71 The Arctic region is warming more rapidly than the global average (AMAP 2017) and it is well established that this warming is at least partially responsible for the Greenland ice sheet losing mass at an accelerating rate, raising concern worldwide (e.g. Kahn et al. 2015; Rahmstorf et al. 2015). It is essential to monitor the changes of the Greenland ice sheet to be able to assess the potential environmental, social and economic implications around the globe, and to provide decision-makers with reliable data. The annual mass-budget deficit of the Greenland ice sheet has grown over the past two decades due to increases in surface melting (Van den Broeke et al. 2017) and ice-f low acceleration (Kahn et al. 2015). Currently, and for the last two decades, the Greenland ice sheet is the single largest Arctic cryospheric contributor to global sea-level rise and the Greenland ice-surface melt rates are projected to increase as the Arctic continues to warm (AMAP 2017). The snowline is here defined as the maximum elevation during the melt season at which snow remains from the pre- vious accumulation season (Cogley et al. 2011). The snow- line is a valuable climate indicator as its position integrates the competing effects of melt (increasing snowline elevation) and snow accumulation (decreasing snowline elevation). Thus the snowline provides a key holistic variable indicating climate change. We have developed a methodology that determines snow- line elevation utilising the moderate resolution imaging spec- troradiometer (MODIS) sensor on the Terra satellite. The MODIS sensor produces a global dataset on a daily basis, with a resolution varying between 250 m and 1 km, in 36 bands covering the visible to thermal wavelengths. Using MODIS, we derived the maximum snowline altitude for the Greenland ice sheet for the years 2000–2017. We are pro- ducing a freely available, consistent dataset that provides an important tool for the monitoring of the long-term impact of climate change on the Greenland ice sheet. Direct compari- son with field observations from automatic weather stations (AWSs) from the Programme for Monitoring of the Green- The Greenland ice sheet – snowline elevations at the end of the melt seasons from 2000 to 2017 Robert S. Fausto and the PROMICE team* * Signe B. Andersen, Andreas P. Ahlstrøm, Dirk van As, Jason E. Box, Daniel Binder, Michele Citterio, William Colgan, Konstanze Haubner, Karina Hansen, Nanna B. Karlsson, Kenneth D. Mankoff, Allan Ø. Pedersen, Anne Solgaard and Baptiste Vandecrux 500 km THU UPE KPC TAS KAN NUK SCO QAS EGP Land 185.3 370.6 555.9 741.2 926.5 1112 1297 1482 1668 1853 2038 2224 2409 2594 2779 2965 3150 Snowline 2016 PROMICE GIMP DEM Fig. 1. Greenland map showing the location of PROMICE automatic weather stations and the 2016 snowline as derived from Terra satellite data using the moderate resolution imaging spectro- radiometer (MODIS) sensor. The locations of the ground-control automatic weather stations (PROMICE) are indicated. EGP: East GRIP. K AN: Kangerlussuaq. KPC. Kronprins Christian Land. NUK: Nuuk. QAS: Qassimiut. SCO: Scoresby Sund. TAS: Tasiilaq. THU: Thule. UPE: Upernavik. DEM: Digital elevation model. GIMP: Greenland ice mapping project. © 2018 GEUS. Geological Survey of Denmark and Greenland Bulletin 41, 71–74. Open access: www.geus.dk/bulletin http://www.geus.dk/bulletin 7272 land Ice Sheet (PROMICE) network validates the snowline dataset derived from MODIS. We use the services of the CryoClim internet portal, providing an operational and per- manent service for long-term systematic climate monitoring of the cryosphere, to distribute our snowline product. More specifically, end-of-melt season, 1 km2 resolution raster grids illustrating snow and bare-ice surfaces, and snowline shape files can be downloaded via CryoClim. Here, we describe the snowline classification algorithm, its