16_JEB_cet.indd 53 Albedo, Latin for ‘whiteness’, is a term used to describe the amount of sunlight reflected by the ground. Fresh snow al- bedo can exceed 85%, making it among the most reflective natural substances. Warm conditions promote snow crys- tal metamorphosis that, like the presence of liquid water, bring snow albedo down below 65%. With the darkening, caused by the metamorphosis, absorbed solar energy thus increases by roughly a factor of two. Seasonal snow melts over the lower reaches of a glacier leading to the exposure of bare ice with albedo below 55%. Impurities such as dust, black carbon or microbes can bring glacier-ice albedo be- low 30%, meaning that snow ablation gives way to impu- rity-rich, bare glacier ice which increases absorbed sunlight by more than a factor of three. The thickness of the winter snow layer and the intensity of spring melt are important determinants of the annual glacier-ice melt, as the amount of snow cover governs the timing of darker ice exposure; the earlier the exposure, the more ice can melt. Because snow and ice albedo proper- ties make it an amplifier of climate change, surface albedo has been designated as an Essential Climate Variable and a Target Requirement for climate monitoring (WMO 2011). Polar orbiting satellites facilitate albedo mapping with Arctic coverage multiple times per day in clear-sky con- ditions. Satellite-based retrievals of surface albedo depend on accurate compensation of the intervening atmosphere. Thus, without ground truth, the satellite retrievals are un- certain. In Greenland, snow and ice albedo is monitored by automatic weather stations (AWSs) from The Greenland Climate Network (GC-Net; Steffen et al. 1996) since 1995 and after 2007 from The Programme for Monitoring of the Greenland Ice Sheet (PROMICE; van As et al. 2013). Using the GC-Net data, satellite-derived albedo values are compared with ground data (e.g. Stroeve et al. 2013). Here, we present comparisons of daily GC-Net and PROMICE albedo data to satellite-derived albedo from the NASA Moderate Resolution Imaging Spectroradio- meter (MODIS) MOD10A1 product (Hall et al. 1995). MOD10A1 data have been available since May 2000 and are here de-noised, gap-filled and calibrated into a daily 500 × 500 m grid covering Greenland, Iceland and the Canadian Arctic glaciers (Fig. 1). Daily albedo from MODIS The MOD10A1 product contains daily snow extent, snow albedo, fractional snow cover and a data quality assessment at 500 × 500 m resolution (Hall et al. 1995). Both NASA Terra and Aqua satellites are equipped with MODIS sen- sors. Here, Terra data are chosen over Aqua data as they Greenland, Canadian and Icelandic land-ice albedo grids (2000–2016) Jason E. Box, Dirk van As, Konrad Steffen and the PROMICE project team* * Robert S. Fausto, Andreas P. Ahlstrøm, Michele Citterio and Signe B. Andersen Fig. 1. An example (15 July 2016) of the daily 500 m × 500 m enhanced MOD10A1 Collection 6 albedo product presented here. The circles show positions of ground-truth observations. 84 80 76 72 68 64 60 56 52 48 44 40 36 32 28 Albedo (%) PROMICE GC-Net KCP JAR © 2017 GEUS. Geological Survey of Denmark and Greenland Bulletin 38, 53–56. Open access: www.geus.dk/publications/bull 5454 give longer temporal coverage, and Aqua MODIS band 6 detectors (useful in cloud discrimination) have become degraded or non-functional (MODIS Characterization Support Team, NASA, updated February 2017). An alter- native MODIS albedo product (MCD43) was not chosen due to its reduced temporal resolution of eight days. The MOD10A1 data used here span the Arctic melt season 15 March (74th day of year) to 26 October (299th day of year) for the 17 year period 2000 to 2016. The two latest MOD10A1 versions are evaluated; Collection 5 (Hall et al. 2011) and Collection 6 (Riggs & Hall 2015; Hall & Riggs 2016), hereafter C5 and C6. MOD10A1 de-noising, smoothing and gap-filling Inspection of the C5 and