Validation of the MODIS Bidirectional Reflectance ... - IEEE Xplore

IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, VOL. 44, NO. 6, JUNE 2006

1555

Validation of the MODIS Bidirectional Reflectance Distribution Function and Albedo Retrievals Using Combined Observations From the Aqua and Terra Platforms Jonathan G. Salomon, Member, IEEE, Crystal B. Schaaf, Member, IEEE, Alan H. Strahler, Member, IEEE, Feng Gao, Member, IEEE, and Yufang Jin

Abstract—We evaluate the performance of the MODerate Resolution Imaging Spectroradiometer (MODIS) Bidirectional Reflectance Distribution Function (BRDF)/Albedo algorithm using observations from MODIS instruments aboard NASA’s Terra (EOS AM-1) and Aqua (EOS PM-1) platforms. This “combined” albedo product is evaluated against continuous field measurements from SURFace RADiation Budget Network (SURFRAD) and Atmospheric Radiation Measurement Southern Great Plains (ARM/SGP) stations, and through an internal analysis of the product’s quality assurance (QA) fields. The combined product is compared to the initial MODIS albedo product, which used observations from the Terra satellite only. During the spring and summer months, the combined product showed a slight improvement over the original Terra-only albedo product, with a root mean square error (RMSE) of 0.0130 and a bias of about 0 02. As with the Terra-only product, accuracy drops during the fall and winter months at some sites. Jin et al. found that increased heterogeneity of validation sites during the fall and winter months is partially responsible for this drop in accuracy. The additional data provided by the Aqua platform changes high-quality albedo estimations only slightly, which underscores the stability of the MODIS algorithm. The most significant benefit of the combined product is a near 50% decrease in lower quality backup algorithm retrievals for the entire globe. A decrease in backup algorithm retrievals improves the overall accuracy of the MODIS albedo product, as it reduces algorithm reliance upon an a priori determination of the underlying surface anisotropy that is not entirely data derived. Index Terms—Remote sensing.

I. INTRODUCTION

S

URFACE albedo is an important parameter of the surface energy budget, and its accurate quantification is of major interest to the global climate modeling community [1]. Until re-

Manuscript received June 29, 2005; revised November 23, 2005. This work was supported in part by NASA, Washington, DC, under Contract NNG04HZ14 as part of the EOS-MODIS Project. J. G. Salomon, C. B. Schaaf, and A. H. Strahler are with the Department of Geography and Environment and the Center for Remote Sensing, Boston University, Boston, MA 02215 USA (e-mail: [email protected]; [email protected]; [email protected]). F. Gao was with Boston University, Boston, MA 20015 USA. He is now with Earth Resources Technology, NASA Goddard Space Flight Center, Greenbelt, MD 20771 USA (e-mail: [email protected]). Y. Jin was with Boston University, Boston, MA 20015 USA. She is now with the Department of Earth System Science, University of California at Irvine, Irvine, CA 92697-3100 USA (e-mail: [email protected]). Digital Object Identifier 10.1109/TGRS.2006.871564

cently, maps of global albedo were not available. The MODerate Resolution Imaging Spectroradiometer (MODIS) Bidirectional Reflectance Distribution Function (BRDF)/Albedo product was specifically designed to fill the need of the climate modeling community for global albedo measurements with an absolute accuracy of 0.02 to 0.05 units [2], [3]. Jin et al. in 2003 evaluated the performance of the MODIS BRDF/Albedo algorithm using only observations from the MODIS instrument aboard NASA’s Terra (EOS AM-1) platform [4]. The purpose of this study is to validate the algorithm using combined observations from both NASA’s Terra and Aqua (EOS PM-1) platforms. This “combined” MODIS albedo product is evaluated through comparisons with independent field measurements and through an investigation of the product’s internal quality assurance (QA) fields. Validation is the process of determining the degree to which a model provides an accurate representation of the real world [5]. While independent field measurements are typically only representative of small areas on the Earth, they remain the primary source of “ground truth” data for validation of remote sensing models. Validation of moderate resolution satellite measurements is difficult because a single satellite measurement can measure energy from a very large area relative to field measurements [6]. Most regions of the globe would require a large number of field samples to capture the mean and variance for the area covered by a single satellite pixel [7]. This creates additional logistic problems for albedo validation. Unlike more static satellite products, such as land cover, albedo varies constantly with solar geometry, viewing geometry, and atmospheric conditions [7]. Thus, albedo validation requires long-term simultaneous field and satellite observations. Currently, there are few field sites in the world with tower albedometers that can be used for albedo validation, and even fewer that provide a large number of field samples distributed across a single region. An alternative strategy to large field samples is to collect measurements from relatively homogenous surface types where a single field measurement can more accurately represent the mean albedo of a pixel at the satellite scale [7]. II. METHODS A. Surface Albedo Measurements The SURFace RADiation Budget Network (SURFRAD), part of the worldwide Baseline Surface Radiation Net-