validation and its inter- annual variations for 18 years spanning 2000–2017. Snowline classification algorithm We processed all MODIS MOD12KM and MOD03 scenes covering Greenland from late July to the beginning of Sep- tember 2000–2017. We used the surface-type detection algo- rithm of Fausto et al. (2015) that distinguishes between bare- ice and snow surfaces. Fausto et al. (2015) uses normalised thresholds (Th) from calibrated radiances (MOD021KM) between the near-infrared band 5 (1230–1250 nm) and the visible band 10 (483–493 nm) with surface-type thresholds Th dry snow ≤0.86, 0.86265–2.1×LAT, where LAT is latitude. Th bare ice is defined as: T h b a r e i c e = c 0 + c 1 × b 1+ c 2 × b 2 + c 3 × b 3 + c 5 × b 5 + c 7 × b 7 where c0=–0.0015, c1=0.160, c2=0.291, c3=0.243, c5= 0.112, c7=0.081 and b1 to b 7 designate band 1 to band 7. Cloud-covered regions are removed using the MOD35_L2 dataset. Subsequently pixels are classified for every MODIS scene as either snow or bare ice for the whole Greenland ice sheet. Daily classification scenes are aggregated to yield a maxi- mum extent of bare ice to define an end-of-melt-season snowline. Snowlines from peripheral glaciers are generally excluded, and the snowline products are based on an algo- rithm success rate of over 95% classified pixels. Validation To help validate the MODIS data we make use of the PROM- ICE automatic weather station network that currently con- sists of two or three stations primarily in the ablation area in eight ice sheet regions. Each automatic weather station records a suite of meteorological and glaciological measure- ments, supplemented by e.g. surface-height changes due to accumulation or ablation (Fig. 1; Van As et al. 2016). To validate the classified snowline elevation at the end of the melt season, we use the mass-budget values from the PROMICE weather stations (Fig. 2; e.g. Fausto et al. 2012) at different elevations to calculate the vertical surface mass-bal- ance gradient for all eight PROMICE transects to determine the equilibrium line altitude (ELA, zero mass budget), for di- rect comparison with MODIS estimated snowline elevation (Fig. 1). AWS balance profiles from the Upernavik region, and those indicating an ELA above 2000 m are excluded as we find them unrealistic. The location of the upper AWS should be close to the actual ELA to get the best balance pro- files. In total, we exclude 25% or 17 out of 67 balance profiles. Fig. 2. South Greenland PROMICE automatic weather station at the end of the 2013 melt season. From the stakes to the left of the weather station, the amount of melt (c. 4 m) is directly visible. The melt is also measured with a pres- sure transducer system drilled into the ice. 73 Figure 3 illustrates the performance of the MODIS end-of- melt-season snowline algorithm for all PROMICE regions in Greenland. The correlation (r=90%, p=0.0001, n=50) and the root-mean-square error (RMSE=200 m) are reasonable as the ELA and snowline elevation can be different due to superimposed ice formation (Cogley 2011). The mean differ- ence between snowline altitude and ELA is −104 m. Results and discussion Figure 1 illustrates the location of remotely sensed snowline plotted on top of the digital elevation model (DEM) from the Greenland Ice Mapping Project (GIMP, Howat et al. 2014). The snowline is easily visible in the southern, western, and northern parts of Greenland due to the relatively even terrain, while the snowline shows a more complicated pat- tern in the mountainous terrain in East Greenland (Fig. 1). The snowline separates bare ice from snow areas and can therefore be used to document the change in bare-ice areas. We find the extent of bare-ice exposure to be increasing in the period 2000–2017 at an average rate of c. 500 km2 per year (Fig. 4), which roughly corresponds to the size of the Danish island of Bornholm. This