PUBLICATIONSJournal of Geophysical Research: AtmospheresRESEARCH ARTICLE10.1002/2014JD021667Key Points: Land albedo climatologies were derivedfrom nine global satellite data sets Most satellite albedos can be used formodel calibration and validation Large differences were found at highlatitudes between satellite data setsSupporting Information: Readme Figure S1Correspondence to:T. He,[email protected]:He, T., S. Liang, and D.-X. Song (2014),Analysis of global land surface albedoclimatology and spatial-temporal variationduring 1981–2010 from multiple satelliteproducts, J. Geophys. Res. Atmos., 119,10,281–10,298, doi:10.1002/2014JD021667.Received 19 FEB 2014Accepted 16 AUG 2014Accepted article online 20 AUG 2014Published online 11 SEP 2014Analysis of global land surface albedo climatologyand spatial-temporal variation during 1981–2010from multiple satellite productsTao He1, Shunlin Liang1,2, and Dan-Xia Song11Department of Geographical Sciences, University of Maryland, College Park, Maryland, USA, 2State Key Laboratory ofRemote Sensing Science, College of Global Change and Earth System Science, Beijing Normal University, Beijing, ChinaAbstract For several decades, long-term time series data sets of multiple global land surface albedo productshave been generated from satellite observations. These data sets have been used as one of the key variables inclimate change studies. This study aims to assess the surface albedo climatology and to analyze long-termalbedo changes, from nine satellite-based data sets for the period 1981–2010, on a global basis. Results showthat climatological surface albedo data sets derived from satellite observations can be used to validate, calibrate,and further improve surface albedo simulations and parameterizations in current climate models. However,the albedo products derived from the International Satellite Cloud Climatology Project and the Global Energyand Water Exchanges Project have large seasonal biases. At latitudes higher than 50 , the maximal differencein winter zonal albedo ranges from 0.1 to 0.4 among the nine satellite data sets. Satellite-based albedo data setsagree relatively well during the summer at high latitudes, with a standard deviation of 0.04 for the 70 –80 zone in both hemispheres. The fine-resolution (0.05 ) data sets agree well with each other for all the land covertypes in middle to low latitudes; however, large spread was identified for their albedos at middle to highlatitudes over land covers with mixed snow and sparse vegetation. By analyzing the time series of satellite-basedalbedo products over the past three decades, albedo of the Northern Hemisphere was found to be decreasing inJuly, likely due to the shrinking snow cover. Meanwhile, albedo in January was found to be increasing, likelybecause of the expansion of snow cover in northern winter. However, to improve the albedo estimation athigh latitudes, and ultimately the climate models used for long-term climate change studies, a still betterunderstanding of differences between satellite-based albedo data sets is required.1. IntroductionSurface albedo, a variable defined as the ratio of the solar radiation reflected from Earth’s surface to the solarradiation incident upon it, is critical to the regulation of Earth’s surface energy budget [Liang et al., 2010,2013a]. Land surface albedo is highly variable, both spatially and temporally. Significant changes in surfacealbedo are accompanied by variations in land cover and surface conditions, such as snow [He et al., 2013],vegetation [Loarie et al., 2011; Lyons et al., 2008], urbanization [Offerle et al., 2005], and soil moisture [Govaertsand Lattanzio, 2008; Zhu et al., 2011]. B. Ghimire, et al. (Global albedo change and radiative cooling fromanthropogenic land-cover change, 1700 to 2005 based on MODIS, land-use harmonization, radiative kernels,and reanalysis, submitted to Geophysical Research Letters, 2014). estimated global surface albedo change,taking into account land cover change. They found that the global surface albedo increased 0.0012 during1700–2005, which had a net cooling effect of a top-of-atmosphere (TOA) radiative forcing of 0.23 W m 2.Aerosols like dust and soot may also contaminate snow and greatly reduce its albedo [Hansen and Nazarenko,2004; Xu et al., 2009]. Accurate surface albedo data are needed to better characterize climate systems and tohelp develop climate models with improved predictive power.Satellite data provide a unique opportunity for monitoring surface albedo on a global basis [Liang et al., 2012].Algorithms for albedo estimation have been developed for various different types of remote sensingsensors and platforms, from broadband sensors [e.g., Li and Garand, 1994; Pinty et al., 2000] to multispectral[e.g., Csiszar and Gutman, 1999; He et al., 2012; Schaaf et al., 2002; Shuai et al., 2014; Wang et al., 2013], multiangle[e.g., Maignan et al., 2004; Martonchik et al., 1998], and hyperspectral [He et al., 2014a] sensors. During thepast few decades, many long-term albedo products with global coverage have been derived from satellite data(Tables 1 and 2).HE ET AL. 2014. American Geophysical Union. All Rights Reserved.10,281

Journal of Geophysical Research: Atmospheres10.1002/2014JD021667Table 1. Global Satellite Albedo Products Used in This StudyAlbedo Data SetsGLASSGlobAlbedoMERISMODISCLARA-SALERBEInput SourceResolutionFrequencyTemporal CoverageType of AlbedoAVHRR and MODISAATSR, MERIS, VGT, and MODISMERISMODISAVHRRERBE0.05 0.05 0.25 0.05 0.25 2.5 8 dayMonthlyMonthly8 day10 day and MonthlyMonthly1981 to present1998–20112002–20062000 to present1982–20091985–1989BSA and WSABSA and WSABSABSA and WSABSABSASurface albedo products with an absolute accuracy of 0.02–0.03 are generally required for regional and globalclimate studies [Dozier et al., 1989; Sellers et al., 1995]. As climate modeling techniques have advanced in recentdecades by introducing temporally involved albedo parameterizations, albedo accuracy requirements havegenerally changed to be more specific based on the time and location of the applications [e.g., Loew et al., 2014;Wang et al., 2007; Widlowski et al., 2011]. Comparisons of satellite-based albedo data sets with generalcirculation models (GCMs) have been made in previous studies [Oleson et al., 2003; Wang et al., 2006; Wanget al., 2004; Wang and Zeng, 2010; Zhang et al., 2010; Zhou et al., 2003]. For example, Wang et al. [2006]compared results from the Coupled Model Intercomparison Project Phase 3 (CMIP3) models with data fromthe International Satellite Cloud Climatology Project (ISCCP) for the period 1984–1999 and found that themodeled albedo was systematically overestimated by as much as 0.05 in summer, compared to the ISCCPdata. Zhang et al. [2010] compared the Moderate Resolution Imaging Spectroradiometer (MODIS) albedoproducts with the results from CMIP3 models for the period 2000–2008 and found that differences inannually averaged global albedo can be as large as 0.06. However, a thorough comparison between globallong-term satellite-based albedo data sets has not yet been attempted.Surface albedo varies both spatially and temporally. Fang et al. [2007] calculated the variation of albedoover the continent of North America for different plant function types (PFTs) using MODIS time series andgenerated the shortwave albedo climatology for each of the PFTs. They observed that the within-classstandard deviation (SD) shows a strong seasonal character; that is, the SD increases in winter and spring anddecreases in the growing season. They also found the strong link between maximal variation in surfacealbedo and events such as winter snowfall and spring snowmelt. Zhang et al. [2010] analyzed a MODIS albedotime series to map global albedo variation and found a decrease of 0.01 in land surface albedo for theNorthern Hemisphere for the period 2000–2008. Gao et al. [2005] found that the interannual variation ofthe MODIS shortwave albedo is less than 0.01 over snow-free surfaces suggesting that the albedo trend ispossibly associated with land surface changes in cryosphere.Surface albedo may change with land cover dynamics caused by deforestation, afforestation, urbanization,snowfall, snowmelt, etc. Previous studies have demonstrated changes in surface albedo at various differentlocations throughout the past three decades [e.g., He et al., 2013; Shi and Liang, 2013]. The Northern Hemisphere,which contains most of Earth’s land surface and about 90% of the total population, is believed to have beenaffected by recent climate changes. Surface albedo changed dramatically for the Northern Hemisphere