0196-2892/$20.00 © 2006 IEEE

1556


work (BSRN), provides validation data in the United States for satellite-derived estimates of the surface energy budget, including measurements of surface albedo (see http://www.srrb.noaa.gov/surfrad/index.html) [8]. SURFRAD has been providing measurements of the surface radiation budget since 1995 [9]. Six SURFRAD stations were used in this analysis. Grasslands dominate the station at the Fort Peck Tribes Reservation in Montana. The Table Mountain Test Facility of the National Oceanic and Atmospheric Administration’s (NOAA) Air Resources Laboratory station in Colorado overlooks a mixture of grasslands and shrublands. The Pennsylvania State University’s agricultural research station, and the Bondville, IL, station are both situated in agricultural land. The station near Goodwin Creek, MS, is located on rural pastureland, and the station near Desert Rock, NV, overlooks sparse vegetation [4]. These sites were chosen to represent the diverse climates of the United States [9]. However, the land cover types associated with these stations are typically grasslands or agricultural lands rather than forests or dense shrublands. Each station is located in a large, relatively homogenous region that allows for comparisons to coarse-scale satellite measurements [4]. In addition to the six SURFRAD stations, the Department of Energy’s (DOE’s) Atmospheric Radiation Measurement Program: Southern Great Plains Facility (ARM/SGP) also provides field measurements of the surface radiation budget (see http://www.arm.gov/) [9]. Two ARM/SGP stations were chosen for this validation experiment: station #1 at the Central Facility in Lamont, OK, and station #15 of the Extended Facility in Ringwood, OK [10]. Each SURFRAD and ARM/SGP station contains two broadband (0.28 to 3 m) pyranometers: a skyward looking pyranometer mounted on a horizontal platform, and a downward looking pyranometer mounted on a nearby 10-m tower. The downward-looking pyranometer has a restricted 45% field of view and observes an area roughly 18 m in diameter on the ground. Combining radiation measurements from the upward and downward looking pyranometers provides an accurate estimate of surface albedo. A normal incidence pyrheliometer mounted on an automatic sun tracker measures direct solar radiation incident upon the site, and a shaded pyranometer riding on top of the sun tracker measures diffuse solar radiation [8]. Estimates of cloud fraction as viewed from the skyward-looking pyranometer were measured for certain dates [11]. The SURFRAD and ARM/SGPP stations record data continuously, with average radiation measurements taken once every 3 min. The Clouds and the Earth’s Radiant Energy System (CERES) ARM Validation Experiment (CAVE) averages SURFRAD and ARM/SGP radiometric data into 30-min intervals (web site: http://snowdog.larc.nasa.gov/cave/) [12]. The CAVE data set contains averaged radiation measurements for numerous validation sites, including SURFRAD and ARM/SGP [13]. CAVE measurements have the advantage of a smaller data volume, and the 30-min time interval provides a more than adequate temporal resolution for validation of the MODIS albedo product. A yearly time series of surface albedo measurements can be quickly tabulated from the CAVE database and compared directly to the MODIS albedo product.

B. MODIS-Derived Albedo Measurements The MODIS BRDF/Albedo algorithm uses a kernel-driven semiempirical RossThick-LiSparse-Reciprocal (RTLSR) BRDF model. The BRDF is defined as the ratio of the radiance scattered by a surface into a specified direction to the unidirectional (collimated) irradiance incident on a surface. Satellite measurements estimate the BRDF through integration of radiance measurements over multiple and viewing geometries [14]. Using MODIS directional reflectance measurements as inputs, the RTLSR model describes surface reflectance in any direction as a function of illumination and view angles at a particular wavelength. Surface albedo describes the total fraction of incoming solar energy reflected at a given point and time. This intrinsic albedo, which describes the characteristics of the surface, can be reconstructed from the BRDF model [15]–[17]. The MODIS albedo product (reprocessed version 4) generates operationally a set of two albedo measurements for each point on the ground. The two measurements represent components of the actual surface albedo. The first component is the directional hemispherical reflectance, or “black-sky albedo,” which is derived by integrating total surface reflectance for any one direction of illumination. For the operational MODIS albedo product, the illumination direction is the solar zenith angle at local solar noon (LSN). If other solar zenith angles are needed, the MODIS BRDF model parameters product can be used to compute black sky albedos at any angle. This component represents the scattering of direct-beam radiation from the sun, omitting incident light from the rest of the sky. The second component is the diffuse bihemispherical reflectance, or “white-sky albedo,” which is derived by integrating surface reflectance for uniform illumination incident upon the surface from all directions. This component of albedo represents scattered radiation from diffuse skylight, omitting direct beam radiation. In a perfectly overcast sky, where illumination could be considered isotropic, white-sky albedo is the sole component of albedo [15]. Black-sky and white-sky albedo represent the extremes of a completely clear and strongly turbid atmosphere, respectively [18]. The actual albedo measurement on the ground, or “blue-sky albedo,” can, therefore, be estimated as a sum of black-sky and white-sky albedos weighted by the proportions of direct and diffuse solar radiation arriving at the ground. Optical type and depth measurements can also be used to compute the proportions of beam and diffuse radiation [19], [20]. It is recognized that this approach ignores a multiple scattering component of albedo and assumes solar illumination is uniformly distributed; however, multiple scattering and anisotropic solar irradiance have been estimated to represent only a few percent of total blue-sky albedo under clear conditions [21], [16]. The MODIS albedo product provides these black-sky and white-sky albedos at 1-km resolution for MODIS bands one through seven, and for three broad bands (0.3–0.7, 0.7–5.0, and 0.3–5.0 m) [15]. The shortwave broadband albedo (0.3– 5.0 m) was used for this analysis, as it is the closest spectral match to the broadband pyranometers used by SURFRAD and


1557

Fig. 1. Time series of surface albedo at LSN for the SURFRAD station at Fort Peck, MT, for the years 2003 and 2004. The dotted line is daily surface albedo at the station. The solid line is a 16-day average of surface albedo matching the sampling scheme used by the MODIS BRDF/albedo algorithm. Filled and open shapes indicate retrievals using the main and backup algorithms, respectively. Circles are from the Terra-only MODIS albedo product, and triangles are from the combined albedo product. Missing data is omitted from the plot. Large changes in MODIS albedo during snow periods (changes in surface albedo greater than 0.5) are due to reversals of the QA snow flag (see text).

Fig. 2. Time series of surface albedo at LSN for the SURFRAD station at Table Mountain, CO, for the years 2003 and 2004. Setup is the same as Fig. 1. Large changes in MODIS albedo during snow periods (changes in surface albedo greater than 0.5) are due to reversals of the QA snow flag (see text).

ARM. The MODIS BRDF/albedo algorithm uses 16 days of sequential multiangle observations from the cloud-screened, snow-flagged, and atmospherically-corrected MODIS surface reflectance product (MOD09) to estimate the RTLSR BRDF model parameters. Due to the wide swath width of MODIS and its frequent overpasses at higher latitudes, multiple observations can be acquired for many pixels every day for each MODIS instrument [19]. MODIS surface reflectances are also screened for outliers before further processing. If the majority of observations for a 16-day period are recorded as snow-covered, then the algorithm uses only snow-covered observations for the parameter retrieval. Conversely, the algorithm uses only snow-free observations for parameter retrieval if the majority of MOD09 observations during the period are snow-free. A full retrieval of the parameters for the RTLSR BRDF model is attempted if seven or more observations survive the screening process. A backup algorithm retrieval is used if less than seven observations survive the screening process or if a robust full retrieval can not be made. The backup algorithm uses a priori estimates of the BRDF shape for each pixel around the globe, and fits these predetermined shapes to any MOD09 observations in order to estimate the model parameters. These a priori

estimates come from a global database of previously obtained high-quality MODIS BRDF/albedo retrievals for that pixel. Whereas backup algorithm retrievals can be quite robust, they are flagged as lowest quality results. A fill value is stored if no observations are retrieved during a 16-day interval. Information on snow and snow-free retrievals, full retrievals, and backup algorithm retrievals are recorded for each pixel in the extensive quality information embedded within the product [19]. C. Comparison of Surface Albedo With MODIS Albedo This study uses MODIS observations and field measurements acquired from the SURFRAD and ARM/SGP stations for the years 2003 and 2004, when complete years of MODIS data from both the Terra and Aqua platforms were first available. Any dates with missing or corrupted field data, or completely missing MODIS data were omitted from this analysis. Surface albedo was derived at each field site using 30-min averaged upwelling and downwelling surface radiation measurements at local solar noon. Daily surface albedos were filtered for clouds to better simulate the cloud-screened observations from MODIS. Any station measurement with a cloud fraction greater than 0.3 at

1558


Fig. 3. Time series of surface albedo at LSN for the SURFRAD station at Pennsylvania State University’s agricultural research farm, PA, for the years 2003 and 2004. Setup is the same as Fig. 1.

Fig. 4. Time series of surface albedo at LSN for the SURFRAD station at Bondville, IL, for the years 2003 and 2004. Setup is the same as Fig. 1.

LSN was removed from analysis. When estimates of cloud fraction were unavailable, any measurement with a direct/diffuse ratio of solar radiation less than three was removed. Daily surface albedos were then averaged over the same 16-day intervals used for the MODIS Terra-only and combined Terra-Aqua albedo retrievals. If a snow retrieval was performed by the combined MODIS albedo product, then any field measurement of albedo less than 0.4 was omitted from the 16-day field average. If the combined albedo product performed a snow-free retrieval, then than any field measurement of albedo greater than 0.4 was omitted from the field average. (If snow retrieval information from the combined product was unavailable, then snow retrieval information from the Terra-only albedo product was used.) Station measurements of the direct/diffuse ratio were used to determine the proportion of black-sky and white-sky components needed to estimate a MODIS blue-sky albedo. The 16-day averaged surface albedos from the station pyranometers were then compared to blue-sky albedos derived from the MODIS product. D. Global QA Comparisons In addition to utilizing field measurements for validation, performance of the combined Terra and Aqua MODIS albedo product was evaluated globally by comparing it to the Terra-only MODIS albedo product previously examined [4].

We analyzed the performance of these products through a comparison of the embedded QA records for each pixel, including whether the full retrieval or backup retrieval algorithms were used. Full retrievals and backup algorithm retrievals were tabulated globally for both products. Retrieval results were used to evaluate changes in the algorithm’s performance resulting from the use of additional looks and solar geometries provided by the Aqua platform. III. RESULTS A. Surface Albedo Time Series Figs. 1–8 compare the yearlong trend of surface albedo for field sites and the MODIS albedo product. The dotted line depicts daily surface albedo at LSN for the station. Daily albedos can fluctuate widely, especially during the fall and winter months. Analysis by Jin et al. revealed that high peaks surface albedos are typically caused by snow cover on the ground. Distinct low peaks albedos are caused by rainfall, which darken the soil [22]. These rain events have little effect on 16-day averaged surface albedos [23]. The solid line is a 16-day average of surface albedo simulating the sampling initially used by the MODIS BRDF/albedo algorithm. Station averages use only snow-covered or snow-free looks, depending on the snow flag in the MODIS albedo product embedded QA. The snow flag of the Terra-only MODIS albedo product


Fig. 5.

Time series of surface albedo at LSN for the SURFRAD station at Goodwin Creek, MS, for the years 2003 and 2004. Setup is the same as Fig. 1.

Fig. 6.

Time series of surface albedo at LSN for the SURFRAD station at Desert Rock, NV, for the years 2003 and 2004. Setup is the same as Fig. 1.

Fig. 7.

Time series of surface albedo at LSN for the Central Facility (station #1) of ARM/SGP for the years 2003 and 2004. Setup is the same as Fig. 1.

is used for this analysis. Averaging and snow filtering makes the 16-day average and MODIS albedo values considerably smoother than daily albedo values. Filled circles represent full retrievals and open circles represent backup algorithm retrievals for the Terra-only MODIS albedo product. Filled inverted triangles represent full retrievals, whereas open inverted triangles represent backup algorithm retrievals of the combined MODIS albedo product. The most notable feature of this analysis is the stability of the MODIS BRDF/albedo algorithm across all terrains and dates. Of the observations that changed between the Terra-only

1559

and combined products, the shift in albedos was less than 10%, and in many cases, the shift was small or nonexistent. Albedo was relatively stable even when the MODIS BRDF/albedo algorithm switched from backup algorithm retrieval to full retrieval. Large variations in MODIS albedo retrievals during ), such as those snow periods (changes in surface albedo observed during the winter months at Fort Peck, Fig. 1, and Table Mountain, Fig. 2, are due to a switch of the snow flag in the albedo product embedded QA. Additional observations from the Aqua platform were enough to shift the majority of observations during the 16-day period from snow to snow-free,

1560


Fig. 8. Time series of surface albedo at LSN for Extended Facility #15 of ARM/SGP for the years 2003 and 2004. Setup is the same as Fig. 1.

or vice-versa. It should be noted, however, that changes in the snow flag either improved the MODIS albedo retrieval or kept it within range of daily albedo values observed at the station. The MODIS albedo product performed admirably over grasslands and shrublands. MODIS albedos agreed closely with the average albedo at Fort Peck, Fig. 1, Table Mountain, Fig. 2, and the sparely vegetated site at Desert Rock, Fig. 6. At Fort Peck, surface albedo is high during the winter months when snow is on the ground. MODIS and field albedos drop slightly in late May (Julian day 129). Albedo remains stable until mid-July when foliage is at its peak. Surface albedo steadily increases from August onward (after Julian day 209) as senescence causes visible albedo to increase. By November (Julian day 321), snow has settled in and albedos are once again high. MODIS albedo is also in close agreement at the Table Mountain site, Fig. 2. MODIS albedos closely track the dip in surface albedo during the summer, as well as the slight rise in surface albedo as senescence occurs. MODIS albedo does not capture many of the sporadic snow events that occur at Table Mountain, providing only snow-free backup retrievals for many of the 16-day periods. However, the MODIS albedos are still within the range of daily surface albedos during the winter months. At the ARM/GSP Central Facility and Extended Facility, Figs. 7 and 8, MODIS albedos drop slightly in early winter and late spring but are in close agreement with field albedos for the rest of the year. The MODIS algorithm performed very well for the spring and summer moths at the rest of the SURFRAD sites, but less well during the fall and winter months at the agricultural sites of Penn State University, Fig. 3, and Bondville, Fig. 4, and at the pasturelands of Goodwin Creek, Fig. 5. During the fall and winter, MODIS albedos are 0.07 lower than station albedos at Goodwin Creek, and 0.15 lower than station albedos at the agricultural sites. Jin et al. suggested that the discrepancy may be explained by increased subpixel heterogeneity in the fall and winter months through various processes such as nonuniform patterns of snowmelt at these sites [4]. An assessment of the Terra-only MODIS albedo product by Stroeve et al. concluded that MODIS albedos were largely accurate over regions of homogenous snow [24], and discrepancies have been found during months of fall when no snow was present. Further study is needed to fully categorize the drop in accuracy over heterogeneous surfaces during the fall and winter months. The

Fig. 9. Scatter plot of surface albedo (at LSN) from the MODIS Terra-only albedo product versus field measurements for the spring and summer of 2003 and 2004. Open symbols indicate backup algorithm retrievals, and filled symbols indicate full retrievals (of all quality). The solid line is the one-to-one line and the dashed lines are 0:02 and 0:05 units.

6

6

effect of scaling on MODIS albedos will be discussed further in the next section. To explore the accuracy of the MODIS algorithm at these sites, scatter plots were made of MODIS albedo versus field albedo at all sites from April (Julian day 97) to September (Julian day 257); typically snow-free periods of the year. In Fig. 9, MODIS albedos are derived from the Terra-only albedo product. In Fig. 10, MODIS albedos are derived from the combined albedo product. A root mean square error (RMSE) were found with the of 0.014 and a bias of around Terra-only albedo product. The combined product showed a slight improvement, with an RMSE of 0.013 and a bias of . The minor change in bias and RMSE further around demonstrates the overall stability of the MODIS algorithm.


Fig. 10. Scatter plot of surface albedo (at LSN) from the MODIS combined albedo product versus field measurements for the spring and summer of 2003 and 2004. Open symbols indicate backup algorithm retrievals, and filled symbols indicate full retrievals (of all quality). The solid line is the one-to-one line and the dashed lines are 0:02 units and 0:05 units.

6

6

B. Scaling Effects on Surface Albedo Validation Jin et al. previously examined the scaling effects on albedo validation in 2003. Their analysis used Landsat observations to simulate 30-m albedo data for the area of the MODIS pixel overlooking the site. Jin et al. found that simulated albedos for the area immediately surrounding the station were relatively homogenous in July and more heterogeneous in September. They found that simulated albedos in the 30-m pixel containing the station were 0.027 units higher than the mean albedo for the entire 1-km pixel for the month of September [4]. Their discovery of subpixel heterogeneity is further corroborated by a recent analysis of land cover surrounding the Bondville station. The BigFoot project created 30-m resolution land cover maps for its field sites [25], [26]. Fig. 11 shows the land cover map for the region around the Bondville station. While soybean crops dominate the area immediately surrounding the station (marked by the white cross in Fig. 11), the region surrounding the tower is not completely homogenous. A mixture of soybean, corn, and urban land covers are present within 1 km of the tower (marked by the box in Fig. 11). Subpixel heterogeneity around the Bondville station may explain some of the seasonal discrepancy at the Bondville station. Future work will analyze the spatial heterogeneity in detail. C. Global QA Comparison Counts of full retrievals and backup algorithm retrievals were tabulated globally for both albedo products during a sample 16-day period from April 22, 2004 to May 8, 2004. Due to the large size of a global albedo data set at 1-km resolution, only one 16-day period was investigated. The period beginning in April

1561

Fig. 11. IGBP land cover map of the region surrounding the Bondville, IL, SURFRAD station. The image is skewed due to the sinusoidal projection. The light color indicates urban/built-up land, blue indicates water, orange indicates grasslands, dark green indicates soybean fields, and light green indicates cornfields. The white cross indicates the exact position of the SURFRAD station, and the square represents the location of the overlying 1-km MODIS pixel [25].

22, 2004 was selected due to the relatively cloud-free conditions globally during this time. Only the equatorial regions (10 N–10 S) and extreme southern latitudes (70 S–80 S) were consistently cloud covered during the 16-day time period. The results of this analysis are presented in Fig. 12. From 20 N northward, the combined albedo product shows a substantial increase in the number of full retrievals. This increase is accompanied by a similar decrease in backup algorithm retrievals as shown in Fig. 13. Note that overall there is a reduction of pixels with increasing latitude. From 20 N to 20 S, there are overall more backup algorithm retrievals than full retrievals. The MODIS Albedo product has few high-quality retrievals over tropical forest regions [27], which is primarily an effect of equatorial cloud cover. Cloud cover in the tropics can severely limit the number of looks available to the MODIS BRDF/albedo algorithm [16]. The slight increase in backup algorithm equatorial retrievals is due to a switch of some pixels from fill values to the backup algorithm. In other words, the Terra-only albedo product obtained no good looks during the 16-day period for most equatorial pixels, and no albedo retrieval was attempted for them. However, one to six looks were obtained during the same 16-day period using the combined albedo product, and the additional looks allowed for a backup algorithm retrieval. A substantial increase in full retrievals is once again observed from 20 S to 40 S. The decrease in pixels from 40 S to 60 S is due to the lack of large land masses in this region. The large spike in fill values below 60 S is due to a lack of clear MODIS observations over Antarctica during the sample 16-day period. Overall, the combined albedo product supplied enough additional information to increase the number of full retrievals by 50% globally, as illustrated in Fig. 14. The loss of observations due to cloud cover is the main reason for the use of the backup algorithm [28], and additional looks from the aqua

1562


Fig. 12. Distribution of full retrievals, backup algorithm retrievals, and fill values with latitude for the 16-day period from April 22, 2004 to May 8, 2004. Dotted lines indicate retrievals from the Terra-only albedo product. Solid lines denote retrievals from the combined albedo product. Note the reduction in backup retrievals and increase in full retrievals for all latitudes except near the equator.

Fig. 13. Change in full retrievals, backup algorithm retrievals, and fill values with latitude for the 16-day period from April 22, 2004 to May 8, 2004. Note the consistent increase of full retrievals with a complimentary decrease in backup algorithm retrievals. The slight increase in backup algorithm retrievals near the equator and extreme southern latitudes is driven by a complimentary decrease in fill values.

platform is the primary reason for the reduction in backup algorithm retrievals. Since the backup algorithm relies upon an a priori determination of the underlying surface anisotropy that is not derived from data during the 16-day period, backup algorithm retrievals are flagged as a lower quality result. Thus, full retrievals reduce the uncertainty of the albedo estimate [15].

IV. CONCLUSION We evaluated the performance of the MODIS BRDF/Albedo algorithm using combined observations from both NASA’s Terra and Aqua platforms. The “combined” albedo product was evaluated against continuous field measurements, through an


1563

Fig. 14. (Top) Distribution of full retrievals, backup algorithm retrievals, and fill values across the globe for the Terra-only albedo product for the 16-day period from April 22, 2004 to May 8, 2004. Green indicates full retrievals, red indicates backup algorithm retrievals, black is a fill value, and blue is water. Fill values over the north and south poles are due to a lack of data from the satellite over these regions during the testing period. (Bottom) Distribution of retrievals for the combined albedo product for the same 16-day period. Note the significant increase in full retrievals (green) when compared to the Terra-only albedo product.

internal analysis of the product’s QA fields, and against the original albedo product that uses observations from the Terra satellite only. During the spring and summer months, both MODIS albedo products met the absolute accuracy requirement of 0.02 for all eight field stations. The Terra-only albedo product had . The combined an RMSE of 0.014, and a bias of around product showed a slight improvement, with an RMSE of 0.013 . Accuracy dropped during the fall and a bias of around and winter months at some sites. Jin et al. found that increased heterogeneity of the validation sites during the winter months is partially responsible for the drop in accuracy [4]. Our anal-

ysis of land cover at the Bondville site corroborates their findings. An assessment of the Terra-only MODIS albedo product by Stroeve et al. concluded that MODIS albedos were largely accurate over regions of homogenous snow [24]. Further study is needed to fully categorize the drop in accuracy over heterogeneous surfaces during the fall and winter months. Comparisons of the Terra-only albedo product with the combined albedo product demonstrate the stability of the MODIS BRDF/albedo algorithm. The additional looks and solar geometries provided by the Aqua platform change the product’s albedo estimations only slightly. Albedo changed only slightly when

1564


the MODIS BRDF/albedo algorithm switched from a backup algorithm retrieval to a full retrieval, underscoring the stability of backup algorithm retrievals as well. The most significant benefit of the combined product is a 50% increase in full retrievals across the globe, which is due to the near doubling of observations at every pixel with the additional data from the Aqua platform. An increase in full retrievals improves the accuracy of the MODIS albedo product, as backup algorithm retrievals rely upon an a priori determination of the underlying surface anisotropy that is not derived from timely data and is thus considered a lower quality result. This study is a preliminary evaluation of the combined BRDF/albedo product. Complete validation will require several much larger field campaigns, with measurements taken at multiple scales of resolution by field instruments, aircraft, and satellites. Spatial coverage of the analysis will have to increase from the US to the entire globe. We intend to continue validation at additional locations and at additional scales, as more data from field tower sites and campaigns becomes available. ACKNOWLEDGMENT The authors would like to thank J. C. F. Hodges for his help with subpixel evaluation of the MODIS Albedo product. They would also like to thank CAVE for supplying the SURFRAD and ARM datasets, and the BigFoot project for providing land cover maps of the Bondville, IL, tower site. They are dependent on the datasets supplied by the fine work of the SURFRAD, ARM, BigFoot, and CAVE science teams. SURFRAD data is made available through NOAA’s Air Resources Laboratory/Surface Radiation Research Branch. ARM data is made available through the U.S. Department of Energy as part of the Atmospheric Radiation Measurement Program. REFERENCES [1] R. E. Dickinson, “Land processes in climate models,” Remote Sens. Environ., vol. 51, pp. 27–38, 1995. [2] A. Henderson-Sellers and M. F. Wilson, “Surface albedo data for climatic modeling,” Rev. Geophys. Space Phys., vol. 21, pp. 1743–1778, 1983. [3] P. J. Sellers, “Remote sensing of the land surface for studies of global change,” in NASA/GSFC Int. Satellite Land Surface Climatology Project Report. Greenbelt, MD., 1993. [4] Y. Jin, C. B. Schaaf, C. E. Woodcock, F. Gao, X. Li, A. H. Strahler, W. Lucht, and S. Liang, “Consistency of MODIS surface bidirectional reflectance distribution function and albedo retrievals, 2. Validation,” J. Geophys. Res., vol. 108, no. D5, 4159, 2003. [5] Department of Defense Dictionary of Military and Associated Terms, Nov. 2004. JP 1-02. [6] Y. Tian, C. E. Woodcock, Y. Wang, J. L. Privette, N. V. Shabanov, L. Zhou, Y. Zhang, W. Buermann, J. Dong, B. Veikkanen, T. Häme, K. Andersson, M. Ozdogan, Y. Knyazikhin, and R. B. Myneni, “Multiscale analysis and validation of the MODIS LAI product I. Uncertainty assessment,” Remote Sens. Environ., vol. 83, pp. 414–430, 2002. [7] W. Lucht, A. H. Hyman, A. H. Strahler, M. J. Barnsley, P. Hobson, and J.-P. Muller, “A comparison of satellite-derived spectral albedos to ground-based broadband albedo measurements modeled to satellite spatial scale for a semidesert landscape,” Remote Sens. Environ., vol. 74, pp. 85–98, 2000. [8] J. A. Augustine, J. J. DeLuisi, and C. N. Long, “SURFRAD—A national surface radiation budget network for atmospheric research,” Bull. Amer. Met. Soc., vol. 81, no. 10, pp. 2341–2358, 2000. [9] J. J. DeLuisi, J. A. Augustine, C. Cornwall, and G. Hodges, “Contrasting ARM’s SRB measurements with six SURFRAD stations,” in Proc. 9th ARM Science Team, San Antonio, TX, 1999, pp. 1–6.

[10] T. P. Ackerman, A. D. Del Genio, G. M. McFarquhar, R. G. Ellingson, P. J. Lamb, R. A. Ferrare, C. N. Long, S. A. Klein, and J. Verlinde, Atmospheric Radiation Measurement Program Science Plan: Current Status and Future Directions of the ARM Science Program. Washington, DC: Office of Science, Office of Biological and Environmental Research, United States Department of Energy, 2004. DOE/ER-ARM-0402. [11] C. N. Long and T. P. Ackerman, “Identification of clear skies from broad-band pyranometer measurements and calculation of downwelling short-wave cloud effects,” J. Geophys. Res., vol. 105, no. D12, pp. 15609–15626, 2000. [12] D. A. Rutan, F. G. Rose, N. M. Smith, and T. P. Charlock, “Validation data set for CERES surface and atmospheric radiation budget (SARB),” WCRP/GEWEX Newsletter, vol. 11, no. 1, pp. 11–12, 2001. [13] D. A. Rutan, F. G. Rose, N. Smith, and T. P. Charlock, “CERES ARM validation experiment,” in Proc. 11th ARM Science Team Meeting, Atlanta, GA, 2001, pp. 1–4. [14] L. D. Girolamo, “Generalizing the definition of the bidirectional reflectance distribution function,” Remote Sens. Environ., vol. 88, pp. 479–482, 2003. [15] Z. Wang, X. Zeng, M. Barlage, R. E. Dickinson, F. Gao, and C. B. Schaaf, “Using MODIS BRDF and albedo data to evaluate global model land surface albedo,” J. Hydrometeorol., vol. 5, pp. 3–13, 2004. [16] W. Lucht, C. B. Schaaf, and A. H. Strahler, “An algorithm for the retrieval of albedo from space using semiempirical BRDF models,” IEEE Trans. Geosci. Remote Sens., vol. 38, no. 2, pp. 977–997, Feb. 2000. [17] W. Wanner, A. H. Strahler, B. Hu, P. Lewis, J.-P. Muller, X. Li, C. B. Schaaf, and M. J. Barnesley, “Global retrieval of bidirectional reflectance and albedo over land from EOS MODIS and MISR data: Theory and algorithm,” J. Geophys. Res., vol. 102, no. D14, pp. 17 143–17 161, 1997. [18] W. Lucht, “Expected retrieval accuracies of bidirectional reflectance and albedo from EOS-MODIS and MISR angular sampling,” J. Geophys. Res., vol. 103, no. D8, pp. 8763–8778, 1998. [19] C. B. Schaaf, F. Gao, A. H. Strahler, W. Lucht, X. Li, T. Tsang, N. C. Strugnell, X. Zhang, Y. Jin, J. P. Muller, P. Lewis, M. Barnsley, P. Hobson, M. Disney, G. Roberts, M. Dunderdale, C. Doll, R. d’Entremont, B. Hu, S. Liang, and J. L. Privette, “First operational BRDF albedo nadir reflectance products from MODIS,” Remote Sens. Environ., vol. 83, pp. 135–148, 2002. [20] Y. Luo, A. P. Trishchenko, R. Latifovic, K. Khlopenkov, and Z. Li, “BRDF/Albedo retrievals from the Terra and Aqua MODIS systems at 500-m spatial resolution and 10-day intervals,” Proc. SPIE Remote Sens. Atmospheric Pollution Monitoring and Control, vol. 5549, pp. 194–201, 2004. [21] P. Lewis and M. J. Barnsley, “Influence of the sky radiance distribution on various formulations of the earth surface albedo,” in Proc. Conf. Phys., Measures, Signals, Val d’Isere, France, 1994, pp. 707–715. [22] M. J. Barnsley, P. D. Hobson, A. H. Hyman, W. Lucht, J.-P. Muller, and A. H. Strahler, “Characterizing the spatial variability of broadband albedo in a semidesert environment for MODIS validation,” Remote Sens. Environ., vol. 74, pp. 58–68, 2000. [23] M. J. Chopping, “Large-scale BRDF retrieval over new mexico with a multiangular NOAA AVHRR dataset,” Remote Sens. Environ., vol. 74, pp. 163–191, 2000. [24] J. Stroeve, J. E. Box, F. Gao, S. Liang, A. Nolin, and C. Schaaf, “Accuracy assessment of the MODIS 16-day albedo product for snow: Comparisons with Greenland in situ measurements,” Remote Sens. Environ., vol. 94, pp. 46–60, 2005. [25] W. B. Cohen, T. K. Maiersperger, Z. Yang, S. T. Gower, D. P. Turner, W. D. Ritts, M. Berterretche, and S. W. Running, “Comparisons of land cover and LAI estimates derived from ETM+ and MODIS for four sites in North America: A quality assessment of 2000/2001 provisional MODIS products,” Remote Sens. Environ., vol. 88, pp. 233–255, 2003. [26] W. B. Cohen, T. K. Maiersperger, and D. Pflugmacher. (2004) BigFoot Land Cover Surfaces for North and South American Sites, 2000–2003 [Online]. Available: http://www.daac.ornl.gov [27] F. Gao, C. B. Schaaf, A. H. Strahler, A. Roesch, W. Lucht, and R. Dickinson, “MODIS bidirectional reflectance distribution function and albedo climate modeling Grid products and the variability of albedo for major global vegetation types,” J. Geophys. Res., vol. 110, no. D01104, 2005. [28] Y. Jin, C. B. Schaaf, F. Gao, X. Li, A. H. Strahler, W. Lucht, and S. Liang, “Consistency of MODIS surface bidirectional reflectance distribution function and albedo retrievals, 1. Algorithm Performance,” J. Geophys. Res., vol. 108, no. D5, 4158, 2003.


Jonathan G. Salomon (M’06) received the B.A. degree in geology from Brown University, Providence, RI, in 2002 and the M.A. degree in geography from Boston University, Boston, MA, in January, 2006. He is currently a Research Assistant at the Department of Geography and Environment and the Center for Remote Sensing, Boston University, Boston, MA. His current research focuses on validation of the BRDF/albedo data products for MODIS.

Crystal B. Schaaf (M’92) received the S.B. and S.M. degrees in meteorology from the Massachusetts Institute of Technology, Cambridge, in 1982, the M.L.A. degree in archaeology from Harvard University, Cambridge, MA, in 1988, and the Ph.D. degree in geography from Boston University, Boston, MA, in 1994. She is currently a Research Associate Professor of Geography and Researcher in the Center for Remote Sensing, Boston University, and is a member of the MODIS and NPP science teams. Her research interests cover remote sensing of land surfaces and clouds, with particular emphasis on surface anisotropy and albedo.

Alan H. Strahler (M’86) received the B.A. and Ph.D. degrees in geography from The Johns Hopkins University, Baltimore, MD, in 1964 and 1969, respectively. He is currently Professor of Geography and Researcher in the Center for Remote Sensing, Boston University, Boston, MA. He has held prior academic positions at Hunter College of the City University of New York, at the University of California, Santa Barbara, and at the University of Virginia. Originally trained as a biogeographer, he has been actively involved in remote sensing research since 1978. He has been a Principal Investigator on numerous NASA contracts and grants. His primary research interests are directed toward modeling the bidirectional reflectance distribution function (BRDF) of discontinuous vegetation covers and retrieving physical parameters describing ground scenes through inversion of BRDF models using directional radiance measurements. He is also interested in the problem of land cover classification using multitemporal, multispectral, multidirectional, and spatial information as acquired in reflective and emissive imagery of the Earth’s surface. Dr. Strahler was awarded the AAG/RSSG Medal for Outstanding Contributions to Remote Sensing in 1993 and the honorary degree Doctorem Scientarum Honoris Causa from the Université Catholique du Louvain, Belgium, in 2000. He was also honored as a Fellow of the American Association for the Advancement of Science in 2003.

1565

Feng Gao (M’99) received the B.A. degree in geology and the M.S. degree in remote sensing from Zhejiang University, Hangzhou, China, in 1989 and 1992, respectively, the Ph.D. degree in geography from Beijing Normal University, Beijing, China, in 1997, and the M.S. degree in computer science from Boston University, Boston, MA, in 2003. From 1992 to 1998, he was a Research Assistant with the Nanjing Institute of Geography and Limnology, Chinese Academy of Science, Nanjing, China. From 1998 to 2004, he was a Research Associate Professor with the Department of Geography and a Researcher in the Center for Remote Sensing, Boston University. He joined the Goddard Space Flight Center, Greenbelt, MD, through a contract with Earth Resources Technology (ERT), Inc. in August 2004. His research interests include remote sensing modeling and retrieving vegetation parameters through inversion of remote sensing models.

Yufang Jin received the B.S. and M.S. degrees in atmospheric physics and environmental sciences from Peking University, Beijing, China, in 1995 and 1998, respectively, and the Ph.D. degree in geography from Boston University, Boston, MA, in 2002. She is currently an Assistant Researcher in the Department of Earth System Science, University of California, Irvine, working on the integration of remote sensing data with biogeochemical models for carbon studies. She was an Assistant Research Scientist in the Department of Geography, University of Maryland, College Park, from 2003 to 2005. Her previous work includes the surface bidirectional reflectance/albedo retrievals and burned area detection with MODIS observations. She is interested in Earth’s radiative energy budget, global carbon cycle, remote sensing of the biosphere and the atmosphere, and data assimilation in climate models.