ClickHere

for

FullArticle

Comparison of MISR and MODIS land surface albedos:Methodology

M. Taberner,1 B. Pinty,1,2 Y. Govaerts,3 S. Liang,4 M. M. Verstraete,1 N. Gobron,1

and J.‐L. Widlowski1

Received 15 June 2009; revised 17 September 2009; accepted 26 October 2009; published 3 March 2010.

[1] The broadband white sky surface albedo (bihemispherical reflectance) productsavailable from the Moderate Resolution Imaging Spectroradiometer (MODIS) arecompared at regional and continental scales with similar products generated from theMultiangle Imaging Spectroradiometer (MISR) land surface bidirectional reflectancefactor (BRF) parameters. This paper describes the methodology applied to derive MISRwhite sky albedos over four spectral broadbands of interest, namely, 0.3–0.7 mm,0.4–1.1 mm, 0.7–3.0 mm, and 0.3–3.0 mm, as well as an evaluation of the strategy adopted tocompare the MODIS and MISR products. The results are very encouraging since the twodata sets show very good statistical agreement over large areas and over a full year ofmeasurements, despite the many differences that exist in the suite of algorithms applied toretrieve these surface quantities from each of these instruments separately.

Citation: Taberner, M., B. Pinty, Y. Govaerts, S. Liang, M. M. Verstraete, N. Gobron, and J.‐L. Widlowski (2010), Comparisonof MISR and MODIS land surface albedos: Methodology, J. Geophys. Res., 115, D05101, doi:10.1029/2009JD012665.

1. Introduction

[2] Modern sensors such as the Moderate ResolutionImaging Spectroradiometer (MODIS) and the MultiangleImaging Spectroradiometer (MISR) offer a unique oppor-tunity to derive surface albedo maps with unprecedentedaccuracy at medium spatial resolution over the globe. Sig-nificant efforts have been devoted over the last decade toachieve the challenging task of retrieving surface albedoquantities from these two instruments by optimally exploitingtheir specific multispectral and multiangular capabilities.Specifically, these high‐performance instruments havestimulated the development of dedicated algorithms to infersurface albedo quantities from sets of spectral and multiangleradiances measured at the top of the atmosphere. The suite ofalgorithms that have been adopted by the MODIS [e.g.,Schaaf et al., 2002] and MISR [e.g.,Martonchik et al., 1998]teams to estimate land surface albedos have, indeed, verylittle in common; this ultimately leads to the production oftwo distinct sets of independent products which can becompared and ultimately serve as benchmark for other, earlieror subsequent products.

[3] The main objective of this research is to document, inthe context of regional and large‐scale climate applications,the eventual differences and biases that may exist at conti-nental and global scales between the MODIS and MISRproducts over three angularly and spectrally integrated albedoquantities, namely, the broadband visible (0.3–0.7 mm), thenear infrared (0.7–3.0 mm) and the shortwave (0.3–3.0 mm)spectral domains. For the purpose of this comparison, allthree albedos are bihemispherical reflectances (BHRs) cal-culated under the assumption of an isotropic illumination ofthe surface, i.e., white sky albedo, as routinely generated bythe MODIS ground segment. The main motivation for se-lecting this particular product is twofold: (1) focusing onsurface albedo quantities that depend on intrinsic surfaceproperties only and (2) comparing products at the mostintegrated/averaged level in both the angular and spectraldomains. The current investigation aims at establishing andanalyzing statistical differences, if any, between the samephysical surface products derived independently from bothinstruments and thus complements previous results obtainedfrom a detailed comparison of the MISR versus MODISindividual spectral bands conducted at local scale over a fewselected sites [e.g., Lyapustin et al., 2007].[4] This paper focuses on the description of the necessary

methodological steps required to generate albedo quantitiesfrom MISR that are physically similar and thus comparablewith those available from MODIS. These steps include, inparticular, estimating the broadband white sky albedo valuesfrom the standard spectral products delivered by the MISRground segment. They must also address a number of issuesrelated to the differences in spatial and temporal sampling aswell as to the upstream data processing chains of both in-struments as described in section 2. The applicability of this

1Global Environment Monitoring Unit, Institute for Environment andSustainability, DG Joint Research Centre, European Commission, Ispra,Italy.

2On leave at Earth Observation Directorate, ESA‐ESRIN, Frascati,Italy.

3EUMETSAT, Darmstadt, Germany.4Department of Geography, University of Maryland, College Park,

Maryland, USA.