withdifferent trends in winter and summer during the past decades. Many studies have reported warming trendsdue to climate change in the Northern Hemisphere in recent decades [e.g., Flanner et al., 2011; He et al., 2013;Jeong et al., 2011]. It is therefore useful to offer a brief trend analysis to help identify the magnitude andcontributors of surface albedo changes from multiple data sets used in this study and to further improveclimate models to better capture the surface changes.The main objective of this study is to identify the differences and potential issues of the existing global satellitealbedo data sets and make possible suggestions to modeling communities when facing data set selectionfor model validation and calibration purposes. Because most previous studies focused on a relatively shortTable 2. Global Satellite Surface Shortwave Radiation Products Used in the ComparisonAlbedo Data SetsGEWEXISCCPCERESHE ET AL.Full NameSpatial ResolutionTemporal CoverageGlobal energy and water exchanges projectInternational satellite cloud climatology projectClouds and Earth’s radiant energy system1 280 km (2.5 )1 1983–20071983–20092000–2012 2014. American Geophysical Union. All Rights Reserved.10,282

Journal of Geophysical Research: Atmospheres10.1002/2014JD021667period in analyzing the climatological surface albedo, we did so for a period of 30 years (1981–2010) inthis study. We compared nine long-term global satellite-based albedo data sets, discussed the issues of eachdata set, and performed a brief trend analysis of land surface albedo. Section 2 introduces each albedo dataset used for in the comparison as well as the methodology for albedo calculation and comparison, which isfollowed by the climatological and trend analysis and discussions in section 3.2. Data and Method2.1. Global Long-Term Satellite Albedo Products2.1.1. MODISThe MODIS sensors on Terra and Aqua provide measurements on a global basis every 1 or 2 days with sevenspectral bands in the shortwave range for land applications. Currently, MODIS albedo products are generatedevery 8 days since early 2000 [Gao et al., 2005; Schaaf et al., 2002]. The atmospherically corrected surfacereflectance data during a 16 day compositing period are collected as the input of the kernel models to calculatesurface albedo. Albedo and the corresponding quality data are available at 500 m resolution in sinusoidalprojection and 0.05 in latitude/longitude projection. Both black-sky albedo (BSA) and white-sky albedo (WSA)are provided through the MCD43 data set. The Collection 5 MCD43C product was used in this study.MODIS albedo products have been validated extensively [Cescatti et al., 2012; He et al., 2012; Liu et al.,2009; Roman et al., 2013; Stroeve et al., 2005; Z Wang et al., 2012]. They have been widely used as a benchmarkfor evaluating other satellite albedo products.2.1.2. MERISA global albedo climatology was generated using the Medium-Resolution Imaging Spectrometer (MERIS)data for the period 2002–2006 [Popp et al., 2011], which was designed for the estimation of cloud fractionfor the Fast Retrieval Scheme for Clouds from the Oxygen A-band (FRESCO ) algorithm. MERIS BSA data fromthe period October 2002 to October 2006 were aggregated to a grid of 0.25 0.25 for each month of theyear and for different spectral channels [Popp et al., 2011]. In this study, the conversion from spectral toshortwave broadband albedo used for MERIS was based on the empirical equation derived by Liang [2001].2.1.3. GLASSThe Global Land Surface Satellites (GLASS) albedo product from 1981 is produced from advanced very highresolution radiometer (AVHRR) and MODIS data [Liang et al., 2013b]. The GLASS albedo product from MODISobservations is based on two direct albedo estimation algorithms, one designed for surface reflectance andone for TOA reflectance [Qu et al., 2014], and the statistics-based temporal filtering fusion algorithm is usedto integrate these two albedo products [Liu et al., 2013a]. The GLASS albedo product from AVHRRobservations is based on a direct estimation algorithm using the surface reflectance with radiometriccalibration and atmospheric correction [Pedelty et al., 2007] comparable to that used on MODIS data [Liu et al.,2013b]. Global albedo maps derived from AVHRR and MODIS are available at a resolution of 0.05 ( 5 km)every 8 days. In addition, albedo data at 1 km resolution in sinusoidal projection derived from MODISobservations are provided at the same temporal resolution. Both BSA and WSA are provided in the GLASSalbedo product.The GLASS albedo product has been evaluated using ground measurements and the MODIS albedo productin previous studies [He et al., 2013; Liu et al., 2013b]. The evaluations show that GLASS albedo has anaccuracy that is comparable to that of MODIS albedo product and that this has been the case consistentlythroughout the period of 1981–2012.2.1.4. GlobAlbedoThe GlobAlbedo project aims to provide global surface albedo data for the period of 1998–2011 based onEuropean satellites at three different spatial resolutions from 1 km, 0.05 to 0.5 . Data derived from theAdvanced Along-Track Scanning Radiometer (AATSR), SPOT4-VEGETATION, SPOT5-VEGETATION2, and MERISare integrated using an optimal estimation approach [Lewis et al., 2013] and a gap-filling technique basedon the MODIS surface anisotropy data set [Lewis et al., 2013; Muller et al., 2012]. Both BSA and WSA areprovided in GlobAlbedo product. In this study, the monthly GlobAlbedo data at a 0.05 resolution were used.Preliminary validation of GlobAlbedo products showed a generally good agreement against MODIS albedoproducts with an R2 of 0.85 on a global basis; however, problems with snow detection at high latitudes ( 70 )were found to cause significant artifacts in the final product [Muller, 2013].HE ET AL. 2014. American Geophysical Union. All Rights Reserved.10,283

Journal of Geophysical Research: Atmospheres10.1002/2014JD0216672.1.5. CLARA-SALThe time series from the Clouds, Albedo, and Radiation-Surface Albedo (CLARA-SAL) provide the globalshortwave BSA for the period 1982–2009, derived from AVHRR sensors [Riihela et al., 2013]. The CLARA-A1-SAL(AVHRR First Release) was used in this study. Atmospheric effects were corrected using the simplified methodfor atmospheric correction [Rahman and Dedieu, 1994], assuming a constant value of 0.1 for the aerosol opticaldepth (AOD) and an ozone constant of 0.35 cm. For vegetated surfaces, shortwave broadband albedo wasconverted from spectral albedo [Liang, 2001] after removing the surface anisotropic effects [Wu et al., 1995].For snow and ice surfaces, broadband albedo was directly converted from surface reflectance at a spatialresolution of 0.05 following the method of Xiong et al. [2002] and then averaged to calculate the monthlymeans at a spatial resolution of 0.25 . Efforts were made to provide a temporally consistent data set fromdifferent AVHRR sensors, by considering corrections for sensor calibration and orbital drift [Heidinger et al.,2010]. This product has been validated thoroughly against ground measurements at spatially representativesites around the world showing a relative uncertainty of 11% in the monthly albedo estimation [Riihela et al.,2013]. They also found that using a constant AOD value of 0.1 at 550 nm led to a typical overestimation of5–10% in surface albedo [Riihela et al., 2013].2.1.6. ERBESurface albedo has been generated from the broadband Earth Radiation Budget Experiment (ERBE) sensorson board two polar-orbiting NOAA satellites and one Earth Radiation Budget Satellite (ERBS) [Li and Garand,1994]. Estimated from the satellite observations by the scene-dependent angular models, the ERBE TOAalbedo was used to derive surface albedo with the empirical relationship established based on radiativetransfer simulations with a constant AOD of 0.05 at 550 nm. Validation of the instantaneous surface albedofrom the ERBE observations at two agricultural sites showed a bias of 0.01 with a root-mean-square error(RMSE) of 0.03 [Li and Garand, 1994]. Jin et al. [2003] reported a 0.90 correlation with an RMSE of 0.047between MODIS and ERBE albedos. Monthly climatology of the ERBE surface albedo used in this study wascalculated from the clear-sky observations obtained during 1985–1989 and available at a spatial resolution of2.5 covering the regions from 60 N to 60 S [Li and Garand, 1994].2.2. Albedo From Global Long-Term Surface Shortwave Radiation Data Sets2.2.1. ISCCPThe International Satellite Cloud Climatology Project’s (ISCCP) flux data set-monthly mean of profiles ofradiative fluxes data set provides monthly global surface