JOURNAL OF GEOPHYSICAL RESEARCH, VOL. 110, D01104, doi: /2004jd005190, 2005

Transcription

1 JOURNAL OF GEOPHYSICAL RESEARCH, VOL. 110,, doi: /2004jd005190, 2005 MODIS bidirectional reflectance distribution function and albedo Climate Modeling Grid products and the variability of albedo for major global vegetation types Feng Gao, 1,2,3 Crystal B. Schaaf, 1 Alan H. Strahler, 1 Andreas Roesch, 4 Wolfgang Lucht, 5 and Robert Dickinson 6 Received 1 July 2004; revised 8 October 2004; accepted 24 October 2004; published 8 January [1] Global land surface albedo data sets derived from the Terra Moderate-Resolution Imaging Spectroradiometer (MODIS) from March 2000 to present have been completed for ready use by the global modeling community. This paper describes these albedo and bidirectional reflectance distribution function Climate Modeling Grid products and their variability within major global vegetation types. Preliminary results based on collection 4 data reveal that these coarse resolution (0.05 ), geographic (latitude/ longitude), global albedos have spatial and temporal patterns appropriate for the underlying land cover classes, further encouraging modelers to introduce albedos as functions of ground cover, geographic location, temporal season, and spatial resolution in the various climate-modeling schemes. Citation: Gao, F., C. B. Schaaf, A. H. Strahler, A. Roesch, W. Lucht, and R. Dickinson (2005), MODIS bidirectional reflectance distribution function and albedo Climate Modeling Grid products and the variability of albedo for major global vegetation types, J. Geophys. Res., 110,, doi: /2004jd Introduction [2] Land surface albedo is the ratio of the light flux in solar wavelengths reflected from a surface area into a hemisphere to the total incoming incident flux. It is one of the most important parameters characterizing the Earth s radiative regime and thus has an impact on biospheric and climate processes [Dickinson, 1995]. Most climate models currently rely on the land cover types present in a grid box to compute or deduce land surface albedo [Dickinson, 1995]. The albedos of various land cover types are primarily based on field measurements [Henderson-Sellers and Wilson, 1983] or coarse spatial resolution climatologies such as those derived from Earth Radiation Budget Experiment (ERBE) observations [Li and Garand, 1994]. [3] The Moderate-Resolution Imaging Spectroradiometer (MODIS) instruments on both of NASA s Terra and Aqua platforms now acquire daily images of the globe and routinely provide remotely sensed global land surface 1 Department of Geography and Center for Remote Sensing, Boston University, Boston, Massachusetts, USA. 2 Also at Research Center for Remote Sensing and GIS, School of Geography, Beijing Normal University, Beijing, China. 3 Now at Earth Resources Technology, Inc., Jessup, Maryland, USA. 4 Department of Environmental Sciences, Swiss Federal Institute of Technology Zurich, Zurich, Switzerland. 5 Potsdam Institut für Klimafolgenforschung, Potsdam, Germany. 6 School of Earth and Atmospheric Sciences, Georgia Institute of Technology, Atlanta, Georgia, USA. Copyright 2005 by the American Geophysical Union /05/2004JD005190$09.00 albedos. MODIS albedo products are calculated by integrating directional surface reflectances over the viewing hemisphere and are therefore dependent on the anisotropy of the angular surface reflectance function (the bidirectional reflectance distribution function, or BRDF) [Wanner et al., 1995]. The MODIS instrument has a wide field of view that enables the same ground location to be observed from different viewing geometries as the daily satellite orbit changes. A 16-day period of inputs is usually sufficient to acquire enough observations to capture and model the BRDF of a 1-km land pixel. Albedo describes the reflectance in the entire broadband shortwave regime; seven MODIS land bands which span the visible (with 3 bands) to the nearinfrared and midinfrared (with 4 bands) allow for accurate conversions from these spectral bands to broadband (visible, near-infrared and shortwave) albedos [Liang, 2001]. All MODIS products incorporate extensive quality control flags that allow both users and higher-level MODIS product algorithms to select the most accurate inputs from the lower level products. These attributes of the MODIS data also assist in retrieving accurate global albedo at a coarser resolution. [4] The MODIS BRDF/albedo products have been initially validated and the recent collection 4 reprocessed data are characterized as Validated Stage 1 (product accuracy has been estimated using a small number of independent measurements obtained from selected locations and time periods and ground-truth/field program effort). Validation results thus far show good agreement with field measurements, typically within 10% (relative) [Liang et al., 2002; Jin et al., 2003a, 2003b; K. Wang et al., 2004]. Focused efforts to compare the albedo produced by climate models 1of13

2 and the MODIS albedo show that albedo agreements depend on the regions, seasons and surface types [Oleson et al., 2003; Zhou et al., 2003; Z. Wang et al., 2004; Roesch et al., 2004]. High-quality characterizations of surface albedo that accurately capture both spatial and temporal features across the entire globe continue to be a high priority for modeling efforts. [5] While we have previously investigated the albedo of areas of high spatial heterogeneity, such as snow-covered vegetation [Jin et al., 2002] or desert regions [Tsvetsinskaya et al., 2002], we now expand this work to address the albedo of major global vegetation types with a particular focus on their spatial, temporal and within-class variability in the Climate Modeling Grid (CMG) albedo product. We discuss the major causes for this variability and the implications for the representation of albedo at coarser spatial resolutions in climate models, where albedo feedback drives some of the most important climate change processes. 2. MODIS BRDF/Albedo Climate Modeling Grid (CMG) Products [6] The attributes of the standard MODIS BRDF/albedo 1-km resolution products (called MOD43B in the MODIS land products family) have been well documented (Strahler et al. [1999] (available at MODIS/LAND/#albedo-BRDF), Lucht et al. [2000], Schaaf et al. [2002], and MOD43 User s Guide, available at These products are provided to the user community by the EROS Data Center (EDC) in equal-area tiles of 1200 pixels by 1200 pixels in a sinusoidal projection. In addition to these 1-km standard products, Climate Modeling Grid (CMG) albedo products, specially designed for the global modeling community, are also provided as global files in a geographic latitude/longitude projection at a 0.05 spatial resolution (called MOD43C in the MODIS land products family). This paper highlights the availability of these CMG products and discusses the temporal and spatial within-class variability of albedo for major vegetation types. Both the 1-km resolution products and the CMG products have been reprocessed (Version V004) and are provided for every 16-day period from March 2000 to present. A forward sampling process is used in the production of the MODIS CMG BRDF/albedo product such that each pixel in a 1-km resolution tile is reprojected and aggregated into the appropriate CMG 0.05 resolution grid box in a geographic projection. All available 1-km albedo values for that grid box are then averaged and the overall product quality is assigned by applying the most frequently occurring quality flag. The MODIS BRDF model parameters also can be averaged because of the linearity of the semiempirical BRDF model retrieved. In addition to the overall quality, the MODIS BRDF/albedo CMG product also includes extensive additional quality control (QC) information about the underlying 1-km resolution values used in the aggregation, such as the percentage of snowcovered pixels present in each grid box, the majority quality value of the BRDF model inversions, and the percentage of valid 1-km data spatially available to contribute to the CMG grid box value. [7] The 1-km resolution MODIS BRDF/albedo algorithm uses all high-quality cloud-cleared and atmospherically corrected 1-km surface reflectances or bidirectional reflectance factors (MOD09 in the MODIS product family) available over a 16-day period and determines the best BRDF parameters for each location [Schaaf et al., 2002]. If, because of cloud contamination, there is no clear MODIS observation acquired during a 16 day period, the corresponding pixel is set to a fill value in the MOD43B product and is excluded from the computation of the CMG grid box value and quality. However, all other nonfill albedo values are averaged for the CMG grid box (regardless of the QA flag associated with each underlying 1-km value and thus including all magnitude inversions [Strugnell et al., 2001] as well as all full inversions) and that grid box assumes the QA flag of the majority of the underlying pixels contributing to it. Note that the underlying 1-km product retrieves albedo values out beyond the coastline to the shallow water boundary (5 km in extent or 50 m deep); thus if albedos associated with water (or coastal zones, or even occasionally snow and ice) are retrieved, then they are included in the average for that CMG grid box. Narrow to broadband conversion is implemented in the MOD43B 1-km products and thus the MODIS BRDF/albedo CMG products also include all seven MODIS land bands and three broad bands (visible, near-infrared, and total shortwave) for both black-sky albedo (or directional hemispherical reflectance) at local solar noon and white-sky albedo (or bihemispherical reflectance under conditions of isotropic illumination). Black-sky albedo at other solar zenith angles can be computed using the CMG BRDF parameters. [8] The actual albedo at a given solar zenith angle, which includes both direct beam and diffuse skylight, can be approximately expressed as a linear, weighted combination of white-sky albedo and black-sky albedos [Lewis and Barnsley, 1994]. The weights (or proportions of diffuse skylight and direct beam) are determined by aerosol parameters, such as optical depth and aerosol type, and can be calculated by any widely used atmospheric radiative transfer package such as 6S (Second Simulation of the Satellite Signal in the Solar Spectrum) [Vermote et al., 1997]. The diffuse part of actual albedo (white-sky albedo times the percentage of diffuse skylight assuming isotropic illumination) approximates the diffuse albedo as calculated by traditional radiative transfer models. The direct part of actual albedo (black-sky albedo times the percentage of direct beam at a particular illumination angle) represents the traditional direct albedo. [9] Archived global MODIS BRDF/albedo CMG products include white-sky albedo and black-sky albedo at local solar noon, BRDF model parameters, and nadir BRDFadjusted reflectance (NBAR) as three separate outputs (MOD43C1, MOD43C2, MOD43C3). As mentioned earlier, each of the outputs is accompanied by quality flags that include the percentage of snow and percentage contribution of fine resolution data. Products are saved in hierarchical data format (HDF-EOS), which includes additional global metadata. [10] Figure 1 shows examples of global albedo mapped from the standard MODIS 0.05 albedo CMG products. The two 16-day periods selected for illustration in Figure 1 represent the white-sky albedo in January and July, Images from the top to the bottom represent the white-sky albedo for visible, near-infrared, and total shortwave broad 2of13

3 Figure 1. Collection 4 MODIS CMG albedo products for January and July Black areas represent missing values due to lack of clear observations. bands and the percentage of snow of each 0.05 box for the associated period. Black identifies missing values. We choose white-sky albedo for demonstration since whitesky albedo is independent of view and solar angles and thus is comparable globally and temporally. The global albedo spatial patterns in the figures depend on both the underlying surface types and their structures. The semiarid and desert regions have much higher albedos than other snow-free regions but still capture the large intraclass differences present in deserts [Tsvetsinskaya et al., 2002]. The seasonal changes of albedo are strongly dependent on the vegetation types and their location, especially in the 3of13

4 near-infrared waveband. Snow and ice (shown by the percentage of snow cover in the CMG QC bits) cause large changes of albedo at high latitudes, especially in the Northern Hemisphere during the winter season [Jin et al., 2002]. The magnitude of the albedo increase under snow conditions is also dependent on the vegetation type. Snow increases least the albedo of a surface with high canopy density and vertical structure (e.g., forest) and decreases most the albedo of a surface with sparse and/or short vegetation (e.g., grassland). 3. Land Surface Albedo Spatial Variations [11] The land surface albedo of any geographic location varies depending on the predominant land cover type, the density and structure of that land cover, the spectral properties of the components composing a landscape, and the temporal season that is being observed. There is a great deal of interest in comparing model results with the typical albedos observed by MODIS for any of these predominant land cover types. Thus this section focuses on the global spatial variability of albedo for major global vegetation types as observed by the MODIS CMG types and section 4 focuses on the temporal variability of the same global vegetation types Spatial Variations for Major Vegetation Types [12] The most recent Collection 4 (V004) MODIS land cover and albedo CMG products were used to analyze the variations of albedo for major vegetation types. The MODIS Land Cover product (MOD12Q1) is produced from nadir BRDF-adjusted reflectance (NBAR) values using a decision tree classifier [Friedl et al., 2002]. The MODIS Land Cover CMG product includes the fraction of each IGBP type that occurs in the underlying 1-km data. This fraction allows users to select relatively pure land cover grid boxes from the CMG product. In our analysis related to the CMG land cover types, only those locations dominated by a single class (those with 90% or higher of a land surface type fraction in the CMG boxes) were used. Figure 2 shows histograms of global snow-free albedo for those CMG grid boxes dominated by evergreen needleleaf forest, deciduous broadleaf forest and grasslands in July Evergreen needleleaf forest shows the smallest within class albedo variance. Deciduous broadleaf forest shows a bimodal histogram, which reflects the different growing stages of deciduous broadleaf forest in the Northern and Southern Hemispheres. Deciduous broadleaf forest in the Northern Hemisphere shows higher albedos in near-infrared and shortwave broadband but lower albedos in visible broadband in July Grassland albedo variability is a direct consequence of the large variability of grasses, gap probability, leaf area, dead material under the grass and underlying soil types, which respond quickly to changes in such environmental parameters as soil water status. The histograms include all data regardless of quality to ensure a maximum number of pixels. [13] To further illustrate the snow-free albedo spatial patterns for these three IGBP types, the underlying 1-km white-sky albedo is displayed by latitude in Figure 3. Solid lines show the mean values computed from all data and dashed lines show the mean values computed only from overall high-quality data, as selected according to the QC bits. As seen in Figure 2, the albedo of deciduous broadleaf forest in July is significantly higher in the Northern Hemisphere than in the Southern Hemisphere. Grassland albedos appear to increase toward the equator and peak at the N latitude belt, which may be due to dry soils and sparse vegetation typical for the steppes south of the deserts in the Sahara and the southern Arabian Peninsula. Note that although the MODIS BRDF/albedo products are cloudcleared and atmospherically corrected, undetected cloud sometimes can still occur in the low-quality retrievals because of the difficult nature of cloud detection over tropical regions. The big differences of white-sky albedo between high-quality retrievals and all retrievals for grasslands between 10 and 20 N may be due to such residual cloud contamination in the low-quality retrievals. Evergreen forests in the N latitude belt and between 10 S and 10 N latitudes also show a significant bias in their albedos for all data compared to high-quality data. Except for these tropical area retrievals, the mean albedos of all-quality and high-quality retrievals are very close in the figure, which implies that the mean albedos from high-quality data capture the spatial variation well for this period Variations for Snow-Covered Vegetation Types [14] Snow cover strongly affects surface albedo. Jin et al. [2002] found that the magnitude of the albedo change under snowy conditions varies with IGBP class. Forest classes have lower mean snow-covered albedos, while nonforest classes have higher albedos [Jin et al., 2002]. Our work also confirms this result, shown in Table 1. The statistics in the table are produced from high-quality retrievals of 3 years ( ) of standard 16-day MODIS CMG albedo products. The mean values are 10 latitude-averaged values under snow-free conditions. The values denoted S max are 10 latitude-averaged albedos of the annual maximum values with 100% snow cover fraction for 3 years. More specifically, for those pixels with 100% snow coverage the annual maximum albedos are first selected from each year (twenty-three 16-day production periods). The 10 -latitudeaveraged values are then computed from the annual maximum albedos. The maximum values in the table are chosen from the three annual-maximum latitude-averaged albedos. Thus computation of S max can be denoted as n S max ¼ max average 10 degrees max 23 periods ðcmg pixels with high quality and 100 % snow coverþ o : [15] Therefore the maximum albedos in the table show the high-surface-albedo upper boundaries for snow-covered condition for major IGBP types. Table 1 shows that the maximum snow-covered albedos of forest types including deciduous types are generally lower than those of nonforest types, showing how canopy structure and extent masks snow cover. For example, maximum albedos under snow cover conditions for grasslands in the N latitude band are 0.839, 0.595, and for visible, NIR and shortwave broadband respectively. In contrast, the respective mean snow-free albedos (0.077, and 0.158) are substantially lower. For evergreen needleleaf forest, the 4of13

5 Figure 2. (a f) Histograms of global snow-free white-sky albedo of three IGBP classes during July The grayscale in the two-dimensional histograms (Figures 2a, 2c, and 2e) shows point density. A scale factor of 1000 is applied in the two-dimensional histograms (Figures 2a, 2c, and 2e). maximum albedos under snowy conditions are 0.278, 0.236, and for the three broad bands in the same latitude band. These values agree with field albedo measurements over boreal forests, where the albedo of the conifer sites in winter rarely reaches 0.3 [Betts and Ball, 1997]. Climate modelers have also recognized this impact by altering snow cover fraction parameterizations in GCMs. Roesch et al. [2001] proposed to considerably reduce the albedo of snowcovered forests in the ECHAM4 GCM. [16] Table 1 clearly shows that the albedo of snowcovered vegetation strongly depends on the vegetation type. In addition, the albedo of any surface type depends also on the snow depth. Figure 4 shows the relations of the shortwave white-sky albedos and snow fractions extracted 5of13

6 Figure 3. Variations of snow-free albedo with latitude for typical IGBP vegetation types for July Solid lines show means computed from all-quality retrievals, and dashed lines show data from highquality retrievals only. 6of13

7 Table 1. Three Complete Years ( ) of High-Quality, Latitude-Averaged, Snow-Free White-Sky Albedo and the Latitude- Averaged Maximum Totally Snow-Covered Albedo for Visible, Near-Infrared, and Total Shortwave Broadband Listed by IGBP Class and Latitude a IGBP Class b Mean SNO max Mean SNO max Mean SNO max Mean SNO max N N N N Visible Broadband 1 EverNdlF DeciNdlF DeciBdlF MixedF OShrub WSavannas Grasslands Croplands CVmosaic Near-Infrared Broadband 1 EverNdlF DeciNdlF DeciBdlF MixedF OShrub WSavannas Grasslands Croplands CVmosaic Shortwave Broadband 1 EverNdlF DeciNdlF DeciBdlF MixedF OShrub WSavannas Grasslands Croplands CVmosaic a High-quality, latitude-averaged, snow-free white-sky albedo is denoted as Mean in the table, and latitude-averaged maximum totally snow-covered albedo is denoted as SNO max in the table. b IGBP classes are as follows: 1, evergreen needleleaf forest (EverNdlF); 2, evergreen broadleaf forest (EverBdlF); 3, deciduous needleleaf forest (DeciNdlF); 4, deciduous broadleaf forest (DeciBdlF); 5, mixed forest (MixedF); 6, closed shrublands (CShrub); 7, open shrublands (OShrub); 8, woody savannas (WSavannas); 10, grasslands; 12, croplands; 14, cropland/natural vegetation mosaic (CVmosaic). from global high-quality retrievals in the MODIS CMG albedo products of 2001 (23 production periods). Snow fractions were unpacked from the quality flags embedded in the MODIS albedo CMG products. The figure demonstrates that the sensitivity of albedo with respect to the snow cover fraction is much higher for grassland than for evergreen needleleaf forest because of the masking of snow by canopy. 4. Temporal Variations of Major Vegetation Types [17] Changes in vegetation seasonal cycle, soil moisture, or vegetation type lead to surface albedo changes, which in turn change the radiation balance and thus have an impact on the climate models [Dickinson, 1995]. Therefore it is important to investigate both seasonal and interannual albedo variations Intra-annual Variations [18] Intra-annual variations of surface albedo over snowfree vegetated areas are mainly caused by seasonal changes. Figure 5 shows seasonal albedo changes for three typical vegetation covers between 30 and 40 N latitude from high- Figure 4. Albedos vary with snow cover fractions. Points in the figure were extracted from global high-quality retrievals during the whole year of 2001 (23 production periods). Each point in the figure represents a CMG resolution (0.05 ) pixel. Note that some snow-covered albedos in the figure are larger than those shown in Table 1 since the maximum albedos in Table 1 are the 10 -latitudeaveraged values of yearly maximum albedos with 100% snow coverage. 7of13

8 Figure 5. Temporal variations of snow-free white-sky albedo for typical vegetation types between 30 N and 40 N for visible, near-infrared, and shortwave broadbands. quality retrievals of The phenological cycles of deciduous broadleaf forest can be clearly seen in the figure. The near-infrared albedo shows more variation than visible broadband and total shortwave broadband albedo. The differences between minimum and maximum albedos for the near-infrared broadband can be as large as a factor of 2 for the deciduous forest type. Total shortwave albedo shows the smallest variation. This implies that it is not sufficient to characterize the variability of land cover types by computing only broadband shortwave albedo in order to correctly capture seasonal albedo variations in GCMs. [19] Figure 6 shows the range of albedos observed for the year The data shown are computed from the high-quality snow-free observations during the twenty-three 16-day periods of The differences of boreal forests, croplands and semi-desert areas are obvious. Only highquality retrievals are considered in the compositing process. Since residual cloud contamination and soil exposure usually increase the surface reflectance of vegetation canopies in the visible band, the minimum albedo map (Figure 6a) tends to capture values selected from the clearest period with the highest vegetation canopy covers. This map would therefore represent the albedo lower bound for use in climate models. Figure 6b shows the maximum albedo for the shortwave broadband. Some tropical regions, such as Central America, Central Africa and South Asia, are missing because of the lack of high-quality retrievals for the year Desert areas, such as those of the Sahara and central Australia, show the highest values, while boreal forests, northern Eurasian forests and southeastern Chinese forests show the smallest shortwave albedo values over land surface. [20] Figures 7a 7d are true color composite maps of the minimum and maximum albedo from red ( nm), green ( nm) and blue ( nm) spectral bands. Figures 7a and 7b are true color composite of snow-free spectral albedo from high-quality retrievals, while Figures 7c and 7d show the composite from allquality retrievals. The albedo retrieval qualities embedded in the standard CMG product are used to calculate the number of high-quality retrievals available over the entire year 2001 (Figure 7e). The quality figure reveals that the product has few high-quality retrievals over tropical forest regions and those grid boxes not associated with highquality retrievals should not be used. The true color composite of the maximum snow-free albedo from all-quality data (Figure 7d) highlights the residual cloud contamination that unfortunately still leaks through as low-quality results in tropical forest regions such as the Amazon basin, central Africa, and Southeast Asia. As cloud contamination generally increases the derived surface albedo, these maximum albedos in the tropics are related to cloudy periods. However since these low-quality, cloud-affected maximum values represent the worst (brightest) cases for each grid box through out the year, the standard CMG albedo products for a region for an individual 16-day period does, of course, reflect much less anomalous cloud. Other approaches, such as taking the second highest maximum value may have less cloud contamination problem in the tropics but will lose some cloud-free high albedos in other regions. This emphasizes the importance of checking the associated QA flags when using the product. We suggest that high-quality data in the MODIS BRDF/albedo CMG product should have the value of majority processed, good quality for the mandatory QA flag. Furthermore, using a threshold of 75% for the value of percent input QA flag will ensure that a CMG grid box captures the underlying spatial heterogeneity. [21] Table 2 summarizes the minimum and maximum albedo with latitude for major snow-free IGBP classes from 3 years of high-quality albedo data ( ). Open shrublands and grasslands have the largest albedo variation globally of all. The global yearly differences between the averaged maximum albedo and minimum albedo for grasslands in visible, NIR and shortwave broadband are 0.160, and 0.168, respectively, and are 0.136, and for open shrublands, respectively. As mentioned earlier, the high albedo variations of grasslands and open shrublands may be caused by different underlying soil types. Table 2 also shows that grassland and shrubland albedo values increase toward the equator as is seen in Figure 3. The third highest variation of albedo among vegetation types is that of cropland, which is likely due to 8of13

9 Figure 6. (a) Minimum and (b) maximum snow-free land surface white-sky albedo during 2001 with high-quality retrievals only. Note that tropical regions have fewer high-quality retrievals because of cloud contamination, which causes some tropical pixels to be missing (black areas) from the maps. the seasonal cultivation change that exposes the soil background during fallow periods. In most spectral bands, the next highest variability after croplands (with variability almost as great) is savannas, deciduous broadleaf forest and mixed forest, undoubtedly for the same reasons that grasslands and cropland have high variability. Although we have provided the statistics for the IGBP land cover classes, other statistics based on the specific climatic-ecological 9of13

10 Figure 7. True color composite maps of the minimum (Figures 7a and 7c) and maximum (Figures 7b and 7d) snow-free white-sky albedo during 2001 with high-quality (Figures 7a and 7b) and all-quality (Figures 7c and 7d) retrievals. Albedos are stretched from 0 to 0.1 and from 0 to 0.2 for minimum and maximum albedo, respectively. The number of high-quality retrievals (Figure 7e) during 2001 explains why land pixels are missing in high-quality pictures (Figures 7a and 7b). The true color composite map of maximum albedo (Figure 7d) from all-quality data highlights the residual cloud contamination that unfortunately still leaks through as low-quality results in tropical forest regions. 10 of 13

11 Table 2. Three Complete Years ( ) of Averaged High-Quality Minimum and Maximum White-Sky Snow-Free Albedo for Visible, Near-Infrared, and Total Shortwave Broadband Listed by IGBP Class and Latitude a N N 20 N to0 0 to 20 S S Global IGBP Class Min Max Min Max Min Max Min Max Min Max Min Max Difference Visible Broadband EverNdlF EverBdlF DeciNdlF DeciBdlF MixedF CShrub OShrub WSavannas Savannas Grasslands Croplands CVmosaic Near-Infrared Broadband EverNdlF EverBdlF DeciNdlF DeciBdlF MixedF CShrub OShrub WSavannas Savannas Grasslands Croplands CVmosaic Total Shortwave Broadband EverNdlF EverBdlF DeciNdlF DeciBdlF MixedF CShrub OShrub WSavannas Savannas Grasslands Croplands CVmosaic a IGBP classes are the same as in Table 1. The numbers in bold (open shrublands and grasslands) show the largest albedo variation globally. regions or on continents would also be useful for modeling albedo variations Interannual Variations [22] MODIS standard land products started routine production in March We selected three complete years ( ) of data for the interannual variation analysis. Table 3 summarizes the changes of minimum albedo and maximum albedo for major snow-free IGBP classes by latitude for two consecutive years. The annual differences from both all-quality retrievals and high-quality retrievals are included. As shown in Figure 7d, the maximum albedo can contain some cloud leaks in the tropical region from year to year, and thus the differences of maximum albedo between two consecutive years are higher from all-quality data than high-quality data. However, the spatial and temporal variations could be lost while filtering high-quality data as shown in Figures 7a and 7b, which causes some statistical biases (higher differences of minimum albedo from the high-quality data) for some types, such as deciduous broadleaf forests and mixed forests. The statistics from high-quality data between 20 north and south latitude and statistics from all-quality data beyond should be close to the true changes if we consider both data quality and completeness. Table 3 shows that interannual differences of minimum and maximum shortwave white-sky albedo are mostly less than However, some variations of minimum shortwave white-sky albedo can be observed, with changes of more than 0.01 from both high-quality data and all-quality data for open shrublands, savannas and grasslands. The open shrublands and grasslands are the most variable IGBP classes for albedo, not only spatially but also intra-annually and interannually. [23] Table 3 reveals that visible, near-infrared, and shortwave albedos averaged over 20 latitude belts barely vary between consecutive years, an observation that supports the conclusion that the statistics in Tables 1 and 2 from three consecutive years are reliable. Note that even with three complete years of albedo data included in the statistics, we consistently lack high-quality retrievals for tropical regions because of cloud contamination. We continue to work on improving albedo retrieval quality over tropical regions 11 of 13

12 Table 3. Difference of Averaged Minimum and Maximum White-Sky Snow-Free Albedo for Shortwave Broadband Listed by IGBP Class and Latitude Between Two Consecutive Years a N N 20 Nto0 0 to 20 S S Min Max Min Max Min Max Min Max Min Max IGBP Class High All High All High All High All High All High All High All High All High All High All Between 2002 and 2001 EverNdlF EverBdlF DeciNdlF DeciBdlF MixedF CShrub OShrub WSavannas Savannas Grasslands Croplands CVmosaic Between 2003 and 2002 EverNdlF EverBdlF DeciNdlF DeciBdlF MixedF CShrub OShrub WSavannas Savannas Grasslands Croplands CVmosaic a High values from high-quality data; all values from all-quality data. IGBP classes are the same as in Table 1. (An albedo scale factor of is used for clearer presentation.) for the next generation of MODIS products [Gao et al., 2002]. 5. Conclusions [24] The MODIS BRDF/albedo CMG algorithm produces global albedo for each 16-day period at 0.05 spatial resolution. Collection 4 (V004) data are available from March 2000 to present, thus providing over four years of consistent global data to users. [25] The variability of albedo depends on the ground cover type and background, geographic location, season, and snow cover. The yearly cycle snow-free minimum albedo represents the lower limit of possible surface albedo values while the maximum albedo under snow cover conditions represents the upper limit of annual surface albedos. However, residual cloud contamination from the poorest quality retrievals can still affect the snow-free maximum albedo maps, especially over tropical forest regions. Therefore it is quite important to use the accompanying quality flags to guarantee use of the highest-quality albedo values. The annual snow-free minimum albedos from all-quality data (which would tend toward cloud-free values) can form a good basis for the detection of albedo interannual changes. High-quality albedos are strongly suggested for computing maximum albedos, especially in tropical regions. More investigation is needed to provide more high-quality retrievals over tropical regions. [26] The MODIS albedo CMG products capture spatial and temporal patterns that are appropriate for the underlying land covers as expected. Grasslands, open shrublands, and croplands show the highest albedo variations both spatially and temporally among all IGBP vegetation classes. The statistical interannual changes among four consecutive years are very small. Most differences between minimum and maximum shortwave white-sky snow-free albedo among four consecutive years are less than [27] The CMG products also capture the large effects of snow cover on vegetation albedos. However, these effects are different for different vegetation types. Grassland albedos show the highest sensitivity to the snow cover. The response of grasslands albedo to snow cover fraction is 4 to 5 times stronger than for evergreen needleleaf forest canopies. The latitude-averaged maximum shortwave albedos for grassland can be as high as 0.729, but only for evergreen needleleaf forests. By exploring these effects in the MODIS albedo data, climate modelers can also better represent surface albedos in their models. Such investigations will lead to better characterizations of vegetation models in the future. [28] Acknowledgments. This work was funded by EOS-MODIS contract NAS We gratefully acknowledge all MODIS land product technical team members. Their hard work and support are directly responsible for the high quality of MODIS data. Collection 4 CMG albedo data are available from either USGS EDC gateway ( pub/imswelcome/) in HDF format or Boston University ( bu.edu/brdf_albedo/datasets.html) in binary format. References Betts, A. K., and J. H. Ball (1997), Albedo over the boreal forest, J. Geophys. Res., 102, 28,901 28,909. Dickinson, R. E. (1995), Land processes in climate models, Remote Sens. Environ., 51, of 13

13 Friedl, M. A., et al. (2002), Global land cover from MODIS: Algorithms and early results, Remote Sens. Environ., 83, Gao, F., Y. Jin, X. Li, C. B. Schaaf, and A. H. Strahler (2002), Bidirectional NDVI and atmospherically resistant BRDF inversion for vegetation canopy, IEEE Trans. Geosci. Remote Sens., 40(6), Henderson-Sellers, A., and M. F. Wilson (1983), Surface albedo data for climate modeling, Rev. Geophys., 21, Jin, Y., C. B. Schaaf, F. Gao, X. Li, A. H. Strahler, X. Zeng, and R. E. Dickinson (2002), How does snow impact the albedo of vegetated land surfaces as analyzed with MODIS data?, Geophys. Res. Lett., 29(10), 1374, doi: /2001gl Jin, Y., C. B. Schaaf, F. Gao, X. Li, A. H. Strahler, W. Lucht, and S. Liang (2003a), Consistency of MODIS surface bidirectional reflectance distribution function and albedo retrievals: 1. Algorithm performance, J. Geophys. Res., 108(D5), 4158, doi: /2002jd Jin, Y., C. B. Schaaf, C. E. Woodcock, F. Gao, X. Li, A. H. Strahler, W. Lucht, and S. Liang (2003b), Consistency of MODIS surface bidirectional reflectance distribution function and albedo retrievals: 2. Validation, J. Geophys. Res., 108(D5), 4159, doi: /2002jd Lewis, P., and M. J. Barnsley (1994), Influence of the sky radiance distribution on various formulations of the Earth surface albedo, paper presented at 6th International Symposium on Physical Measurements and Signatures in Remote Sensing, Int. Soc. for Photogramm. and Remote Sens., Val d Isere, France. Li, Z., and L. Garand (1994), Estimation of surface albedo from space: A parameterization for global application, J. Geophys. Res., 99, Liang, S. (2001), Narrowband to broadband conversion of land surface albedo. I. Algorithms, Remote Sens. Environ., 76, Liang,S.,H.Fang,M.Chen,C.J.Shuey,C.Walthall,C.Daughtry, J. Morisette, C. Schaaf, and A. Strahler (2002), Validating MODIS land surface reflectance and albedo products: Methods and preliminary results, Remote Sens. Environ., 83, Lucht, W., C. B. Schaaf, and A. H. Strahler (2000), An algorithm for the retrieval of albedo from space using semiempirical BRDF models, IEEE Trans. Geosci. Remote Sens., 38(2), Oleson, K. W., G. B. Bonan, C. B. Schaaf, F. Gao, Y. Jin, and A. H. Strahler (2003), Assessment of global climate model land surface albedo using MODIS data, Geophys. Res. Lett., 30(8), 1443, doi: / 2002GL Roesch, A., M. Wild, H. Gilgen, and A. Ohmura (2001), A new snow cover fraction parameterization for the ECHAM4 GCM, Clim. Dyn., 17, Roesch, A., C. Schaaf, and F. Gao (2004), Use of Moderate-Resolution Imaging Spectroradiometer bidirectional reflectance distribution function products to enhance simulated surface albedos, J. Geophys. Res., 109, D12105, doi: /2004jd Schaaf, C. B., et al. (2002), First operational BRDF, albedo and nadir reflectance products from MODIS, Remote Sens. Environ., 83, Strahler, A. H., W. Lucht, C. B. Schaaf, T. Tsang, F. Gao, X. Li, J.-P. Muller, P. Lewis, and M. J. Barnsley (1999), MODIS BRDF/albedo product: Algorithm theoretical basis document, NASA EOS-MODIS document, v5.0, 53 pp., NASA Goddard Space Flight Cent., Greenbelt, Md. Strugnell, N. C., W. Lucht, and C. B. Schaaf (2001), A global albedo data set derived from AVHRR data for use in climate simulations, Geophys. Res. Lett., 28(1), Tsvetsinskaya, E., C. B. Schaaf, F. Gao, A. H. Strahler, R. E. Dickinson, X. Zeng, and W. Lucht (2002), Relating MODIS-derived surface albedo to soils and rock types over northern Africa and the Arabian Peninsula, Geophys. Res. Lett., 29(9), 1353, doi: /2001gl Vermote, E. F., D. Tanre, J. L. Deuze, M. Herman, and J. J. Morcrette (1997), Second simulation of the satellite signal in the solar spectrum: An overview, IEEE Trans. Geosci. Remote Sens., 35(3), Wang, K., J. Liu, X. Zhou, M. Sparrow, M. Ma, Z. Sun, and W. Jiang (2004), Validation of the MODIS global land surface albedo product using ground measurements in a semidesert region on the Tibetan Plateau, J. Geophys. Res., 109, D05107, doi: /2003jd Wang, Z., X. Zeng, M. Barlage, R. E. Dickinson, F. Gao, and C. Schaaf (2004), Using MODIS BRDF and albedo data to evaluate global model land surface albedo, J. Hydrometeorol., 5, Wanner, W., X. Li, and A. H. Strahler (1995), On the derivation of kernels for kernel-driven models of bidirectional reflectance, J. Geophys. Res., 100, 21,077 21,090. Zhou, L., et al. (2003), Comparison of seasonal and spatial variations of albedo from MODIS and common land model, J. Geophys. Res., 108(D15), 4488, doi: /2002jd R. Dickinson, School of Earth and Atmospheric Sciences, Georgia Institute of Technology, 221 Bobby Dodd Way, Atlanta, GA , USA. (robted@eas.gatech.edu) F. Gao, NASA Goddard Space Flight Center, Code 923, Greenbelt, MD 20771, USA. (fgao@ltpmail.gsfc.nasa.gov) W. Lucht, Potsdam Institut für Klimafolgenforschung, Postfach , D Potsdam, Germany. (wlucht@pik-potsdam.de) A. Roesch, Department of Environmental Sciences, Swiss Federal Institute of Technology Zurich, Winterthurerstrasse 190, CH-8057 Zurich, Switzerland. (andreas.roesch@env.ethz.ch) C. B. Schaaf and A. H. Strahler, Department of Geography and Center for Remote Sensing, Boston University, 675 Commonwealth Avenue, Boston, MA , USA. (schaaf@bu.edu; alan@bu.edu) 13 of 13