Copyright 2010 by the American Geophysical Union.0148‐0227/10/2009JD012665$09.00

JOURNAL OF GEOPHYSICAL RESEARCH, VOL. 115, D05101, doi:10.1029/2009JD012665, 2010

D05101 1 of 13

method and expected performance of the comparison ofthese two data sets are addressed in section 3. Conditionslimiting the impact of the difficulties in deriving similarphysical albedo quantities from both MODIS and MISR arealso illustrated and discussed.

2. Comparison Strategy

[5] The standard MODIS and MISR products both refer toratios between upward and downward radiant fluxes butthey differ in the selection of the atmospheric and Sun angleconditions adopted to generate the standard products. TheMODIS products provide surface albedo values that areidealized intrinsic surface properties: (1) a directional hemi-spherical reflectance (DHR) also called black sky albedo,which is associated with direct illumination only (no contri-bution from atmospheric scattering) and thus changes withthe Sun zenith angle, and (2) albedo values corresponding to aperfectly isotropic illumination (no dependency with respectto Sun zenith angle), i.e., a bihemispherical reflectance(BHRiso) also called white sky albedo. Our investigation usesthe broadband white sky albedo values available fromMODIS collection 5 products (MCD43B3) at 1 km (0.01°)spatial resolution for successive 16 day periods (https://lpdaac.usgs.gov/).[6] By contrast, the MISR standard albedo products in-

clude the directional hemispherical reflectance and the bi-hemispherical reflectance calculated under actual ambientatmospheric conditions and for the particular solar illumi-nation geometry at the time of the satellite measurement.The first task to achieve in the current MISR‐MODIS com-parison thus consists in deriving, from MISR, surface albedoquantities that are similar to MODIS, that is, (1) the estima-tion of narrowband white sky albedos from the surfacespectral characterization provided by MISR surface standardproducts which is presented in section 2.1 and (2) the spectralconversion of MISR narrowband to broadband white skyalbedo which is discussed in section 2.2. Collection 6 (gen-erated by version 17 of the standard processor) of the MISRlevel 2 Aerosol and Land products (MIL2ASLS) originallyavailable in a Space Oblique Mercator projection (http://eosweb.larc.nasa.gov/) were used to generate theMISRwhitesky albedos in the current investigation.

2.1. Estimating Narrowband White Sky AlbedoFrom MISR

[7] The MISR standard products include the surface bi-directional reflectance factors (BRFs) and the parametervalues of the Modified‐Rahman‐Pinty‐Verstraete (MRPV)empirical model [Engelsen et al., 1996; Martonchik et al.,1998] that optimally approximate these BRF fields in eachof the four MISR spectral bands. These model parametervalues can be used in forward mode in the MRPV model togenerate white sky spectral albedos that are comparable toMODIS. The MRPV model splits a BRF field into a scalaramplitude component, r0, and an angular function, ��sfc(itself implemented as the product of three functions) todescribe the anisotropy of the surface,

�sfc �0;�; �0; �c; b; kð Þ ¼ �0 ��sfc �0;�; �c; b; kð Þ: ð1Þ

[8] The parameter k controls the bowl or bell shape of theBRF fields [Pinty et al., 2002], the parameter b establishesthe degree of forward versus backward scattering and theparameter rc accounts for the hot spot effect, which may beespecially significant at or near the backscattering direction.Any direction W is characterized by a zenith angle whosecosine is equal to m and an azimuth angle noted � (thesubscript 0 identifies the particular direction of the Sun).[9] The contribution of the BRF angular field to the bihe-

mispherical reflectance, BHRiso, is estimated by integratingequation 1 over all upward directions under perfectly iso-tropic downward illumination conditions,

� �c; b; kð Þ ¼ 1

�2

Z 2�

0

Z 1

0

Z 2�

0

Z 1

0��sfc �

0; �

0; �

0 0; �

0 0; �c; b; k

� �

� �0d�

0d�

0�

0 0d�

0 0d�

0 0; ð2Þ

and the white sky spectral albedo is thus simply given by

BHRiso �0; �c; b; kð Þ ¼ �0 � �c; b; kð Þ: ð3Þ

[10] Note that, in practice, rc is set equal to r0 in theroutine MISR processing, which reduces the number of freeparameters to 3. Furthermore, equation (2) is calculatedusing an expansion as a cosine Fourier series with respect tothe relative azimuth angle (limited to the first two terms) assuggested by Pinty et al. [2005, equation (14)].[11] Note also that equations (2) and (3) are strictly valid

only if the BRF model parameters are not functions of theillumination angle, a condition that may not be verified insome circumstances [see Pinty et al., 2002].