shortwave radiation budget estimates at a spatialresolution of 2.5 ( 280 km) from July 1983 to December 2009 [Zhang et al., 1995; Zhang et al., 2004].Comparisons of surface albedo from ISCCP against other radiation data sets indicate that ISCCP data have ageneral underestimation of land surface albedo on a global basis, particularly in tropical regions [Stackhouseet al., 2012].2.2.2. GEWEXThe Global Energy and Water Exchanges Project’s (GEWEX) surface radiation budget (SRB) product version 3.0was produced by the National Aeronautics and Space Administration (NASA)/GEWEX to facilitate study ofEarth’s radiation budget under global and regional climate change. The SRB data set has a temporal coverageof 24.5 years, from July 1983 to December 2007, at a spatial resolution of 1 . The shortwave radiation dataare estimated using the algorithms developed by Pinker and Laszlo [1992], which requires cloud propertyinputs from ISCCP, reanalysis temperature and moisture from the fourth Goddard Earth Observing SystemModel (GEOS-4), and ozone observations from multiple satellites. Validations of GEWEX surface albedo withother radiation data sets showed an underestimation of GEWEX albedo over snow/ice surfaces [Qin et al.,2011; Stackhouse et al., 2012].2.2.3. CERESThe latest monthly shortwave radiation data from the Clouds and the Earth’s Radiant Energy System (CERES)sensors on board Terra and Aqua are available at a spatial resolution of 1 , from March 2000 to September 2012,using MODIS surface anisotropy data as background information. CERES products include both clear-sky andall-sky shortwave radiation estimates. We chose the most recent CERES Edition 2.7r all-sky surface downwardand upward radiation from the Energy Balanced and Filled (EBAF) data set [Kato et al., 2013] to calculate theblue-sky albedo in this study. Broadband albedos derived from CERES data have been validated in previousstudies, which revealed an underestimation of 0.003–0.008 relative to MODIS albedos [Hudson et al., 2010; Rutanet al., 2009]. Kato et al. [2013] evaluated the uncertainties in the CERES-derived irradiance with satellite-derivedHE ET AL. 2014. American Geophysical Union. All Rights Reserved.10,284

Journal of Geophysical Research: Atmospheres10.1002/2014JD021667cloud and aerosol properties and found that the monthly mean upward shortwave radiation over land couldhave uncertainties of 12 W m 2 at a grid level, which can be translated into an error of 0.06 in surface albedogiven the mean downward shortwave radiation of 203 12 W m 2.2.3. Calculation of Global and Regional Average AlbedoEarth’s surface albedo (blue-sky albedo) is the ratio of the reflected solar radiation to the incident solar radiation.Thus, it is temporally and spatially scalable. To calculate the blue-sky surface albedo over a region, monthlyaveraged data are spatially aggregated following equation (1), for which we need to consider the downwardradiation. For those data sets that only provided BSA, equation (2) was used to calculate the spatial mean albedo.X i iiα bs þ f difα iwsAi F di 1 f difXα¼;(1)Ai F diXAi F di α ibsα¼ X i i ;A Fd(2)where α is the spatially aggregated shortwave albedo. For pixel i, Ai is the area of the pixel, F di is the surfaceiis the diffuse skylight ratio, and α ibs and α iws are the BSA anddownward radiation under all-sky condition, f difWSA, respectively.Although downward radiation data from GEWEX have been used for the purpose of spatial aggregation ofalbedo before [Zhang et al., 2010], estimation of downward radiation over bright surfaces [Gui et al., 2010]may result in inaccurate albedo estimates at high latitudes. Instead, we used the monthly downwardradiation data from the Modern-Era Retrospective Analysis for Research and Applications (MERRA) reanalysisto calculate the spatial mean albedo in equations (1) and (2) at a finer spatial resolution of 1/2 2/3 [Rieneckeret al., 2011].Differences between BSA and blue-sky albedo have been shown to be very small over snow-free surfaces whenthe solar zenith angle is less than 70 [Liu et al., 2009]. However, the importance of the coupling of diffusedownward radiation in surface albedo estimation has been emphasized [Pinty et al., 2005; Roman et al., 2010],particularly for snow-covered surfaces. In previous studies, the diffuse skylight