C6 albedo imagery reveals that, despite some cases when pixel quality is coded ‘best’ or ‘good’, cloud artifacts resembling shadows, aircraft con- densation trails, thin clouds, and cloud edges can persist, often over the brightest areas presumably where there is less distinction between clouds and clean snow. Fortunately, because the artifacts introduce abrupt tem- poral departures in the albedo (α) time series, it is possible to reject them on a pixel by pixel basis using temporal sta- tistics from multi-day albedo (αN-days) samples. Here, an 11 day αN-days sample size is selected; five days before and after each day i. On a pixel-by-pixel basis, statistics are computed from αN-days. The number of days N does not always repre- sent 11 albedo values because some days a pixel may already be dismissed as cloudy, missing or of inadequate quality. Only cases with at least four samples per 11-day window are considered sufficient for an albedo estimate for that day and pixel. The final pixel by pixel daily albedo values are taken as the 11-day average of available values below a fractional noise threshold (D) value of 0.4, with D computed as: Di = | (αN-days - median(αN-days)) / median(αN-days) | For low albedo variability areas, for example the dry snow area, when the standard deviation αN-days is under 0.03, then a more strict D threshold of 0.1 is used. The procedure has both a smoothing and a gap filling effect on the albedo time series. The resulting data product can be viewed at https://tinyurl.com/PROMICE-albedo-Greenland. MODIS validation using ground-truth albedo Daily PROMICE and GC-Net AWS albedo values are com- pared with the nearest 500 × 500 m resolution MOD10A1 C5 and C6 values for all stations in each of 9 or 16 years that span 2007 to 2016 or 2000 to 2015, respectively. Fig- ure 2 provides an example for C6 data illustrating a typical result of the de-noising procedure, yielding an increased number of MODIS values, increased correlation, reduced root mean squared difference (RMSD) shifting regression slope closer to unity and no real change in bias. Table 1 lists summary statistics for the multi-year, multi- station comparison with MOD10A1 C5 and C6. The MOD10A1 skill either improves or is stable in the C5 to C6 update. The average bias and root mean squared differ- ence decrease and the correlation and average count of days increase. The number of compared station-years increases. From raw to de-noised, there is also a consistent improve- ment in agreement between the satellite and ground data (Table 2). In the de-noised product, the RMSDs are 0.08 for PROMICE stations that are concentrated in the ablation Day of year Al be do (% ) 100 150 200 250 90 80 70 60 50 40 de-noised MOD10A1 PROMICE N = 117 Regression: Correlation = 0.962 PROMICE = MODIS*0.872 + 0.090 bias = 0.008 RMSD = 0.043 PR O M IC E alb ed o (% ) MOD10A albedo (%) 40 50 60 70 80 90 90 80 70 60 50 40 Fig. 2. Year 2013 example comparison of daily de-noised albedo from satellite (NASA MODIS MOD10A1 Collection 6 data) and the ground (PROMICE) for the KPC_L (Fig. 1) station on the north-eastern Greenland ice sheet. 55 area or 0.05 for GC-Net stations that are concentrated in the accumulation area. e lower GC-Net correlation and RMSD values result from the mostly dry snow areas where albedo variability is small. For PROMICE stations which are concentrated in the ablation area (and for the GC-Net Jakobshavn Ablation Region (JAR) stations), the larger MOD10A1 pixel footprint includes a complex contribution from some combination of e.g., crevasses, snow patches and concentrated or distributed snow and ice impurities such as cryoconite. e root mean squared di erence is probably more attributable to the ground data because they have a c. 5–10 m2 footprint, four orders of magnitude smaller than the MOD10A1 500 m × 500 m footprint. MODIS validation using GC-Net albedo GC-Net albedo data, having a time coverage longer