2.2. Narrowband to Broadband Conversion of MISRWhite Sky Albedos

[12] The conversion from narrowband to broadband al-bedos is ensured by sets of conversion formulae that arespecific to each instrument. These conversion formulae takethe form of polynomials whose coefficients are optimized toaccount for wide varieties of simulated and/or measuredhigh spectral resolution reflectance spectra and for a diver-sity of atmospheric scenarios, paying particular attention tothe aerosol load and properties. The diversity of such con-version formulae available from the literature reflects thedifferences in the selection and representativeness of thetraining data sets as well as the radiation transfer modelsused to simulate the top of the atmosphere radiances. Forglobal applications, the derivation of these formulae is ratherchallenging since the training data sets must cover the fulldiversity of geophysical conditions that may be encounteredin actual measurements, but also be properly area weightedto limit the introduction of biases toward particular set ofconditions, e.g., bright soils versus dense vegetation. Also,under completely snow covered conditions, the MODISalbedo products are obtained using snow specific narrow-band to broadband conversion formulae [e.g., Stroeve et al.,2005].[13] In essence, broadband conversions can be success-

fully applied under conditions where the broadband regionis adequately sampled by the narrowband channels. This istypically the case for the visible (0.3–0.7 mm) domain, and/or when the spectral correlation between bands is significant

TABERNER ET AL.: MISR‐MODIS SURFACE ALBEDO COMPARISON D05101D05101

2 of 13

enough. In the former situation, the conversion plays therole of a weighted averaging function across the availablenarrowband channels while in the latter, the conversion isused as a spectral extrapolation scheme; as happens withMISR which has only one single narrowband in the near‐infrared region of the spectrum (0.85–0.87 mm) available toestimate the broadband near‐infrared (0.7–3.0 mm) albedovalues. This point is illustrated in Figure 1 which shows therelative spectral responses of MISR and MODIS land bandstogether with observed spectral variations of soils and typ-ical green vegetated surface reflectances (available from theASTER Spectral Library at http://speclib.jpl.nasa.gov/). Itappears indeed quite challenging to account for these sur-face spectral variations with a limited band set, as is the casewith MISR in the broadband near‐infrared region.[14] Given that this spectral conversion step may generate

significant (generally uncontrolled) uncertainties, we per-formed a number of tests on actual MODIS and MISRproducts to evaluate the conversion formulae applicable toMISR that ultimately provide the least biases between thebroadband products from both instruments. In addition, wehave considered a fourth broadband region (0.4–1.1 mm)which encompasses the four MISR spectral narrowbands inorder to limit the impact of the extrapolation in the far near‐infrared part of the solar spectrum.[15] Since the MODIS standard broadband albedos are

estimated using inherent albedo conversion formulae pub-lished by Liang et al. [1999] and Stroeve et al. [2005], thisspectral conversion concerns essentially the MISR products.Table 1 gives two published sets of spectral coefficients[Liang et al., 2002; Govaerts et al., 2006] entering first

degree polynomials that were found to provide limited al-bedo biases over large regions of the globe [Pinty et al.,2004] and over the four broadband regions (see section 3).Our confidence in these formulae is also enhanced by thefact that they yield comparable broadband albedo resultsalthough they were established using different training datasets and atmospheric radiation transfer models. The differ-ence in the coefficients including their signs optimized tothe same goal reflects the high degree of correlation thatexists in the spectral domain and the nonuniqueness of thesolutions to this spectral conversion problem.[16] According to Govaerts et al. [2006, Tables I and II],

the absolute uncertainties in surface albedo due to thespectral conversion of MISR and MODIS narrowbands tothe broadband 0.4–1.1 mm region are similar for both in-struments and close to 0.004. These uncertainty values in-crease to 0.0092 and 0.0060, for MISR and MODIS,respectively, when performing the conversion to the short-wave 0.3–3.0 mm broadband region. The latter absoluteuncertainty values should correspond to relative un-certainties on the order of 3 to 6% for the MODIS productsand 5 to 10% for the MISR derived products in most cases,i.e., assuming broadband shortwave albedo values between0.1 and 0.2. Any difference between the inherent andapparent albedo conversion formulae would, in all likelihood,be lost in the overall uncertainty of the spectral conversionprocedure. The same orders of magnitude are likely pertinentto the broadband near‐infrared albedo values as well. Thesestatistical uncertainties related to the estimation of broadbandproducts from measurements made in a limited number ofspectrally narrow bands are somewhat comparable to theknown cross calibration differences of about 3% (whichgenerally falls within the calibration uncertainties of bothinstruments) between the MISR bands and correspondingMODIS land bands [e.g., Bruegge et al., 2004; Thome et al.,2004; Xiong et al., 2005].

2.3. Sampling Issues

[17] Differences in the spatiotemporal sampling associatedwith the surface albedo products is another critical issue toconsider when performing comparison between these in-struments. The MODIS products used in this study aregenerated every 16 days (once per hexadecad) to ensure a

Figure 1. Relative spectral responses of MISR (dark grey)and MODIS (light grey) land bands together with observedspectral variations of soils (dashed lines) and typical greenvegetated surface reflectances (solid line). The spectral re-flectance data sets are available from the ASTER SpectralLibrary (http://speclib.jpl.nasa.gov/). The location and widthof the broadband visible (0.3–0.7 mm), near‐infrared (0.7–3.0 mm), and shortwave (0.3–3.0 mm) spectral domains areindicated above the plot.

Table 1. Values of the Coefficients Used to Convert the MISRNarrowband Surface Albedos Into Apparent Broadband Valuesa

Broadband Domain Blue Green Red Near Infrared Offset

0.3–0.7 mmb 0.3810 0.3340 0.2870 0.00 0.00.7–3.0 mmb −0.3870 −0.1960 0.5040 0.8300 0.0110.3–3.0 mmb 0.0 0.1260 0.3430 0.4150 0.00370.4–1.1 mmc 0.1485 0.2705 0.1788 0.3912 0.00.3–3.0 mmd 0.1713 −0.0217 0.3961 0.4028 0.00640.4–1.1 mmd 0.1852 0.2347 0.2074 0.3613 0.0013

aThe broadband value is the weighted sum of the spectral BHRs usingthose coefficients plus the offset. The coefficients adopted in this studyare those proposed by Liang et al. [2002].

bFrom Liang et al. [2002].cFrom the procedure adopted for estimating the coefficients applied to

generate the MODIS apparent surface albedo conversion algorithm[Liang et al., 1999].

dFrom Govaerts et al. [2006].

TABERNER ET AL.: MISR‐MODIS SURFACE ALBEDO COMPARISON D05101D05101

3 of 13

sufficient accumulation of surface BRF values to adequatelysample the upward radiance fields in the angular domain.This strategy takes advantage of the MODIS large swath andrelies on the variations in viewing conditions that occurduring this period for every ground location. The approachthus assumes that surface conditions remain mostly stablewhile, by contrast, and during the same 16 day accumulationperiods, the surface BRFs delivered by MISR (furtherexploited in terms of broadband white sky albedos) arebased on quasi‐instantaneous measurements. These BRFproducts, originally available in a Space Oblique Mercator

projection, have been first composited over 8 day and 16 dayperiods using a nearest neighbor technique and then re-mapped at 0.01° resolution on a sinusoidal grid identical tothe one selected for MODIS.[18] In all likelihood, the MISR products are more sen-

sitive to rapid changes in surface conditions that mayeventually happen during each of these 16 day periods. Thisissue becomes crucial when fresh snow falls or melts, orwhen other geophysical situations result in drastic andabrupt variations in surface reflectance. To avoid comparingproducts when such situations occur, only MODIS products

Figure 2. Maps of the (top) visible and (bottom) near‐infrared broadband surface albedo estimated overthe region (35°S–37°N; 18°W–52°E) for the first hexadecad (Julian day 1 to 16) of year 2005 over Africafrom (left) MISR and (right) MODIS. The various grey tones are related to the surface albedo values.Only MODIS high‐quality products are shown here.

TABERNER ET AL.: MISR‐MODIS SURFACE ALBEDO COMPARISON D05101D05101

4 of 13

with the highest quality assessment (QA) have been con-sidered in this study. MODIS high‐quality products corre-spond to conditions when the estimate of surface products isbased on the full retrieval algorithm, i.e., when a minimumof seven BRF observations are available to estimate thealbedo.[19] TheMODIS snow indicator associated with the 16 day

composite MODIS surface albedo product has been usedhere to identify snow events in both MODIS and MISR datasets since a snow mask or indicator is not routinely availablefrom MISR. It must be recalled that MISR surface albedovalues might still have been calculated under snowy con-ditions in the particular case where a spatially and tempo-rally limited snow event takes place just before one of the

MISR overpasses, but the rest of the 16 day accumulationperiod for MODIS is in majority snow free. In such a case,this ephemeral snow condition is discarded by the MODISalgorithm and the MODIS product at 500 m resolution willthen be flagged as snow free whereas the MISR product willcorrespond to snowy measurements. Furthermore when theMODIS 1 km resolution product is generated, snow free andsnow covered 500 m pixels are averaged together and whennone of the 500 m pixels are snow covered, the 1 km isflagged as snow free. Conditions where the quasi‐instanta-neous MISR albedo is affected by snow and the MODISalbedo has been derived from the predominantly snow freemeasurements yield outliers in the comparison, in the sameway as those that may be caused occasionally by upstream

Figure 3. Same as for Figure 2 but for hexadecad 13 (Julian day 209 to 224) of year 2005.

TABERNER ET AL.: MISR‐MODIS SURFACE ALBEDO COMPARISON D05101D05101

5 of 13

algorithms for both instruments, for example, inaccurateremoval of aerosol and/or cloud effects for instance.[20] A procedure was implemented to remove outliers

from the regional scale, e.g., over the entire African conti-nent, statistics. As will be seen in section 3, the effect of thisprocedure is largely cosmetic, although it does slightly re-duce the overall noise statistics and slightly improve thequality of the MISR‐MODIS derived relationships. In gen-eral, samples which fall outside a certain threshold of thebidimensional normal distribution, as characterized by thecovariance matrix, are removed. The process is reiterated,with the updated statistics, until the distribution has stabi-lized. For this particular study, a threshold which wouldexclude the outer 1% of the normal distribution was chosen.

3. Evaluation of the Method

3.1. Qualitative Assessment of the MISR DerivedWhite Sky Albedos

[21] Figure 2 exhibits some of the resulting products overa large region including the African continent (35°S–37°N;18°W–52°E). Figure 2 shows the white sky albedos derivedfrom MISR (Figure 2, left) and MODIS (Figure 2, right), forthe visible (Figure 2, top) and near‐infrared (Figure 2,bottom) broadband domains and for the first hexadecad(Julian day 1 to 16) of year 2005. Figure 3 shows the samefour plots for hexadecad 13 (Julian day 209 to 224) to illus-trate the seasonal cycle over that continent. The geographicalcoverage provided by both sensors is largely controlled bycloudiness, and to a certain extent by the limitations of

applicability of the aerosol retrieval algorithms. It can beseen that the overall grey‐scale coverage is broadly similarduring these particular 16 day periods of extreme latitudinaldisplacement of the Intertropical Convergence Zone (ITCZ).The white color indicates the unavailability of surface albedovalues.[22] Locally, differences in coverage result mainly from

the interaction between cloud cover and the samplingscheme of each instrument. During the reference hexadecad,the MODIS instrument may have multiple opportunities tosample any particular location on the ground. However, thesurface albedo algorithm requires multiple successful ob-servations during that period to retrieve a reliable product.By contrast, MISR can generate a surface characterizationon the basis of a single clear‐sky day during that hexadecad,but its limited swath width and revisiting frequency mayprevent the albedo product to be generated in some areas.This results in the striping effect apparent in the left‐handplots of Figures 2 and 3.[23] Figure 4 displays, in greater detail than the previous

two figures, the near‐infrared broadband surface albedoestimated over a region in northeastern Sudan (19°45′N–14°15′N; 31°30′E–35°45′E) by MISR (Figure 4, left) andMODIS (Figure 4, right). The geographical and morpho-logical features of this region are easily recognizable asstrong spatial contrasts. This example permits us to illustratethe overall quality of the custom developed remappingprocedure used to reproject the MISR white sky albedoproducts into the same map projection as MODIS. Theblockiness of the missing MISR surface albedo product,

Figure 4. (left) MISR and (right) MODIS maps of the near‐infrared broadband surface albedo estimatedfor the first hexadecad (Julian day 1 to 16) of year 2005 over a region in northeastern Sudan (19°45′N–14°15′N; 31°30′E–35°45′E). The various grey tones are related to the surface albedo values. The white blocksin the left plot correspond to missing surface albedo products. See text for details.

TABERNER ET AL.: MISR‐MODIS SURFACE ALBEDO COMPARISON D05101D05101

6 of 13

which results from the finite spatial resolution of the aerosolproduct (17.6 × 17.6 km), after remapping and compositingover the hexadecad is also evident in Figure 4 (left). TheMISR and MODIS products visually appear very similar,although the grey scales have been stretched to best show

spatial contrast. Section 3.2 further compares these productsfrom a quantitative point of view.