ratio has been either ignored orassumed constant in the calculation of the blue-sky albedo [Zhang et al., 2010]. To improve the albedoestimation over large regions, we used monthly diffuse and direct downward radiation data from the NationalCenters for Environmental Prediction (NCEP) reanalysis [Kalnay et al., 1996] to derive the diffuse skylight ratio inequation (1). Figure 1 shows the diffuse skylight ratio in the total shortwave range, derived from NCEP data.It indicates that the diffuse skylight ratio is not constant but that it varies over both time and space. We haveaggregated albedos using both MERRA and NCEP downward radiation data as inputs in equation (1). Despitethe difference in the spatial resolution, the correlation between albedos aggregated up to the global levelbased on these two data sets was found to be 0.994 at the 99% confidence level.To consistently prevent ocean pixels from corrupting the land surface albedo calculation over the time periodin this study, we used the MODIS land cover map (MCD12C1) of 2000 to identify and exclude the same oceanpixels for each temporal observation. The land and ocean information from the original MODIS land coverdata at a 0.05 resolution was directly applied on the GLASS, MODIS, and GlobAlbedo data to screen thenonland pixels. We aggregated the land cover data to match the spatial resolutions for all the other data setsbased on a majority rule: if the majority of the fine-resolution pixels to be aggregated are land, thecorresponding coarse-resolution pixels were masked as land. Because CLARA-SAL albedo was first generatedat a spatial resolution of 0.05 including sea ice pixels and then aggregated to 0.25 in its publicly releasedproduct [Riihela et al., 2013], it is highly likely that the CLARA-SAL albedo aggregated in this study will cause anoverestimation compared with other fine-resolution land albedo products if we applied the majority rule inspatial aggregation.Unlike the other data sets that are available on a monthly basis, GLASS and MODIS products are provided at8 day intervals. To get comparable monthly mean albedo from these two data sets and to maintainhigh-quality albedo data, only the albedo data with quality flags of 0 and 1 were included in the calculationfor these two data sets.HE ET AL. 2014. American Geophysical Union. All Rights Reserved.10,285

Journal of Geophysical Research: Atmospheres10.1002/2014JD021667Figure 1. Global diffuse skylight ratio derived from NCEP data for (a) January 2000 and (b) July 2000.3. Results and Discussion3.1. Land Surface Albedo Climatology3.1.1. Global Land Surface AlbedoMonthly albedo averages were calculated for each of the data sets mentioned in section 2 except for theERBE data due to its spatial coverage. As can be seen from the global values shown in Figure 2a, most of thesatellite albedo products agree relatively well and can likely satisfy the accuracy requirements for globalclimate applications, with differences less than 0.02. Fang et al. [2007] found that in North America albedoincreases in winter and spring and decreases in the growing season. It is interesting to find that the meansurface albedo for the Northern Hemisphere peaks in March (Figure 2b). This can be explained by the factthat high-latitude regions on the Northern Hemisphere receive little amount of downward solar radiation( 5 W m 2) in winter. Consequently, strongly reflective surfaces common to these latitudes, such as snowand ice, do not actively contribute to the shortwave radiation budget. From winter to spring, increasinglyhigh surface albedos are observed, because snow and ice contribute to the radiation budget more and moreas a result of increasing exposure to sunlight. Based on the result of our climatology comparison, we canconclude that most of the satellite-based albedo data sets shown in Figure 2 can serve as the backgroundinformation in global long-term climate modeling studies.Although mismatch exists within temporal coverages among the selected data sets, the magnitude ofdifference among the derived albedo climatologies is significantly larger than that of the three decadal changesin global mean land surface albedo (see section 3.2). Thus, it is reasonable to compare the climatologies derivedHE ET AL. 2014. American Geophysical Union. All Rights Reserved.10,286

Journal of Geophysical Research: Atmospheres10.1002/2014JD021667from these data sets with differenttemporal coverages. ISCCPclimatological albedos were found tobe considerably underestimated,particularly from June to September.This confirms the finding of Wang et al.[2006] that the GCM simulation resultsoverestimated surface albedos, relativeto ISCCP albedos by about 0.05 insummer at northern latitudes. Based onresults from other satellite productsshown in this study (Figure 2), it is likelythat ISCCP surface albedos werenegatively biased, particularly fromJune to August for both hemispheres.An intercomparison made among theCERES, GEWEX, and ISCCP surfaceshortwave radiation data suggestedthat the underestimation of ISCCPsurface albedo in middle to lowlatitudes was possibly the result ofinaccurate estimation on atmospherictransmittance over urban and tropicalregions [Stackhouse et al., 2012].Stackhouse et al. [2012] also pointedout that GEWEX had significantunderestimation in albedo over snow/icesurfaces, which is likely the reason ofits underestimation in winter and earlyspring shown in Figure 2.The CLARA-SAL albedo product tends tohave the highest values in most seasonsfor both hemispheres. There are threereasons for this overestimation. First,the CLARA-SAL product includes seaice albedo estimates, which may affect our regional aggregates. Its coarse resolution is likely to result inmixed pixels from sea ice and land cover. It matched the data from CERES very well, which is a coarseresolution data set with pixels mixed from sea ice and land covers and believed to be more accurate than ISCCPand GEWEX [Stackhouse et al., 2012]. Second, the CLARA-SAL product provides only the BSA, which is typicallywith higher value than the actual “blue-sky” albedo, if the solar zenith angle is large. Third, the weaker clouddetection ability of the AVHRR sensors makes it more likely for cloud pixels to be misidentified as snow/ice pixelsat latitudes higher than 50 [Karlsson et al., 2013], leading to overestimation of surface albedo.Figure 2. Monthly climatological surface shortwave albedo derived fromsatellite-based albedo data sets for (a) the globe, (b) the NorthernHemisphere, and (c) the Southern Hemisphere. The “mean” and “SD” arecalculated from all the data sets except the ISCCP.Differences between the albedo data sets range from approximately 3% to 5% (not considering the ISCCP data)depending on the season and location. In the Northern Hemisphere, surface albedo peaks in March and April,when the amount of snow cover exposed to sunlight reaches its maximum. Not surprisingly, differences inalbedo also peak for the same season with SD 0.02 (Figure 2b). For the Southern Hemisphere, these data setswere not in good agreement from October to January with SD 0.015 (not considering the ISCCP data), whichlikely resulted from the large differences in albedo over the Antarctic region (Figure 3).Differences between the albedo data sets were found to be larger for the Northern Hemisphere (Figure 2b)than for the Southern Hemisphere (Figure 2c). This is likely because seasonal snow cover is moreextensive in the Northern than in the Southern Hemisphere. Thus, for the Southern Hemisphere, estimationaccuracy of the aggregated albedos would suffer less from uncertainties in the snow cover detection andparameterization of snow albedo. Disagreement on spring albedo for the northern high latitudes could beHE ET AL. 2014. American Geophysical Union. All Rights Reserved.10,287

Journal of Geophysical Research: Atmospheres10.1002/2014JD021667attributed to a high degree ofsensitivity of the land surface albedo tothe different spatial resolutions (e.g.,partial/subpixel snowmelt) andtemporal composition strategies usedin the different albedo data sets [Heet al., 2013]. In addition, changes inplant phenology, poleward expansionof tree line, and other climate-relatedvariations are more prominent andcomplex in the Northern Hemisphereincreasing the uncertainties in theclimato

Analysis of global land surface albedo climatology and spatial-temporal variation during 1981–2010 from multiple satellite products Tao He 1, Shunlin Liang1,2, and Dan-Xia Song 1Department of Geographical Sciences, University of Maryland, College Park, Maryland, USA, 2State Key Laboratory of Remote Sensing Scien