than 10 years, are compared with C5 and C6 to evaluate accu- racy in year-to-year albedo changes. MOD10A1 Collection 6 (Hall & Riggs 2016) compensates MODIS sensor degra- dation found in Collection 5 (Lyapustin et al. 2014). C6 is found to compensate the temporal trend bias in dry snow areas (Fig. 3A). e trend bias is usually smaller or non- existent for darker targets such as the ablation area (Fig. 3B). e 0.02 albedo o set at the Summit site is partly attributable to the bias described in the following. MOD10A1 sun angle bias Whereas the adjustments to Collection 6 eliminate a spu- rious darkening trend concentrated over snow and in the northern part of Greenland (Polashenski et al. 2015), both Collections 5 and 6 MOD10A1 albedo products have a residual bias based on the angle of the sun above the hori- zon. e bias is evident over nearly 20° of latitude range of the PROMICE and GC-Net data. In April (days 91–120), there is no bias in southern Greenland but a c. 4% bright bias in the northern 2/3 of Greenland (Fig. 4A). By June (days 152–181), the pattern of the bias has shifted to a more uniform dark bias strongest in the south (Fig. 4B). e bias varies over time and latitude (see the animation: https://ti nyurl.com/bias-vs-lat). We correct the Collection 6 bias ac- cording to the daily variation in the regression line (blue dashed line in Fig. 4A, trends in Fig. 4B). e calibration Table 1. Summary statistics for comparison of αMOD10A1 Collection 5 and 6 with available αPROMICE and αGC-Net in the 2000–2016 period (de-noised). Collection 5 Correlation 0.838 0.182 Bias –0.024 0.035 RMSD * 0.086 0.042 A verage count of days 117 37 Collection 6 Correlation 0.832 0.183 Bias –0.002 0.034 RMSD 0.084 0.044 A verage count of days 124 39 Collection 5 Correlation 0.490 0.225 Bias -0.006 0.099 RMSD 0.104 0.090 A verage count of days 114 33 Collection 6 Correlation 0.581 0.259 Bias 0.006 0.038 RMSD 0.050 0.033 A verage count of days 110 32 αMOD10A1 versus αPROMICE 144 station years Average St. dev. αMOD10A1 versus αGC-Net 183 station-years Average St. dev. * Root mean squared difference. Table 2. Summary statistics for comparison of αMOD10A1 Collection 5 and 6 with available αPROMICE and αGC-Net in the 2000–2016 period (de-noised minus raw). Collection 5 Correlation 0.067 0.003 Bias –0.001 0.001 RMSD * –0.022 0.003 Count of days per year 28 13 Collection 6 Correlation 0.070 0.032 Bias –0.002 0.001 RMSD –0.029 0.003 Count of days per year 28 13 Collection 5 Correlation 0.111 –0.048 Bias 0.002 0.026 RMSD 0.016 0.031 Count of days per year 32.000 11.000 Collection 6 correlation 0.170 –0.005 bias –0.005 –0.042 RMSD –0.030 –0.037 Count of days per year 29 12 αMOD10A1 versus αPROMICE 4 station years Average St. dev. αMOD10A1 versus αGC-Net 19 station-years Average St. dev. * Root mean squared difference. Year Year Al be do (% ) Al be do (% ) 2000 2005 2010 2015 2000 2005 2010 2015 Summit (72.5°N, 38.5°W, 3254 m a.s.l.) GC-Net trend: –1.0% MOD10A1 Collection 5 trend: –5.4% MOD10A1 Collection 6 trend: –1.2% 84 83 82 81 80 79 70 60 50 40 30 20 JAR1 (69.4°N, 49.6°W, 962 m a.s.l.) GC-Net trend: –27.0% MOD10A1 Collection 5 trend: –27.0% MOD10A1 Collection 6 trend: –22.9% BA Fig. 3. Examples of July monthly average dry snow area (A) and ablation area (B) MODIS col- lection 5 and 6 compared with GC-Net albedo trends spanning 16 summers. JAR: Jakobshavn Ablation Region. nr38_p53-56.pdf 1 09/08/17 09.18 nr38_p53-56.pdf 1 09/08/17 09.36 https://tinyurl.com/bias-vs-lat 5656 assumes there is no sun-angle-dependent bias in the PRO- MICE and GC-Net data. MODIS albedo for Iceland and the Canadian Arctic The regional product also includes albedo for glaciated ar- eas in Iceland, Jan Mayen and the Canadian Arctic. The occurrence of clouds reduces the coverage of the product, in the case of land ice, especially at the lowest elevations often near oceans. Iceland has data coverage less than half of that of most areas of