3.2. Comparative Statistics Between MISR andMODIS White Sky Albedos

[24] Figure 5 (top) shows density plots of the MODISversus MISR 0.4–1.1 mm broadband albedo values, esti-mated during the eighth hexadecad (Julian day 129 to 144)over the same continental window as Figures 2 and 3. Thesurface albedo values vary in a rather wide range, i.e., fromabout 0.05 to 0.5 over this region, which includes verybright deserts as well as dense vegetation canopies. Thoughthis graph appears muddled, the values of the statisticalindicators (reported at the top of the Figure 5 plots), i.e.,correlation coefficient, slope of the means, linear regressioncoefficients and primary eigenvectors reported in Figure 5(top) confirm an excellent agreement between the two datasets. The values of the statistical indicators are indeedgenerally within the range of uncertainties discussed insection 2. Figure 5 (bottom) illustrates the effect of theprocedure used to screen the outliers (constituting in thatcase less than 1.5% of the complete data set) that largelyreduces the noise of the MISR‐MODIS relationship withoutchanging significantly the values of the statistical indicators.As can be seen from Figure 6 the majority of these outliersare found at locations exhibiting strong geophysical transi-tions, e.g., along coast lines, and otherwise randomly dis-tributed over the entire region (most probably due toimproper or incomplete removal of cloud and aerosol con-tamination in at least one of the two data sets). About 4/5 ofthese outliers are associated with MISR values higher than

Figure 5. Density plots of the MODIS versus MISR 0.4–1.1 mm broadband albedo values estimated over the region(35°S–37°N; 18°W–52°E) for the eighth 16 day period ofyear 2005. Only MODIS high‐quality products are consid-ered here. (top) Statistics over the entire data sets and(bottom) those obtained after screening the outliers. Thesolid, dotted, and dashed lines show the fits obtained byestimating the slope of the means, the linear regression,and the primary eigenvectors, respectively.

Figure 6. Geographical distribution of the outliers screenedon bottom plot of Figure 5. The red (blue) color indicates thelocations where the MISR (MODIS) values are exceeding thecorresponding ones in the MODIS (MISR) data set. The greytones feature the grid points where both MODIS and MISRproducts are available during the eighth hexadecad period.

TABERNER ET AL.: MISR‐MODIS SURFACE ALBEDO COMPARISON D05101D05101

7 of 13

the corresponding MODIS values. The histogram of theabsolute MISR‐MODIS differences between the completeMODIS and MISR data set (Figure 7) over this region forthe eighth hexadecad indicates that the deviations are mostlyconfined within ±0.02 with a slight positive bias close to0.01. A separate inspection of the geographical distributionof this deviation reveals that MISR values tend to exceedthose from MODIS over bright desertic and sparsely vege-tated regions while the opposite situation is generally pre-dominant over darker nonvegetated surface conditions.[25] Similar statistics were produced for the other three

broadband albedo quantities in the shortwave (0.3–3.0 mm),visible (0.3–0.7 mm) and near‐infrared (0.7–3.0 mm) spectraldomains. Figure 8 shows density plots of the MODIS versusMISR broadband albedo values estimated during the eighthhexadecad over the shortwave (Figure 8, top), visible(Figure 8, bottom left) and near‐infrared (Figure 8, bottomright) spectral domains. The values of the statistical in-dicators for the 0.3–3.0 mm band are similar to those ob-tained previously over the 0.4–1.1 mm band. The threeindicators, however, show a slight (underestimation) over-estimation of MISR with respect to MODIS in the (visible)near‐infrared spectral domains. In all cases, however, thecorrelation coefficients are better than 0.98.[26] Figure 9 displays the temporal variation of the cor-

relation coefficients between the MODIS and MISR whitesky albedo data sets in these three broadband domains. Thisfigure indicates that the very consistent relationships dis-cussed previously based on products generated for theeighth 16 day period is extremely stable during the course ofthe year: the correlation coefficients lie in the [0.96–0.98]range. The annual means of the statistics shown on the right

hand side of Figure 9 confirm the robustness of the agree-ment between the two data sets at the continental scale.[27] The values of the primary eigenvectors retrieved for

every hexadecad during year 2005 are displayed in Figure 10.In the shortwave domain (Figure 10, top), the slopes betweenthe two distributions (MODIS over MISR) are slightly butconsistently less than unity, i.e., the annual mean is close to0.92, while the intercept remains positive and close to 0.01.This indicates that MISR‐derived white sky albedos arebiased positively on average over this particular region.Figure 10 middle (visible) and bottom (near‐infrared) plotsreveal, however, that the very good agreement noticed in theshortwave integrated domain, slightly deteriorates whensplitting the shortwave into the two visible and near‐infraredsubdomains. The annual mean of the slopes (intercepts) areclose to 0.96 (0.01) in the visible and 0.89 (0.03) in the near‐infrared domain. The root mean square values of the fits aresystematically larger in the near‐infrared than in the visibledomain. This is expected given the very limited and some-what biased sampling of the near‐infrared broadband domainby MISR (see section 2.2). One may also notice the occur-rence of a well‐defined seasonal variation in the visibledomain with a relative increase in the slope during theNorthern Hemisphere summer season.

3.3. Evaluation in the Case of Partial Snow Conditions

[28] The appearance and disappearance of snow in thewinter and spring seasons during the 16 day periods of theMODIS sequential accumulation periods reduce the spatialand temporal coverage of the midlatitude and high‐latituderegions when considering only the most reliable products(high‐quality products derived from the full retrieval algo-rithm). Figure 11 (top) provides an example of the geo-physical situations that prevailed over Northern Europeduring the fifth hexadecad (Julian day 65 to 81) of year2005. The retrievals in the northern part of this region arepredominantly associated with snow conditions (accordingto the MODIS snow flag), by contrast to the southern partwhere most of the albedo values correspond to snow‐freecases. As expected, the quasi‐instantaneous sampling ofMISR (Figure 11, top left) provides more opportunities toestimate surface albedo values as compared to MODIS(Figure 11, top right). This translates into a significantlymore complete spatial coverage over northern regions suchas Scandinavia than is feasible with a sequential accumu-lation approach, which is not appropriate when surface orcloud coverage conditions change drastically and frequently.[29] Figure 11 (bottom) shows density plots of the MODIS

versus MISR 0.3–0.7 mm broadband albedo values obtainedover the geographical region displayed in Figure 11 (top).The analysis of the complete data sets (Figure 11, bottomleft) reveals a number of outliers associated with rapidlychanging surface conditions discussed previously; this leadsto a number of situations (roughly 80%) where low MODISalbedo values (not contaminated by snow) are related torelatively high MISR albedos (suggesting the presence ofsnow at time of overpass).

4. Concluding Remarks

[30] Broadband white sky surface albedo products fromMODIS are compared against similar products generated

Figure 7. Histogram of the MISR‐MODIS absolute differ-ences estimated from the complete data sets over the fullgeographical window and for the eighth 16 day period ofyear 2005. The solid (dashed) vertical lines correspond tothe value of the mean (principal mode).

TABERNER ET AL.: MISR‐MODIS SURFACE ALBEDO COMPARISON D05101D05101

8 of 13

from MISR land surface BRF model parameters. The latterare used as input to estimate the integral form of the angularfunction of the MRPV BRF model, together with a set ofdedicated narrowband to broadband conversion formulae.Various issues related to different sampling strategies in thespace, time and angular domains between these instrumentsas well as to different approaches in estimating surface BRFvalues are addressed. Due to their significant impact onsurface albedo values, the occurrence of snow events de-serves a particular attention, especially under those condi-

tions where the snow fall/melting happens within thesequential accumulation period required by the MODIS al-bedo algorithm (16 day).[31] Overall, the results shown and discussed here are

very encouraging since rather good statistical agreementsbetween MISR and MODIS surface albedo products arefound, when analyzing large data sets derived for year 2005in four broadband spectral domains namely, 0.3–0.7 mm,0.4–1.1 mm, 0.7–3.0 mm and 0.3–3.0 mm. The relationshipsare, indeed, quasilinear with limited spread (low root mean

Figure 8. Density plots of the MODIS versus MISR broadband albedo values (obtained after screeningthe outliers) estimated over the region (35°S–37°N; 18°W–52°E) for the eighth 16 day period of year2005. The solid, dotted, and dashed lines show the fits obtained by estimating the slope of the means,the linear regression and, the primary eigenvectors, respectively, for the (top) 0.3–3.0, (bottom left)0.3–0.7, and (bottom right) 0.7–3.0 m broadband albedo values.

TABERNER ET AL.: MISR‐MODIS SURFACE ALBEDO COMPARISON D05101D05101

9 of 13

Figure 9. Temporal variation of the correlation coefficient in the broadband (top) shortwave (0.3–3.0 mm),(middle) visible (0.3–0.7 mm), and (bottom) near‐infrared (0.7–3.0 mm) spectral domains between MODISandMISRwhite sky albedo over the (35°S–37°N; 18°W–52°E) for year 2005. The right‐hand side values ineach plot correspond to the annual mean of the statistics, and the shaded areas delineate the standard devi-ation around the annual means.

TABERNER ET AL.: MISR‐MODIS SURFACE ALBEDO COMPARISON D05101D05101

10 of 13

Figure 10. Temporal variation of the primary eigenvectors in the broadband (top) shortwave (0.3–3.0 mm), (middle) visible (0.3–0.7 mm), and (bottom) near‐infrared (0.7–3.0 mm) spectral domains. Theinformation displayed in each plot is the same as indicated in Figure 9.