increase in the bare-ice area is insignificant, but it demonstrates a small average gain of melt over accumulation since 2000. The increasing trend in the bare-ice area is consistent with increasing Greenland mass loss due to surface processes (Van den Broeke et al. 2017). Both independent, in situ observations (Machguth et al. 2016) and remotely sensed observations (Hall et al. 2012; Tedesco et al. 2017) show that the Greenland melt area is ex- panding to higher elevations. Further, the increase in bare ice enhances the positive feedback mechanism of a darkening ice sheet surface (ice is darker than snow), which affects the sur- face mass and energy balance of the Greenland ice sheet (Box et al. 2012). Figure 4 also illustrates the inter-annual variabil- ity of the 2000–2017 snowlines, which is highly dependent on the complicated seasonal weather systems around Green- land. For instance, the below average snowline of the snowy year of 2016/2017 is consistent with positive albedo anoma- lies that reduced melting in 2017 (Tedesco et al. 2017). Uncertainties associated with the different surface-type detection are assessed with the ELAs derived from the AWS surface mass-budget observations. Figure 3 shows a signifi- cant correlation between the MODIS snowline and ELAs derived independently from PROMICE AWSs. A reason for the difference between the two can be that the MODIS data have a spatial resolution of 1 km2, pan-ice sheet coverage and quasi-daily temporal coverage, while the footprints of the in situ measurements are small (5–50 m2), and surface patchi- ness is clear in aerial photography (Stroeve et al. 2006). Fausto et al. (2015) discuss an August anomaly in their monthly surface-type data set during the 2010–2014 peri- od, illustrated by a noisy melting-snow classification in the northern ice sheet, which was most likely due to false clas- sification. However, with the updated bare-ice threshold, we improve the detection of snow and ice surfaces (Fig. 1), visualised by a less noisy snow classification of snow in the northern part of the ice sheet, resulting in a more reliable cli- mate indicator for Greenland. Conclusions Remotely sensed MODIS data can yield daily, automated classification of the Greenland ice sheet surface type (snow and ice). Validation indicates a high correlation (0.9) be- tween MODIS-derived snowline altitudes and ELAs esti- mated from in situ measurements. The end-of-melt-season 1000 1500 1700 1500 1300 1100 900 700 500 Automatic weather station ELA (m a.s.l.) KPC SCO TAS QAS NUK KAN UPE THUSn ow lin e el ev at io n (m a .s. l.) 500 2004 2009 2014 190 000 170 000 150 000 130 000 110 000 90 000 70 000 Year A re a (k m ) 1999 2 Bare-ice area y = 504x – 879029 Fig. 3. The end-of-melt-season snowline elevation for 2000 to 2017 from MODIS vs. the PROMICE AWS-derived equiblibrium line altitude (ELA). The blue line gives the 1:1 relation. The locations of the PRO- MICE automatic weather stations are shown in Fig. 1. Fig. 4. End-of-melt-season bare-ice area for the Greenland ice sheet for the years 2000–2017. 7474 snowline is useful as an ice-sheet climate indicator for the competing processes of surface accumulation and ablation, quantified by an average annual increase of c. 500 km2 of the bare-ice area for the 2000–2017 period. Acknowledgements The Programme for Monitoring of the Greenland Ice Sheet (PROMICE) is funded by the Geological Survey of Denmark and Greenland (GEUS) and the Danish Ministry of Energ y, Utilities and Climate under the Danish Cooperation for Environment in the Arctic (DANCEA), and is conducted in collaboration with DTU Space, Denmark’s National Space Institute, and Asiaq (Greenland Survey). The NUK and K AN stations were/are (co-)funded by the Greenland Climate Research Centre (GCRC) and the Greenland Analogue Project (GAP), respectively. This study was funded by DK ESA-PRODEX