Greenland. Areas of the Canadian Arctic such as the Devon Ice Cap also have reduced time coverage compared to Greenland, which has a stronger cloud clearing effect from its high pressure areas often cen- tered over the ice sheet. A new PROMICE data product The methodology developed here for de-noising, gap fill- ing, and bias correction for the MOD10A1 albedo product yields an enhanced MODIS MOD10A1 Climate Data Re- cord available for download through the PROMICE data- base via the webpage www.promice.dk Acknowledgements PROMICE is supported by the Danish Energy Agency through the DANCEA programme. GC-Net is supported by NASA and US Na- tional Science Foundation grants. References Hall, D.K. & Riggs, G.A. 2016: MODIS/Terra Snow Cover Daily L3 Global 500m Grid, Version 6. Greenland coverage. National Snow and Ice Data Center, NASA Distributed Active Archive Center, Boulder, Colorado USA. http://nsidc.org/data/MOD10A1/versions/6, accessed December 2016. Hall, D.K., Riggs, G.A. & Salomonson, V.V. 1995: Development of methods for mapping global snow cover using moderate resolution imaging spectroradiometer data. Remote Sensing of Environment 54, 127–140. Hall, D.K., Riggs, G.A. & Salomonson, V.V. 2011: MODIS/Terra Snow Cover Daily L3 Global 500 m Grid V004, January to March 2003. Digital media, updated daily. National Snow and Ice Data Center, Boulder, CO, USA. Lyapustin, A. et al. 2014: Science impact of MODIS C5 calibration deg- radation and C6+ improvements. Atmospheric Measurement Tech- niques Discussion 7, 7281–7319. MODIS Characterization Support Team, NASA, updated February 2017: http://mcst.gsfc.nasa.gov/calibration/time-dependent-list-non- functional-or-noisy-detector Polashenski, C.M., Dibb, J.E., Flanner, M.G., Chen, J.Y., Courville, Z.R., Lai, A.M., Schauer, J.J., Shafer, M.M. & Bergin, M. 2015: Neither dust nor black carbon causing apparent albedo decline in Greenland’s dry snow zone: implications for MODIS C5 surface re- flectance, Geophysical Research Letters 42, 9319–9327. Riggs, G.A. & Hall, D.K. 2015: MODIS Snow products collection 6, User Guide. https://nsidc.org/sites/nsidc.org/files/files/MODIS-snow- user-guide-C6.pdf Steffen, K., Box, J.E. & Abdalati, W. 1996: Greenland Climate Net- work: GC-Net. In: Colbeck, S.C. (ed.): CRREL 96-27 Special Re- port on glaciers, ice sheets and volcanoes, tribute to M. Meier, 98– 103. Hannover: U.S. Army. Stroeve, J.C., Box, J.E., Wang, Z., Schaaf, C. & Barrett, A. 2013: Re- evaluation of MODIS MCD43 Greenland albedo accuracy and trends. Remote Sensing of Environment 138, 199–214. Van As, D., Fausto, R.S., Colgan, W.T., Box, J.E. and the PROMICE project team 2013: Darkening of the Greenland ice sheet due to the melt-albedo feedback observed at PROMICE weather stations. Geo- logical Survey of Denmark and Greenland Bulletin 28, 69–72. WMO (World Meteorological Organization) 2011: Systematic observa- tion requirements for satellite-based data products for climate, Up- date. Global Climate Observing System, GCOS-154, 138 pp. Latitude (°N) Day of year Al be do b ias (u ni tle ss ) 65 70 75 80 100 120 140 160 180 200 220 240 PROMICE GC-Net 0.06 0.04 0.02 0.00 –0.20 –0.04 –0.06 0.08 0.06 0.04 0.02 0.00 –0.20 –0.04 Al be do b ias (u ni tle ss ) 60°N 70°N 80°N A B Fig. 4. A: Example of NASA MODIS MOD10A1 Collection 6 difference with ground data from automatic weather stations versus latitude, indicating a sun-angle-dependent bias. B: Bias for three latitude bands. The station names are abbreviated. Authors’ addresses J.E.B., D.V.A. & the PROMICE team, Geological Survey of Denmark and Greenland, Øster Voldgade 10, DK-1350 Copenhagen K, Denmark; E-mail: jeb@geus.dk K.S., Swiss Federal Institute for Forest, Snow and Landscape Research, Zürcherstrasse 111, CH-8903 Birmensdorf, Switzerland. denoised MOD10A1 PROMICE: N 117 Regression Correlation 0962 PROMICE MODIS0872 0090 bias 0008 RMSD 0043: 60N 70N 80N B: undefined: undefined_2: Authors addresses: