. E. The Greenland Ice Mapping Project extremophyte living on glaciers. FEMS Microbiol. Ecol. 79, 638–648 (2012). (GIMP) land classification and surface elevation data sets. Cryosphere 8, 34. Stibal, M., Šabacká, M. & Žárský, J. Biological processes on glacier and ice 1509–1518 (2014). sheet surfaces. Nat. Geosci. 5, 771–774 (2012). 64. Pedregosa, F., Weiss, R. & Brucher, M. Scikit-learn: machine learning in 35. Stibal, M. et al. Microbial abundance in surface ice on the Greenland Ice Sheet. Python. J. Mach. Learn. Res. 12, 2825–2830 (2011). Front. Microbiol. 6, 225 (2015). 65. Canny, J. A computational approach to edge detection. IEEE Trans. Pattern 36. Lutz, S. et al. The biogeography of red snow microbiomes and their role in Anal. Mach. Intell. 8, 679–698 (1986). melting arctic glaciers. Nat. Commun. 7,1–9 (2016). 66. Knap, W. H. & Reijmer, C. H. Anisotropy of the reflected radiation field over 37. Ryser, C. et al. Cold ice in the ablation zone: Its relation to glacier hydrology melting glacier ice: measurements in Landsat TM bands 2 and 4. Remote Sens. and ice water content. J. Geophys. Res. Earth Surf. 118, 693–705 (2013). Environ. 65,93–104 (1998). 38. Lüthje, M., Pedersen, L. T., Reeh, N. & Greuell, W. Modelling the evolution of supraglacial lakes on the west Greenland ice-sheet margin. J. Glaciol. 52, 608–618 (2006). Acknowledgements 39. Tedesco, M. & Steiner, N. In-situ multispectral and bathymetric J.C.R. was funded by an Aberystwyth University Doctoral Career Development Scho- measurements over a supraglacial lake in western Greenland using a remotely larship (DCDS) and is currently funded by the NASA Cryosphere Program grant controlled watercraft. Cryosphere 5, 445–452 (2011). NNX14AH93G. The field camp, logistics and J.B. were supported by the Dark Snow 40. Fitzpatrick, A. A. W. et al. A decade (2002-2012) of supraglacial lake volume Project (http://www.darksnow.org/). We also gratefully acknowledge the UK Natural estimates across a land-terminating margin of the Greenland Ice Sheet. Environment Research Grants NE/H024204/1 and NE/G005796/1, the Aberystwyth The Cryosphere Dicuss. 7, 1383–1414 (2014). University Research Fund and the Danish Villum Young Investigator Programme grant 41. Leeson, A. et al. Supraglacial lakes on the Greenland ice sheet advance inland VKR 023121 awarded to M.S. A.H. acknowledges support from the Centre for Arctic Gas under warming climate. Nat. Clim. Chang. 5,51–55 (2015). Hydrate, Environment and Climate, funded by the Research Council of Norway through 42. Hubbard, A. L., Blatter, H., Nienow, P., Mair, D. & Hubbard, B. Comparison its Centres of Excellence (grant 223259). J.C. acknowledges a Rolex Award for Enterprise of a three-dimensional model for glacier flow with field data from Haut and L.C.S. acknowledges the NASA Cryosphere Program grant NNX14AH93G. The Glacier d’ Arolla, Switzerland. J. Glaciol. 44, 368–378 (1998). automatic weather stations used to validate the MODIS albedo data were funded and 43. van der Veen, C. J. Crevasses on glaciers. Polar Geogr. 23,1–33 (1999). installed by the Greenland Analogue Project (GAP) and maintained by the Geological 44. van de Wal, R. S. W. et al. Large and rapid melt-induced velocity changes in Survey of Denmark and Greenland (GEUS). We thank Peter Sinclair and Gabriel Warren the ablation zone of the Greenland Ice Sheet. Science 321, 111–113 (2008). for their assistance in the field and our three reviewers. 45. Doyle, S. H. et al. Persistent flow acceleration within the interior of the Greenland ice sheet. Geophys. Res. Lett. 41, 899–905 (2014). 46. Christoffersen, P., Bougamont, M., Hubbard, A., Doyle, S. & Grigsby, S. Author contributions Tensile shock triggers cascading lake drainage on the Greenland Ice Sheet. J.C.R. and A.H. designed, tested and built the UAVs. A.H. and J.B. were the project PIs Nat. Commun. https://doi.org/10.1038/s41467-018-03420-8 (2018). and J.B. provided UAV onboard pyranometers and data logger. J.C.R. and J.B. NATURE COMMUNICATIONS (2018) 9:1065 DOI: 10.1038/s41467-018-03353-2 www.nature.com/naturecommunications 9 | | | ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/s41467-018-03353-2 programmed and deployed the UAV. J.C.R. analysed the data and wrote the original Open Access This article is licensed under a Creative Commons manuscript with A.H. who also provided direction/supervision. K.C. provided the Attribution 4.0 International License, which permits use, sharing, ground photographs used for qualitative assessment of surface types. A.H., M.S., J.B., J.C., adaptation, distribution and reproduction in any medium or format, as long as you give T.D.I.-F., L.C.S. and K.C. provided conceptual and technical advice and contributed to appropriate credit to the original author(s) and the source, provide a link to the Creative data interpretation. All authors commented and critically revised the manuscript. Commons license, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the Additional information article’s Creative Commons license and your intended use is not permitted by statutory Competing interests: The authors declare no competing interests. regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this license, visit http://creativecommons.org/ Reprints and permission information is available online at http://npg.nature.com/ licenses/by/4.0/. reprintsandpermissions/ Publisher's note: Springer Nature remains neutral with regard to jurisdictional claims in © The Author(s) 2018 published maps and institutional affiliations. 10 NATURE COMMUNICATIONS (2018) 9:1065 DOI: 10.1038/s41467-018-03353-2 www.nature.com/naturecommunications | | | http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Nature Communications Springer Journals

Dark zone of the Greenland Ice Sheet controlled by distributed biologically-active impurities

Loading next page...
1
 
/lp/springer_journal/dark-zone-of-the-greenland-ice-sheet-controlled-by-distributed-sFgU0dyFBn

References (60)

Publisher
Springer Journals
Copyright
Copyright © 2018 by The Author(s)
Subject
Science, Humanities and Social Sciences, multidisciplinary; Science, Humanities and Social Sciences, multidisciplinary; Science, multidisciplinary
eISSN
2041-1723
DOI
10.1038/s41467-018-03353-2
pmid
29540720
Publisher site
See Article on Publisher Site

Abstract

ARTICLE DOI: 10.1038/s41467-018-03353-2 OPEN Dark zone of the Greenland Ice Sheet controlled by distributed biologically-active impurities 1,2,3 1,4 5,6 1 7 Jonathan C. Ryan , Alun Hubbard , Marek Stibal , Tristram D. Irvine-Fynn , Joseph Cook , 2 8 9 Laurence C. Smith , Karen Cameron & Jason Box Albedo—a primary control on surface melt—varies considerably across the Greenland Ice Sheet yet the specific surface types that comprise its dark zone remain unquantified. Here we use UAV imagery to attribute seven distinct surface types to observed albedo along a 25 km transect dissecting the western, ablating sector of the ice sheet. Our results demonstrate that distributed surface impurities—an admixture of dust, black carbon and pigmented algae— explain 73% of the observed spatial variability in albedo and are responsible for the dark zone itself. Crevassing and supraglacial water also drive albedo reduction but due to their limited extent, explain just 12 and 15% of the observed variability respectively. Cryoconite, con- centrated in large holes or fluvial deposits, is the darkest surface type but accounts for <1% of the area and has minimal impact. We propose that the ongoing emergence and dispersal of distributed impurities, amplified by enhanced ablation and biological activity, will drive future expansion of Greenland's dark zone. 1 2 Centre for Glaciology, Department of Geography and Earth Sciences, Aberystwyth University, Aberystwyth SY23 3DB, UK. Department of Geography, University of California, Los Angeles, Los Angeles, CA 90095, USA. Institute at Brown for Environment and Society, Brown University, Providence, RI 4 5 02906, USA. Centre for Arctic Gas Hydrate, Environment and Climate, Department of Geology, University of Tromsø, 9037 Tromsø, Norway. Department of Ecology, Faculty of Science, Charles University, 12844 Prague, Czech Republic. Department of Geochemistry, Geological Survey of Denmark and 7 8 Greenland, 1350 Copenhagen, Denmark. Department of Geography, University of Sheffield, Sheffield S10 2TN, UK. Institute of Biological, Environmental and Rural Sciences, Aberystwyth University, Aberystwyth SY23 3DB, UK. Department of Glaciology and Climate, Geological Survey of Denmark and Greenland, 1350 Copenhagen, Denmark. Correspondence and requests for materials should be addressed to A.H. (email: abh@aber.ac.uk) NATURE COMMUNICATIONS (2018) 9:1065 DOI: 10.1038/s41467-018-03353-2 www.nature.com/naturecommunications 1 | | | 1234567890():,; ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/s41467-018-03353-2 he Greenland Ice Sheet has become the largest cryospheric the specific ice surfaces across the dark zone and determine their contributor to global sea-level rise predominantly through impact on the mesoscale (1–10 km) albedo distribution during Tincreased surface melt and runoff, which accounts for over peak melt season, as represented by the MODIS albedo product, 1–4 half of its mass loss since 1991 . The dominant energy source MOD10A1. On 8 August 2014, a fixed-wing UAV equipped with for snow and ice melt is direct solar shortwave radiation, the a digital camera and upward and downward facing pyranometers absorption and reflection of which is predominantly modulated was deployed from a field camp based in the vicinity of the K- 5–7 by surface albedo . Accurately constraining spatiotemporal transect, S6 automated weather station (AWS) on a 25 km east- patterns of albedo across the ice sheet is hence fundamental to west transect dissecting the dark zone (Fig. 1). Seven distinct understanding and predicting surface melt and runoff along with surface types were visually identified on the ground by an expert their impact on ice sheet flow dynamics and sea-level rise. A and automatically classified based on their reflectance and conspicuous feature of Greenland’s ablation area is its dark zone, roughness properties. The survey transect was divided into sixty an area of bare ice with particularly low albedo that appears 500 × 500 m segments, co-located to the footprints of corre- across the west and southwest sectors of the ice sheet each sponding MODIS pixels, and the fractional area of each surface 8–10 summer . At the Arctic Circle, in the vicinity of the Kanger- type in each segment was determined using a supervised k- lussuaq (K-) sector, the dark zone extends between 20 and 75 km nearest neighbours (k-NN) classification (see Methods section: from the land-terminating margin where Moderate Resolution Surface Classification for more information) (Figs. 1, 2). Finally, Imaging Spectroradiometer (MODIS) data indicate a regional the mean albedo of each surface type was derived from the digital albedo minimum of ~0.34 . From 2000 to 2012, the spatial extent imagery and the relative contribution of different surface types to of the dark zone increased by 12% but also exhibited considerable mesoscale albedo variability (defined by MOD10A1 pixels) was 11,12 interannual variability . The extent of the dark zone is weakly calculated using principal component regression (PCR). positively correlated with air temperature and negatively corre- 11,12 lated with solar radiation during June, July and August (JJA) . Results This suggests that the ongoing albedo decrease observed during Surface type variation along the transect. Analysis of all UAV the melt season is not simply driven by melting of the winter imagery allowed us to visually identify and automatically classify snowpack to reveal the darker bare ice surface beneath, but, fol- seven distinct surface types across the survey transect: (i) clean lowing exposure, there are changes in the nature of the bare ice ice, (ii) ice containing uniformly distributed impurities, (iii) deep 11,12 itself . However, the surface characteristics of the dark zone water, (iv) shallow water, (v) cryoconite either in holes or fluvial remain unquantified because the spatial resolution of satellite deposits, (vi) crevasses and (vii) snow (Fig. 3). Distinction imagery is insufficient to fully resolve the specific surface types between clean ice and ice containing uniformly distributed that comprise it, and how these surfaces evolve through time, impurities was guided by qualitative assessment of 112 oblique distinguishing the dark zone from brighter ice surfaces adjacent and nadir photographs taken from the ground (<1 cm pixel to it. footprint) at specific study sites around the field camp (Fig. 1). Previous field-based, in situ observations indicate that western These images confirm that distributed impurities across the ice Greenland’s ablation zone is characterized by highly variable non- surface are responsible for bare ice albedo variability at the local 8,13–15 ice constituents and surface structures . These include fea- scale (1–10 m) (Fig. 4). In order to upscale and understand the 16,17 tures such as crevasses, fractures and foliations ; supraglacial impact of these surface impurities on the mesoscale albedo dis- hydrological features, including streams, rivers, ponds and tribution of the ablation zone, we divided bare ice into two 18,19 lakes ; snow patches and fracture cornices; cryoconite, con- categories: clean ice, with very low impurity concentrations, and 20,21 centrated in holes or in supraglacial fluvial deposits ; microbes ice containing some or an abundance of impurities. It is apparent 22–24 and their humic by-products ; mineral dust and aerosols that additional categories could be defined for bare ice given from outcropping or contemporary aeolian deposition including sufficiently high pixel resolution, but for the purpose of this study, 10,25,26 black carbon from wildfires and other aerosols. While the and considering the spectral limitations of the onboard camera, highest resolution optical satellite imagery currently available has we do not attempt to. We note that it would be a fruitful direction facilitated the examination of crevasse fields and surficial with multi- and hyper-spectral sensor payloads. Clean ice has hydrology , a quantitative assessment of the specific ice surface 57.2% aerial coverage in the lower, western half of the survey types that comprise the dark zone, and how they combine to yield transect between 0 and 17 km (Fig. 3). In the eastern half (17–27 observed albedo patterns across the ablation zone of the ice sheet, km), clean ice coverage is lower at 23.0%. Ice containing uni- has yet to be made. formly distributed impurities (Fig. 5a) varies inversely to clean Here, we utilize high-resolution (15 cm pixel size) imagery ice, with a higher fraction in the eastern half (74.5%) compared to acquired from an unmanned aerial vehicle (UAV) to characterize the western half of the transect (40.0%) (Fig. 3). Dark zone 50.0°W 49.5°W 49.0°W 67.1°N 5c Fig.5a 5b Russell Glacier 30 km 6c 6b Fig.6a 15 km 0 km 0 5 km UAV footprint Camp Fig. 1 Overview map showing location of UAV survey transect. The background is a Landsat 8 Operational Land Imager (OLI) true colour image of the Kangerlussuaq sector of the Greenland Ice Sheet from 6 August 2014. The transect was divided into sixty 0.25 km segments for comparison with the MODIS albedo product, MOD10A1. High-resolution aerial imagery and surface classification of six segments (coloured red) are shown in Figs. 5 and 6 2 NATURE COMMUNICATIONS (2018) 9:1065 DOI: 10.1038/s41467-018-03353-2 www.nature.com/naturecommunications | | | NATURE COMMUNICATIONS | DOI: 10.1038/s41467-018-03353-2 ARTICLE MOD10A1 pixel 500 m Fig. 2 Schematic summarizing the aims of the study. Aerial digital imagery are used to characterize the surface types that are found in the ablation zone and assess their impact on mesoscale spatial albedo patterns as represented by MODIS 6a 6b 6c 5a 5b 5c 0.50 0.45 MOD10A1 0.40 albedo 0.35 (unitless) 0.30 0.25 0.60 Clean ice fraction 0.40 (unitless) 0.20 0.00 Distributed 0.85 impurities 0.65 fraction 0.45 (unitless) 0.25 Surface 0.30 water 0.20 fraction 0.10 (unitless) 0.00 0.06 Crevasse 0.04 fraction 0.03 (unitless) 0.01 0.00 0.02 Cryoconite 0.01 fraction 0.01 (unitless) 0.00 0.00 0.02 Snow 0.01 fraction 0.01 (unitless) 0.00 0.00 510 15 20 25 30 W E Distance from most westerly point of transect (km) Fig. 3 Variation of albedo and the fractional area of each surface type across the UAV transect. The albedo and fractional areas derived from MOD10A1 and the UAV imagery, respectively, on 8 August 2014. The x axis is displayed in Fig. 1. The results of the classification for six segments, highlighted by the vertical grey bars, are shown in Figs. 5 and 6 Cryoconite is commonly found in holes but also in fluvial concentrated, rather than distributed, impurities. Cryoconite deposits near supraglacial streams and lakes (Fig. 6c). In this has a maximum aerial coverage of 1.6% at 8 km from the western study, cryoconite is distinguished from ice containing distributed end of the transect and a mean coverage of 0.6% across the entire impurities by its very low albedo, which is indicative of transect (Fig. 3). It is possible that we underestimate the fractional NATURE COMMUNICATIONS (2018) 9:1065 DOI: 10.1038/s41467-018-03353-2 www.nature.com/naturecommunications 3 | | | 500 m ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/s41467-018-03353-2 a b Fig. 4 Photograph showing close-up of bare ice found in the ablation zone. The photographs were taken near the field camp located close to the S6 automated weather station at ~1000 m a.s.l. (Fig. 1). a Ice containing distributed impurities on the surface and b clean ice with cryoconite holes a Original Albedo Classified MOD10A1 Albedo = 0.29 Ice containing Fraction distributed 0.037 impurities 0.951 0.011 200 m 0.001 b MOD10A1 Albedo = 0.27 Fraction 0.014 0.933 0.046 0.007 MOD10A1 Albedo = 0.36 Shallow surface Deep water Fraction surface 0.249 water Clean ice 0.421 0.141 0.187 0 Albedo 1 0.002 Clean ice Distributed impurities Deep surface water Shallow surface water Cryoconite Fig. 5 RGB digital image, albedo map and classification of surface types in three MOD10A1 pixels. The albedo maps were derived from the digital images (Methods). The locations of the segments along the UAV transect are shown in Fig. 1. a Segment characterized by mostly ice containing uniformly distributed impurities. b Segment characterized by similar ice surface to a but with a larger fraction of channelized surface melt-water. c Segment dominated by a supraglacial lake with a previous shore consisting of clean ice 20,21 area of small cryoconite holes due to the limited, 15 cm pixel surrounding them . Coincident field measurements, made resolution of our UAV imagery. However, we note that smaller during UAV image acquisition, indicate that the cryoconite holes cryoconite holes (<15 cm) would also be hidden from virtually all observed in our study were well developed. The implication is aerial and satellite imagery obtained at low solar elevation angles. that once they have attained equilibrium depth, they cease to Furthermore, cryoconite hole depths tend to equilibrate as the absorb additional energy (which would make them deeper) melt season progresses, due to their low albedo and preferential compared to surrounding ice and hence are effectively neutralized radiative absorption in comparison to brighter ice surfaces from the effects of incoming solar radiation. For these reasons, we 4 NATURE COMMUNICATIONS (2018) 9:1065 DOI: 10.1038/s41467-018-03353-2 www.nature.com/naturecommunications | | | NATURE COMMUNICATIONS | DOI: 10.1038/s41467-018-03353-2 ARTICLE a Original Albedo Classified MOD10A1 Albedo = 0.38 Crevasse Fraction 0.558 0.381 0.002 0 Albedo 1 200 m 0.059 b MOD10A1 Albedo = 0.47 Clean ice Fraction 0.710 0.279 0.003 0.008 MOD10A1 Albedo = 0.37 Fraction 0.682 0.293 Cryoconite 0.016 0.009 Clean ice Distributed impurities Cryoconite Shallow surface water Crevasse Fig. 6 RGB digital image, albedo map and classification of surface types in three more MOD10A1 pixels. The albedo maps were derived from the digital images (Methods). The locations of the segments along the UAV transect are shown in Fig. 1. a Segment containing a high fraction of crevasses. b Segment characterized by a much lower relief surface and no crevasses. c Segment characterized by clean ice and numerous cryoconite holes argue that smaller cryoconite holes had a minimal net impact on area of ice containing uniformly distributed impurities explains MODIS-derived albedo compared to the ice surface surrounding 73% of the observed mesoscale albedo variability. Although not them. the darkest surface type observed, distributed impurities dom- At 23 km (all distances refer from the western start point of the inate the mesoscale albedo signal due to their extensive coverage UAV transect), a braided meltwater channel network with a and large variations in their fractional area across the survey fractional area of 5.3%, intersects the transect (Figs. 1 and 5b), but transect. Distributed impurities have been attributed to the out- otherwise, surface water comprises only 1.9% of the survey area