under the CryoClim project. References AMAP 2017: Snow, water, ice and permafrost in the Arctic (SWIPA) 2017, 269 pp. Oslo: Arctic Monitoring and Assessment Programme (AMAP). Box, J.E., Fettweis, X., Stroeve, J.C., Tedesco, M., Hall, D.K. & Steffen, K. 2012: Greenland ice sheet albedo feedback: thermodynamics and atmospheric drivers. The Cryosphere 6, 821–839. Cogley, J.G., Hock, R., Rasmussen, L.A., Arendt, A.A., Bauder, A., Braith- waite, R.J., Jansson, P., Kaser, G., Moller, M., Nicholson, L. & Zemp, M. 2011: Glossary of glacier mass balance and related terms. IHP-VII Technical Documents in Hydrolog y 86, IACS Contribution 2. Paris: UNESCO-IHP. Fausto, R.S., Van As, D. & PROMICE project team 2012: Ablation ob- servations for 2008–2011 from the Programme for Monitoring of the Greenland Ice Sheet (PROMICE). Geological Survey of Denmark and Greenland Bulletin 26, 25–28. Fausto, R.S. et al. 2015: Greenland ice sheet melt area from MODIS (2000–2014). Geological Survey of Denmark and Greenland Bulletin 33, 97–100. Hall, D.K., Comiso, J.C., DiGirolamo, N.E., Shuman, C.A., Key, J.R. & Koenig, L.S. 2012: A satellite-derived climate-quality data record of the clear-sky surface temperature of the Greenland ice sheet. Journal of Climate 25, 4785–4798. Howat, I.M., Negrete, A. & Smith, B.E. 2014: The Greenland Ice Map- ping Project (GIMP) land classification and surface elevation datasets. The Cryosphere 8, 1509–1518. http://dx.doi.org/10.5194/tc-8-1509- 2014 Khan, S.A., Aschwanden, A., Bjørk, A.A., Wahr, J., Kjeldsen, K.K. & Kjær, K.H. 2015: Greenland ice sheet mass balance: a review. Reports on Progress in Physics 78, 046801, 26 pp. http://dx.doi.org/10.1088/0034- 4885/78/4/046801 Machguth, H., MacFerrin, M., van As, D., Box, J.E., Charalampidis, C., Colgan, W., Fausto, R.S., Meijer, H.A.J., Mosley-Thompson, E. & van de Wal, R.S.W. 2016: Greenland meltwater storage in firn limited by near-surface ice formation. Nature Climate Change 6, 390–393. http:// dx.doi.org/10.1038/nclimate2899 Rahmstorf, S., Box, J., Feulner, G., Mann, M., Robinson, A., Rutherford, S. & Schaffernicht, E. 2015: Exceptional twentieth-Century slowdown in Atlantic Ocean overturning circulation. Nature Climate Change 5, 475–480. http://dx.doi.org/10.1038/nclimate2554 Stroeve, J.C., Box, J.E. & Haran, T. 2006: Evaluation of the MODIS (MOD10A1) daily snow albedo product over the Greenland ice sheet. Remote Sensing of Environment 105, 155–171. http://dx.doi. org/10.1016/j.rse.2006.06.009 Tedesco, M., Box, J.E., Cappelen, J., Fausto, R.S., Fettweis, X., Hansen, K., Mote, T., Sasgen, I., Smeets, C.J.P.P., van As, D., Velicogna, I. & van de Wal, R.S.W. 2017: Greenland ice sheet: contribution to the Arctic Report Card 2017. Van As, D., Fausto, R.S., Cappelen, J., Van de Wal, R.S.W., Braithwaite, R.J., Machguth, H. & PROMICE project team 2016: Placing Green- land ice sheet ablation measurements in a multi-decadal context. Geo- logical Survey of Denmark and Greenland Bulletin 35, 71–74. Van den Broeke, M.R., Box, J.E., Fettweis, X., Hanna, E., Noël, B., Te- desco, M., van As, D., van de Berg, W.J. & van Kampenhout, L. 2017: Greenland ice sheet surface mass loss: recent developments in obser- vation and modelling. Current Climate Change Reports 3, 345–356. http://dx.doi.org/10.1007/s40641-017-0084-8 Authors’ address Geological Survey of Denmark and Greenland, Øster Voldgade 10, DK-1350 Copenhagen K, Denmark. E-mail: rsf@geus.dk. http://dx.doi.org/10.5194/tc-8-1509-2014 http://dx.doi.org/10.5194/tc-8-1509-2014 http://dx.doi.org/10.1088/0034-4885/78/4/046801 http://dx.doi.org/10.1088/0034-4885/78/4/046801 http://dx.doi.org/10.1038/nclimate2899 http://dx.doi.org/10.1038/nclimate2899 http://dx.doi.org/10.1038/nclimate2554 http://dx.doi.org/10.1016/j.rse.2006.06.009 http://dx.doi.org/10.1016/j.rse.2006.06.009 http://dx.doi.org/10.1007/s40641-017-0084-8 mailto:rsf@geus.dk