cropping of aeolian dust deposited during the early Holo- 19 9,10,27 (Fig. 3). These results are consistent with Smith et al. who found cene and/or pigmented surface algal blooms and associated 2 22–24 that surface water accounted for 1.4% of a ~5000 km bare ice humic material . area in the K-sector of the ice sheet. At 28 km, a small, 0.83 km Locally, defined at the scale of a single MODIS pixel, supraglacial lake covers 32.8% of the segment (Fig. 5c). Crevasse supraglacial water, contained in both lakes and channels, has a density is highest on the western flank between 2 and 5 km and distinct impact on albedo due to its low albedo (α = 0.19–0.26). attains maximum coverage of 5.9% at 2 km (Figs. 3, 6a). Up- The segment of the transect that corresponds to a large braided glacier of 10 km, crevasses are almost entirely absent (Fig. 6b). channel network at 23 km has an albedo of 0.28 (Fig. 5b). This is Remnant snow patches, which persist within ice fractures and ~0.02 lower than the surrounding segments with <1% surface supraglacial channel incisions, attain a maximum coverage of water yet are otherwise composed of similar ice surfaces (Figs. 3, 1.7% with a mean aerial coverage of only 0.1%, at this time of 5a). However, in comparison to distributed impurities, supragla- year. cial water has a minor impact on the mesoscale albedo pattern and explains only 15% of the albedo variability across our survey transect. Surprisingly, the supraglacial lake located at 28 km is not associated with a significant reduction of MOD10A1 albedo. This Relationship between surface types and albedo patterns. The is because it is relatively narrow (260 m width) and covers just mesoscale albedo distribution, determined from MOD10A1 data, 33% of the segment, while 25% of the remaining segment consists exhibits considerable variability along the survey transect with of very bright clean ice faculae (mean α = 0.58): interpreted as a values between 0.27 and 0.47 (Figs. 1, 3). The dark zone, which is shoreline exposed when the lake level dropped (Fig. 5c). Hence, approximately located between 18 and 27 km along the survey the low albedo of the lake water (mean α = 0.19) itself is offset by transect, has a mean albedo of 0.29 and is characterized by dis- the brightness of the surrounding ice surface adjacent to it tinct and conspicuous banding that specifically relate to foliation yielding minimal change in the net albedo (MOD10A1) signature. structures apparent in Landsat 8 imagery (Fig. 1). Between 80 and Crevassing explains 12% of albedo variability and its impact is 95% of the dark zone is classified as ice containing uniformly distributed impurities (mean albedo (α) = 0.27), with the well illustrated at 3–4 km where a transition into a crevasse zone yields a significant reduction in mesoscale albedo compared to the remaining 5–20% consisting of predominately clean ice (mean α = 0.55) (Fig. 5a). Application of PCR reveals that the fractional adjacent, flatter surface (Fig. 6a, b). Crevasses enhance shortwave NATURE COMMUNICATIONS (2018) 9:1065 DOI: 10.1038/s41467-018-03353-2 www.nature.com/naturecommunications 5 | | | ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/s41467-018-03353-2 radiation absorption; radiative transfer modelling indicates that summer ablation alone . Instead, positive correlations between the presence of crevasses can double the downward energy the dark zone extent and proxies for the availability of liquid absorbed relative to a homogeneous, flat ice surface and reduce water and nutrients are interpreted as evidence that blooms of 28,29 albedo by between 0.10 and 0.25 . The amount of radiation surface ice algae control bare ice albedo across the dark zone. Any absorbed by crevasses is determined by their size, orientation, increase in temperature and/or liquid water production in the density and whether they are water-filled. One of the most presence of dust promotes further colonization of surface algae densely crevassed areas across the transect (5.9% fractional area) yielding an increase in pigmented biomass and net albedo 24,36 (Fig. 6a), with crevasse widths up to ~10 m and depths in excess reduction . The 12% expansion of the dark zone between 2000 of 8 m, yields an albedo reduction of ~0.06 in comparison to the and 2014 in western Greenland, corresponding with an increase segment in Fig. 6b, which has no crevassing but similar fractions in mean summer air temperature of 0.13 °C per year over the of other surface types (Figs. 3, 6b). Elsewhere, at 6–7 km, smaller same period , provides further support for these ongoing crevasses, with mean widths of 5 m, have a reduced impact on processes. mesoscale albedo, lowering it by only 0.02 in comparison to the While our results attest that the variation in the fractional area control segment in Fig. 6b. This observation is at odds with of supraglacial lakes, streams, crevasses and cryoconite do not radiative transfer modelling results because the modelled significantly affect mesoscale albedo, they still play a secondary crevasses were compared against a flat, clean ice surface, which role in determining interannual and seasonal albedo variability. 28 29 is not the case here (Fig. 6b) . Cathles et al. modelled crevasses For example, supraglacial water may act to consolidate or dis- with width to depth ratios similar to the crevasses we observed tribute sediment and impurities across the ice sheet surface . and found that they have a melt enhancement factor of 1.14–1.20 Moreover, a relatively small expansion in the spatial extent of at solar zenith angles of 45°, which is in broad agreement with our surface water would have a disproportionate impact on mesoscale findings. albedo and further amplify ablation. Melt rates at the base of Increases in the fractional area of cryoconite, either in large supraglacial lakes and water bodies are double that of bare ice 38,39 holes or fluvial deposits, are not particularly associated with surfaces due to enhanced shortwave radiation absorption . mesoscale albedo reduction across the survey transect, and Atmospheric warming has been shown to increase the spatial 40,41 surprisingly, the lowest concentration of cryoconite is actually extent and duration of ponded supraglacial water . During observed within the dark zone itself (Fig. 5a). Cryoconite only years with higher summer temperatures, such as 2007, 2010 and occupies a very small fraction of the total coverage (1.6% 2012, supraglacial lakes formed earlier in the season and occupied maximum and 0.6% mean), which can be explained by the nature a 40% larger area than in cooler summers . It follows that of the cryoconite material itself. The thread-like, filamentous increased storage of water in supraglacial lakes will play an structure of cyanobacteria enables them to entangle debris and important role in net albedo reduction across an expanding bare facilitate the formation of granules. These granules absorb more ice area in future. solar radiation and melt down into the ice until they are in Our analysis also demonstrates that crevasses reduce local 30–32 radiative and thermodynamic equilibrium . Although this (0.1–1 km) albedo, and hence any increase in crevasse extent will mechanism means that the cryoconite hole has a very low albedo impact on mesoscale albedo patterns. Crevasses form due to value (mean α = 0.10) when observed from directly above, the localized concentration of tensile stresses which, due to highly hole occupies a relatively small area and is effectively hidden at variable subglacial conditions and longitudinal stress coupling, non-zenith solar illumination resulting in an increase in are spatially and temporally variable across the Greenland Ice 20 16,42,43 mesoscale albedo . Furthermore, cryoconite holes are often . In response to increased surface melt, GPS obser- Sheet covered by an ice lid, caused by the refreezing of water that has vations by van de Wal et al. report reduced net flow over the 31 45 filled the hole during negative net radiation conditions . While marginal zone of the K-transect, whereas Doyle et al. report thin frozen lids may undergo partial or complete ablation during persistent ice flow acceleration above the equilibrium line up to the day, their higher albedo acts to further moderate the impact of 140 km from the ice sheet margin in this same sector. Recent 46 46,47 cryoconite that they cover and render the holes indistinguishable modelling and observations of new crevasses forming over from the adjacent ice surface (much to the dismay of many a field 160 km from the western margin reveals that the ice sheet interior campaigner with sodden feet). is also more dynamically sensitive to transient stress perturba- tions originating from downstream than a previous steady-state model suggests . Regardless of ice dynamics, inland migration of Discussion the equilibrium line caused by atmospheric warming will drive The analysis presented here demonstrates that the dark zone has increased bare ice extent, further exposing existing crevasses that low fractional areas of surface water (<1.0%), cryoconite holes were formerly snow and firn covered. Hence, surface crevasse (<0.5%) and crevasses (<0.2%). Instead, it appears that ice con- extent will likely expand in future, resulting in mesoscale albedo taining uniformly distributed impurities draped over a relatively reduction and enhanced surface absorption of incoming energy flat surface are the primary agent responsible for the low available for melt. (MOD10A1) albedo values observed during the melt season Finally, future spatial expansion of cryoconite does have the (Fig. 5a). Near the S6 AWS, from where the UAV was launched, potential to significantly impact surface albedo. Hodson et al. Stibal et al. report that samples of distributed impurities consist showed that 53% of plot-scale (0.01–0.5 m) variation in albedo of an abundance of ice algae (Fig. 4a), which are characterized by was correlated with the growth of cryoconite holes, and Chandler a grey/brown hue due to the brown-to-purple coloured pigments et al. report that the gradual seasonal reduction in albedo also 24,33,34 surrounding the algae chloroplasts . Correlations between correlates with an increase in cryoconite hole size and number. dust content and the abundance of microbes suggest that the An increase in the extent of cryoconite holes may be caused by melt-out of particulates may provide nutrients for surface ice longer and warmer ablation seasons, which would increase the 22,35 12 algae to grow and indirectly control the extent of dark zone . heat energy to the walls and base of the hole, leading to further Further support for this hypothesis is provided by Tedstone melting and hole expansion . On the other hand, an increase of et al. who argue that the large interannual variability in the meltwater may promote aggregation of distributed impurities and extent of the dark zone, and its significant reduction in 2013 and could have a surface cleaning effect, potentially raising the albedo 20,31 2015, demonstrates that bare ice albedo is not a consequence of of the surrounding bare ice . The growth and development of 6 NATURE COMMUNICATIONS (2018) 9:1065 DOI: 10.1038/s41467-018-03353-2 www.nature.com/naturecommunications | | | NATURE COMMUNICATIONS | DOI: 10.1038/s41467-018-03353-2 ARTICLE cryoconite holes on mesoscale albedo is hence complex and still surface particulates and nutrients fertilizes pigmented ice surface somewhat ambiguous. algae, which drives albedo reduction over the duration of the melt 12,22 In this study, we characterized the spatial variability of surface season . Further research though is required to determine how types across the western ablating margin of the ice sheet towards these factors combine to increase the spatial extent and con- the end of the melt season when bare ice surfaces are most centration of pigmented surface algae, and their interaction with apparent. However, for much of the year (September to May), the the availability of in situ and aeolian-derived nutrients, changing ice sheet is snow-covered and it is likely that snow grain size and atmospheric forcing and enhanced ice melt and runoff. impurity concentration govern mesoscale albedo patterns across the K-transect and elsewhere during this period. Likewise, early Methods melt-season albedo patterns are primarily governed by the rela- MODIS albedo. Albedo patterns were determined from a MOD10A1 C6 broad- tive proportions of snow and ice extent. Accurately determining band (spectral range of 300–3000 nm) albedo product from the 8 August 2014 and available from the National Snow and Ice Data Center (NSIDC) . MOD10A1 is snow melt and the timing of bare ice exposure has therefore been gridded in a sinusoidal map projection and has a resolution of ~500 × 500 m or a priority for surface mass and energy balance models and the 0.25 km . The value of each pixel represents the best single albedo observation in theoretical determinants of snow albedo and melt are relatively 59 the day based on cloud cover and viewing and illumination angles . We estimated 5,50 well established . In contrast, few studies have investigated the that MOD10A1 has a root mean square difference (RMSD) of 7.0% in comparison to albedo measured by CNR1 or CNR4 thermopile pyranometers at the PRO- albedo of the bare ice surface types that characterize the ablation MICE/GAP automatic weather stations KAN-L, KAN-M and KAN-U between zone and they have commonly been treated as temporal and 60 59 2009 to 2014 . This compares well to Stroeve et al. who estimated an RMSD of 51–53 spatial constants in surface melt models . 6.7%. The observed spatiotemporal variability in albedo across the 54,55 ablation zone has motivated a new generation of surface UAV platform. Aerial imagery was acquired by a fixed-wing UAV identical to that energy balance models that assimilate spatial patterns of albedo 14,61,62 used by Ryan et al. The UAV has a 2.1 m wingspan and is powered by a 10 4,56,57 derived from MODIS data . Across our 25 km survey Ah, 16.8 V LiPo battery pack which, with a total weight of 4 kg, yields a 1 h endurance and 60 km range. The autonomous control system is based around an transect, the MODIS-derived surface albedo pattern is dominated Arduino navigation and flight computer updated in real-time by a 10 Hz data by variations in the extent of uniformly distributed impurities, a stream comprising of a GPS, magnetometer, barometer and accelerometer. These result that contradicts previous research attributing it to an data are logged along with a timestamp for each activation of the digital camera increased occurrence of supraglacial water . The source, pro- shutter which automatically triggers when a horizontal displacement threshold is cesses and drivers of distributed impurities are yet to be exceeded. The UAV was hand launched on 8 August 2014 from a base camp at 67.08°N, 49.40°W, located at the site of the Institute for Marine and Atmospheric unequivocally established, with some studies indicating a wind- Research (IMAU), University of Utrecht S6 automatic weather station. It was pre- blown origin, others revealing that they are derived from melt-out programmed to carry out a 25 km survey across the Kangerlussuaq sector of the 9–11 of englacial dust . Recent research has promoted the concept western Greenland Ice Sheet (Fig. 1). The Greenland Ice Mapping Project (GIMP) of bioalbedo, which argues that the melt-out and release of digital elevation model (DEM) was used during the selection of three- 250 250 Snow 200 200 Shallow water Clean ice Deep water 150 150 100 100 Distributed impurities 50 50 Cryoconite 0 0 0 50 100 150 200 250 0 50 100 150 200 250 Green DN Red DN 250 250 200 200 150 150 Crevasse 100 100 50 50 0 0 0 50 100 150 200 250 0 1 Green DN Crevasse Fig. 7 Scatter-plots showing the key attributes of the seven surface types identified in this study. DN is digital number of the Sony NEX-5N calibrated RAW image. Deep and shallow supraglacial water is easily distinguishable because it has low reflectivity in the red band and forms a unique cluster in the feature space. Snow, clean ice, ice containing distributed impurities and cryoconite form another cluster and are distinguishable because they reflect different proportions of RGB visible wavelengths NATURE COMMUNICATIONS (2018) 9:1065 DOI: 10.1038/s41467-018-03353-2 www.nature.com/naturecommunications 7 | | | Red DN Blue DN Green DN Blue DN ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/s41467-018-03353-2 dimensional waypoints to ensure the UAV maintained a constant altitude of 350 m vignette of all images acquired at nadir during the survey period. Barrel distortion above the surface during the autonomous sorties. On return from the sortie, the was corrected using ImageMagick (http://www.imagemagick.org/), which utilized UAV was manually landed into a 10 × 5 m net. the coefficients stored in the image’s ancillary metadata also known as exchange- able image file format data. We then corrected the images for changing illumination conditions during the Digital imagery. Digital imagery was acquired by a Sony NEX-5N digital camera survey using downward irradiance measured by a ground-based upward facing vertically mounted inside the front of the airframe. The camera has a 16 mm fixed Apogee SP-110 pyranometer. To do this, images of a 25 × 25 cm Teflon white focus lens (53.1 by 73.7° field of view) yielding an image footprint of ~525 × 350 m reference target were acquired every 10 min using the UAV digital camera from the during the autonomous sortie. The width of each image is approximately similar to ground. The relationship between the mean RGB DNs of the white reference target the pixel footprint of MODIS. The camera was preset with a fixed shutter speed of and the downward irradiance recorded by the upward facing pyranometer were 1/1000 s, ISO 100 and F-stop of 8, and triggered every 35 m to provide a 90% used to construct a calibration curve using a linear least squares regression (R = forward image overlap. The relatively fast shutter speed minimizes image blur 0.96). The ratio of reflected radiation recorded by the camera and the downward while the low ISO and F-number ensures maximum image quality where even the radiation estimated from the calibration curve enabled the illumination-corrected brightest surfaces do not saturate the image. The camera was set to record the reflectance of each pixel to be defined. images in RAW format, an image format that contains minimally processed data Since snow and ice are non-Lambertian surfaces, a nadir measurement of from the camera’s sensor. During the survey, ~2000 RAW images were acquired at reflectance underestimates albedo by between 1 and 5% in the visible band . The the camera’s maximum (4912 × 3264 pixels) resolution which, once the images illumination-corrected images were therefore calibrated again by multiplying the were corrected for barrel, or geometric, distortion, equates to a ground sampling image pixel numbers by a factor calculated by dividing the mean pixel value of the distance of ~11 cm. illumination-corrected image by the albedo recorded by upward and downward facing Apogee SP-110 pyranometers mounted on the UAV. Ryan et al. found that albedo determined using this method has an accuracy of ±5% over ice sheet Orthomosaic and DEM generation. The R(ed)/G(reen)/B(lue) images were used surfaces typically found in the ablation zone. to produce an orthomosaic and DEM using Agisoft PhotoScan Pro (http://www. agisoft.com/) following the processing sequence described by Ryan et al. The images were georeferenced by providing latitude, longitude and altitude data Data availability. The MODIS (MOD10A1) albedo data are available from the recorded by the flight controller. The orthomosaic was produced in the software’s National Snow and Ice Data Center (NSIDC) at http://nsidc.org/data/MOD10A1. ‘mosaic’ mode, meaning that pixels in the centre of the images were preferentially The UAV images are archived in the PANGAEA repository: https://doi.pangaea. used to provide the output pixel value. The orthomosaic and DEM were nearest de/10.1594/PANGAEA.885798. neighbour resampled to a ground resolution of 15 cm and 50 cm, respectively. We divided the orthomosaic into sixty 0.25 km segments and each segment was assigned a MOD10A1 value. Received: 21 June 2017 Accepted: 7 February 2018 Surface classification. The fractional area of each surface type was calculated by dividing the number of pixels of each surface type by the total number of pixels in each orthomosaic segment. The number of pixels of each surface type was esti- mated using a supervised k-NN classification from the scikit-learn Python mod- References ule . The pixels were classified using a majority vote based on the Euclidean distance to five equally weighted nearest neighbours (Fig. 7). The k-NN was 1. Box, J. E. et al. Changes to Arctic land ice. In Snow, Water, Ice and Permafrost in the Arctic (SWIPA) 137–168 (Arctic Monitoring and Assessment manually trained with seven distinct and visually identified surfaces found in the orthomosaic: (i) clean ice, (ii) ice containing uniformly distributed impurities, Programme (AMAP), Oslo, Norway, 2017). (iii) deep water, (iv) shallow water, (v) cryoconite either in holes or fluvial deposits, 2. Hanna, E. et al. Ice-sheet mass balance and climate change. Nature 498,51–59 (vi) crevasses and (vii) snow. The training samples of the surface types were (2013). manually digitized from 10 orthomosaic segments based on RGB brightness and a 3. Enderlin, E. M. et al. An improved mass budget for the Greenland ice sheet. layer that specified whether or not the pixel was situated within a crevasse or Geophys. Res. Lett. 41, 866–872 (2014). fracture. This roughness layer was determined by calculating the residual between 4. van den Broeke, M. R. et al. On the recent contribution of the Greenland ice the original 50 cm DEM subtracted from a 30 m Gaussian-smoothed DEM. sheet to sea level change. Cryosphere 10, 1933–1946 (2016). Negative anomalies with a vertical displacement >1 m were identified as crevasses. 5. Gardner, A. S. & Sharp, M. J. A review of snow and ice albedo and the Small cracks and fractures were detected on the basis of sharp RGB contrast, and development of a new physically based broadband albedo parameterization. J. were discriminated using an edge detector algorithm . Pixels within 2 m of a linear Geophys. Res. Earth Surf. 115,1–15 (2010). feature were also identified as crevasses. 6. van den Broeke, M. R. et al. Partitioning of melt energy and meltwater fluxes The efficacy of the k-NN classifier was evaluated by comparison with in the ablation zone of the west Greenland ice sheet. Cryosphere 2, 179–189 independently digitized surface types in three orthomosaic segments, at the centre (2008). and extreme ends of the transect. We found that 92% of the pixels were classified 7. Mikkelsen, A. B. et al. Extraordinary runoff from the Greenland ice sheet in accurately. The k-NN classified crevasses with an accuracy of 88% and performed 2012 amplified by hypsometry and depleted firn retention. Cryosphere 10, better for deep and shallow water (96%) than for cryoconite, ice containing 1147–1159 (2016). distributed impurities, clean ice and snow (90%). The classification of supraglacial 8. Knap, W. H. & Oerlemans, J. The surface albedo of the Greenland ice sheet: water is relatively accurate because water has low reflectance in the red band and satellite-derived and in situ measurements in the Søndre Strømfjord area forms a unique cluster in the feature space (Fig. 7). The performance of the k-NN during the 1991 melt season. J. Glaciol. 42, 364–374 (1996). for clean ice and ice containing distributed impurities is less accurate (90%) 9. Wientjes, I. G. M. & Oerlemans, J. An explanation for the dark region in the because these surfaces reflect the RGB visible bands in similar relative proportions western melt zone of the Greenland ice sheet. Cryosphere 4, 261–268 (2010). and the surface classes overlap in the feature space (Fig. 7). Misclassification could 10. Wientjes, I. et al. Carbonaceous particles reveal that Late Holocene dust causes also be caused by shadows, especially in segments with steep topography (eg, the dark region in the western ablation zone of the Greenland ice sheet. J. Fig. 6b). Shadows increase the proportions of the surface classified as ice containing Glaciol. 58, 787–794 (2012). distributed impurities and/or cryoconite. Accounting for shadows would reduce 11. Shimada, R., Takeuchi, N. & Aoki, T. Inter-annual and geographical variations the proportion of ice containing distributed impurities in the crevassed zone in the extent of bare ice and dark ice on the Greenland Ice Sheet derived from (Fig. 6b) and subsequently increase their variation across the transect. This would MODIS satellite images. Front. Earth Sci. 4,1–10 (2016). make the impact of distributed impurities on the mesoscale albedo variability, and 12. Tedstone, A. J. et al. Dark ice dynamics of the south-west Greenland Ice Sheet. the conclusions of this study, more significant. Finally, we used PCR to explore the Cryosphere 11, 2491–2506 (2017). dominant modes of surface type variation along the transect and assess which 13. Moustafa, S. E. et al. Multi-modal albedo distributions in the ablation area of surface type, or set of surface types, best correlate with mesoscale albedo variability, the southwestern Greenland Ice Sheet. Cryosphere 9,1–11 (2015). as represented by MOD10A1. Linear regression was used to yield the correlation 14. Ryan, J. C. et al. Attribution of Greenland’s ablating ice surfaces on ice coefficients between the principle component scores and MOD10A1 albedo. sheet albedo using unmanned aerial systems. The Cryosphere Discuss.1–23. https://doi.org/10.5194/tc-2016-204 (2016). Estimating albedo of digital image pixels. An estimate for the albedo (α) of each 15. Ryan, J. C. et al. Derivation of high spatial resolution albedo from UAV digital surface type was obtained from the UAV digital imagery. An explicit description of imagery: application over the Greenland Ice Sheet. Front. Earth Sci. this method can be found in Ryan et al. , but we briefly summarize it here. Firstly, 5, 40 (2017). the RAW digital numbers of each proprietary Sony RAW image were preserved by 16. Colgan, W. et al. An increase in crevasse extent, West Greenland: Hydrologic converting to a 16-bit TIFF image using dcraw (http://cybercom.net/~dcoffin/ implications. Geophys. Res. Lett. 38, L18502 (2011). dcraw/). A vignette correction mask was universally applied to compensate for 17. Colgan, W. et al. Glacier crevasses: observations, models and mass balance image and lens distortion due to edge effects which were as high as 17.6% at the implications. Rev. Geophys. 54, 119–161 (2016). corners of some images. The correction mask was calculated from the mean 8 NATURE COMMUNICATIONS (2018) 9:1065 DOI: 10.1038/s41467-018-03353-2 www.nature.com/naturecommunications | | | NATURE COMMUNICATIONS | DOI: 10.1038/s41467-018-03353-2 ARTICLE 18. Greuell, W. Melt-water accumulation on the surface of the Greenland Ice 47. Scheick, J. Investigating Greenland Crevasses at Raven Camp. Field Notes, Sheet: effect on albedo and mass balance. Geogr. Ann. A Phys. Geogr. 82, Polar Field Services Newsletter, http://www.polarfield.com/blog/blog/ 489–498 (2000). investigating-greenland-crevasses-raven-camp. Accessed 25 May 2014 (2012). 19. Smith, L. C. et al. Efficient meltwater drainage through supraglacial streams 48. Poinar, K. et al. Limits to future expansion of surface-melt-enhanced ice flow and rivers on the southwest Greenland ice sheet. Proc. Natl Acad. Sci. USA into the interior of western Greenland. Geophys. Res. Lett. 42, 1800–1807 112, 1001–1006 (2015). (2015). 20. Bøggild, C. E., Brandt, R. E., Brown, K. J. & Warren, S. G. The ablation zone in 49. Hodson, A. et al. A glacier respires: quantifying the distribution and northeast Greenland: ice types, albedos and impurities. J. Glaciol. 56, 101–113 respiration CO flux of cryoconite across an entire Arctic supraglacial (2010). ecosystem. J. Geophys. Res. 112, G04S36 (2007). 21. Chandler, D. M., Alcock, J. D., Wadham, J. L., Mackie, S. L. & Telling, J. 50. Wiscombe, W. J. & Warren, S. G. A model for the spectral albedo of snow. I: Seasonal changes of ice surface characteristics and productivity in the ablation Pure snow. J. Atmos. Sci. 37, 2712–2733 (1980). zone of the Greenland Ice Sheet. Cryosphere 9, 487–504 (2015). 51. Zuo, Z. & Oerlemans, J. Modelling albedo and specific balance of the 22. Stibal, M. et al. Algae drive enhanced darkening of bare ice on the Greenland Greenland ice sheet: calculations for the Sondre Strømfjord transect. J. Glaciol. ice sheet. Geophys. Res. Lett. 44, 11463–11471 (2017). 42, 305–316 (1996). 23. Stibal, M. et al. Organic matter content and quality in supraglacial debris 52. Hock, R. Glacier melt: a review of processes and their modelling. Prog. Phys. across the ablation zone of the Greenland ice sheet. Ann. Glaciol. 51,1–8 Geogr. 29, 362–391 (2005). (2010). 53. Alexander, P. M. et al. Assessing spatio-temporal variability and trends in 24. Yallop, M. L. et al. Photophysiology and albedo-changing potential of the ice modelled and measured Greenland Ice Sheet albedo (2000-2013). Cryosphere algal community on the surface of the Greenland ice sheet. ISME J. 6, 8, 2293–2312 (2014). 2302–2313 (2012). 54. Tedesco, M. et al. The role of albedo and accumulation in the 2010 melting 25. Dumont, M. et al. Contribution of light-absorbing impurities in snow to record in Greenland. Environ. Res. Lett. 6, 014005 (2011). Greenland’s darkening since 2009. Nat. Geosci. 7, 509–512 (2014). 55. Box, J. E. et al. Greenland ice sheet albedo feedback: thermodynamics and 26. Keegan, K. M., Albert, M. R., McConnell, J. R. & Baker, I. Climate change and atmospheric drivers. Cryosphere 6, 821–839 (2012). forest fires synergistically drive widespread melt events of the Greenland Ice 56. van Angelen, J. H. et al. Sensitivity of Greenland Ice Sheet surface mass Sheet. Proc. Natl Acad. Sci. USA 111, 7964–7967 (2014). balance to surface albedo parameterization: a study with a regional climate 27. Wientjes, I. G. M., Van de Wal, R. S. W., Reichart, G. J., Sluijs, A. & model. Cryosphere 6, 1175–1186 (2012). Oerlemans, J. Dust from the dark region in the western ablation zone of the 57. Langen, P. L. et al. Liquid water flow and retention on the Greenland Ice Sheet Greenland ice sheet. Cryosphere 5, 589–601 (2011). in the regional climate model HIRHAM5: local and large-scale impacts. Front. 28. Pfeffer, W. T. & Bretherton, C. S. The effect of crevasses on the solar heating Earth Sci. 4, 110 (2017). of a glacier surface. In Proc. of the Vancouver Symposium, August 1987, IAHS 58. Hall, D. K. & Riggs, G. A. MODIS/Terra Snow Cover Daily L3 Global 500m Publ. no. 170, 191–205 (IAHS Publications, Wallingford, UK, 1987). Grid, Version 6 (NASA National Snow and Ice Data Center Distributed Active 29. Cathles, L. M., Abbot, D. S., Bassis, J. N. & MacAyeal, D. R. Modeling surface- Archive Center, Boulder, 2016). roughness/solar-ablation feedback: application to small-scale surface 59. Stroeve, J. C., Box, J. E. & Haran, T. Evaluation of the MODIS (MOD10A1) channels and crevasses of the Greenland Ice Sheet. Ann. Glaciol. 52,99–108 daily snow albedo product over the Greenland ice sheet. Remote Sens. Environ. (2011). 105, 155–171 (2006). 30. Langford, H., Hodson, A., Banwart, S. & Bøggild, C. The microstructure 60. van As, D. et al. Large surface meltwater discharge from the Kangerlussuaq and biogeochemistry of Arctic cryoconite granules. Ann. Glaciol. 51,87–94 sector of the Greenland ice sheet during the record-warm year 2010 (2010). explained by detailed energy balance observations. Cryosphere 6, 199–209 31. Cook, J., Edwards, A., Takeuchi, N. & Irvine-Fynn, T. Cryoconite: the dark (2012). biological secret of the cryosphere. Prog. Phys. Geogr. 40,66–111 (2015). 61. Ryan, J. C. et al. UAV photogrammetry and structure from motion to assess 32. Takeuchi, N. Optical characteristics of cryoconite (surface dust) on glaciers: calving dynamics at Store Glacier, a large outlet draining the Greenland ice the relationships between light absorbency and the property of organic matter sheet. Cryosphere 9,1–11 (2015). contained in the cryoconite. Ann. Glaciol. 34, 409–414 (2002). 62. Ryan, J. C. et al. How robust are in situ observations for validating satellite- 33. Remias, D. et al. Characterization of an UV- and VIS-absorbing, derived albedo over the dark zone of the Greenland Ice Sheet? Geophys. Res. purpurogallin-derived secondary pigment new to algae and highly abundant Lett. 44, 6218–6225 (2017). in Mesotaenium berggrenii (Zygnematophyceae, Chlorophyta), an 63. Howat, I. M., Negrete, A. & Smith, B. E. The Greenland Ice Mapping Project extremophyte living on glaciers. FEMS Microbiol. Ecol. 79, 638–648 (2012). (GIMP) land classification and surface elevation data sets. Cryosphere 8, 34. Stibal, M., Šabacká, M. & Žárský, J. Biological processes on glacier and ice 1509–1518 (2014). sheet surfaces. Nat. Geosci. 5, 771–774 (2012). 64. Pedregosa, F., Weiss, R. & Brucher, M. Scikit-learn: machine learning in 35. Stibal, M. et al. Microbial abundance in surface ice on the Greenland Ice Sheet. Python. J. Mach. Learn. Res. 12, 2825–2830 (2011). Front. Microbiol. 6, 225 (2015). 65. Canny, J. A computational approach to edge detection. IEEE Trans. Pattern 36. Lutz, S. et al. The biogeography of red snow microbiomes and their role in Anal. Mach. Intell. 8, 679–698 (1986). melting arctic glaciers. Nat. Commun. 7,1–9 (2016). 66. Knap, W. H. & Reijmer, C. H. Anisotropy of the reflected radiation field over 37. Ryser, C. et al. Cold ice in the ablation zone: Its relation to glacier hydrology melting glacier ice: measurements in Landsat TM bands 2 and 4. Remote Sens. and ice water content. J. Geophys. Res. Earth Surf. 118, 693–705 (2013). Environ. 65,93–104 (1998). 38. Lüthje, M., Pedersen, L. T., Reeh, N. & Greuell, W. Modelling the evolution of supraglacial lakes on the west Greenland ice-sheet margin. J. Glaciol. 52, 608–618 (2006). Acknowledgements 39. Tedesco, M. & Steiner, N. In-situ multispectral and bathymetric J.C.R. was funded by an Aberystwyth University Doctoral Career Development Scho- measurements over a supraglacial lake in western Greenland using a remotely larship (DCDS) and is currently funded by the NASA Cryosphere Program grant controlled watercraft. Cryosphere 5, 445–452 (2011). NNX14AH93G. The field camp, logistics and J.B. were supported by the Dark Snow 40. Fitzpatrick, A. A. W. et al. A decade (2002-2012) of supraglacial lake volume Project (http://www.darksnow.org/). We also gratefully acknowledge the UK Natural estimates across a land-terminating margin of the Greenland Ice Sheet. Environment Research Grants NE/H024204/1 and NE/G005796/1, the Aberystwyth The Cryosphere Dicuss. 7, 1383–1414 (2014). University Research Fund and the Danish Villum Young Investigator Programme grant 41. Leeson, A. et al. Supraglacial lakes on the Greenland ice sheet advance inland VKR 023121 awarded to M.S. A.H. acknowledges support from the Centre for Arctic Gas under warming climate. Nat. Clim. Chang. 5,51–55 (2015). Hydrate, Environment and Climate, funded by the Research Council of Norway through 42. Hubbard, A. L., Blatter, H., Nienow, P., Mair, D. & Hubbard, B. Comparison its Centres of Excellence (grant 223259). J.C. acknowledges a Rolex Award for Enterprise of a three-dimensional model for glacier flow with field data from Haut and L.C.S. acknowledges the NASA Cryosphere Program grant NNX14AH93G. The Glacier d’ Arolla, Switzerland. J. Glaciol. 44, 368–378 (1998). automatic weather stations used to validate the MODIS albedo data were funded and 43. van der Veen, C. J. Crevasses on glaciers. Polar Geogr. 23,1–33 (1999). installed by the Greenland Analogue Project (GAP) and maintained by the Geological 44. van de Wal, R. S. W. et al. Large and rapid melt-induced velocity changes in Survey of Denmark and Greenland (GEUS). We thank Peter Sinclair and Gabriel Warren the ablation zone of the Greenland Ice Sheet. Science 321, 111–113 (2008). for their assistance in the field and our three reviewers. 45. Doyle, S. H. et al. Persistent flow acceleration within the interior of the Greenland ice sheet. Geophys. Res. Lett. 41, 899–905 (2014). 46. Christoffersen, P., Bougamont, M., Hubbard, A., Doyle, S. & Grigsby, S. Author contributions Tensile shock triggers cascading lake drainage on the Greenland Ice Sheet. J.C.R. and A.H. designed, tested and built the UAVs. A.H. and J.B. were the project PIs Nat. Commun. https://doi.org/10.1038/s41467-018-03420-8 (2018). and J.B. provided UAV onboard pyranometers and data logger. J.C.R. and J.B. NATURE COMMUNICATIONS (2018) 9:1065 DOI: 10.1038/s41467-018-03353-2 www.nature.com/naturecommunications 9 | | | ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/s41467-018-03353-2 programmed and deployed the UAV. J.C.R. analysed the data and wrote the original Open Access This article is licensed under a Creative Commons manuscript with A.H. who also provided direction/supervision. K.C. provided the Attribution 4.0 International License, which permits use, sharing, ground photographs used for qualitative assessment of surface types. A.H., M.S., J.B., J.C., adaptation, distribution and reproduction in any medium or format, as long as you give T.D.I.-F., L.C.S. and K.C. provided conceptual and technical advice and contributed to appropriate credit to the original author(s) and the source, provide a link to the Creative data interpretation. All authors commented and critically revised the manuscript. Commons license, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the Additional information article’s Creative Commons license and your intended use is not permitted by statutory Competing interests: The authors declare no competing interests. regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this license, visit http://creativecommons.org/ Reprints and permission information is available online at http://npg.nature.com/ licenses/by/4.0/. reprintsandpermissions/ Publisher's note: Springer Nature remains neutral with regard to jurisdictional claims in © The Author(s) 2018 published maps and institutional affiliations. 10 NATURE COMMUNICATIONS (2018) 9:1065 DOI: 10.1038/s41467-018-03353-2 www.nature.com/naturecommunications | | |

Journal

Nature CommunicationsSpringer Journals

Published: Mar 14, 2018

There are no references for this article.