TABERNER ET AL.: MISR‐MODIS SURFACE ALBEDO COMPARISON D05101D05101

11 of 13

square errors) and always exhibit high correlation coeffi-cients. This is quite a remarkable result given the fact thatthe MODIS and MISR surface albedo data sets are gener-ated from suites of very different assumptions and algo-rithms, especially regarding the removal of atmosphericcontamination (cloud and aerosol effects) as well as thefurther exploitation of surface BRF values to generatebroadband white sky albedos.

[32] The good statistical consistency between the two datasets has also permitted us to expose some biases suggestingthat, for the studied regions, MISR shortwave albedos tend tobe generally slightly higher than the corresponding MODISproducts. These biases are more pronounced when consid-ering the visible and near‐infrared broadband domains andseem to follow seasonal variations.

Figure 11. (top) Maps of the visible broadband surface albedo for hexadecad 5 (Julian day 65 to 81) ofyear 2005 over Europe from (left) MISR and (right) MODIS. The various grey tones are related to thesurface albedo values. (bottom) Density plots of the MODIS versus MISR 0.3–0.7 mm broadband albedovalues for (left) statistics over the complete data set and (right) those obtained after screening the outliers.The solid, dotted, and dashed lines visualize the fits obtained by estimating the slope of the means, thelinear regression, and the primary eigenvectors, respectively.

TABERNER ET AL.: MISR‐MODIS SURFACE ALBEDO COMPARISON D05101D05101

12 of 13

[33] Acknowledgments. This research was performed at the GlobalEnvironment Monitoring unit of the Institute for Environment and Sustain-ability at the DG Joint Research Centre, an institution of the EuropeanCommission. The authors would like to thank Simon Pinnock (ESA) forhelpful discussions as well as the providers of the remote sensing data setsneeded to perform this research. The MISR products were obtained fromthe NASA Langley Research Center Atmospheric Sciences Data Center.The MODIS products were obtained from the National Snow and Ice DataCenter and the NASA Land Processes Distributed Active Archive Centers.

ReferencesBruegge, C. J., et al. (2004), Validating the MISR radiometric scale for theocean aerosol science communities, paper presented at InternationalWorkshop on Radiometric and Geometric Calibration, Int. Soc. forPhotogramm. and Remote Sens., Gulfport, Miss.

Engelsen, O., B. Pinty, M. M. Verstraete, and J. V. Martonchik (1996),Parametric bidirectional reflectance factor models: Evaluation, improve-ments and applications, Tech. Rep. EUR 16426 EN, Joint Res. Cent., Eur.Comm., Brussels.

Govaerts, Y., B. Pinty, M. Taberner, and A. Lattanzio (2006), Spectralconversion of surface albedo derived from Meteosat First Generationobservations, IEEE Geosci. Remote Sens. Lett., 3, 23–27, doi:10.1109/LGRS.2005.854202.

Liang, S., A. H. Strahler, and C. Walthall (1999), Retrieval of land surfacealbedo from satellite observations: A simulation study, J. Appl. Meteorol.,38, 712–725.

Liang, S., C. J. Shuey, A. L. Russ, H. Fang, M. Chen, C. Walthall,C. S. T. Daughtry, and R. J. Hunt (2002), Narrowband to broadband con-versions of land surface albedo: II. validation, Remote Sens. Environ., 84,25–41.

Lyapustin, A., Y. Wang, R. Kahn, J. Xiong, A. Ignatov, R. Wolfe, A. Wu,B. Holben, and C. Bruegge (2007), Analysis of MODIS‐MISR calibra-tion differences using surface albedo around AERONET sites and cloudreflectance, Remote Sens. Environ., 107, 12–21.

Martonchik, J. V., D. J. Diner, B. Pinty, M. M. Verstraete, R. B. Myneni,Y. Knyazikhin, and H. R. Gordon (1998), Determination of land and

ocean reflective, radiative, and biophysical properties using multiangleimaging, IEEE Trans. Geosci. Remote Sens., 36, 1266–1281.

Pinty, B., J.‐L. Widlowski, N. Gobron, M. M. Verstraete, and D. J. Diner(2002), Uniqueness of multi‐angular measurements: Part 1. A subpixelsurface heterogeneity indicator from MISR, IEEE Trans. Geosci. RemoteSens., 40, 1560–1573.

Pinty, B., et al. (2004), Intercomparison of surface albedo products fromvarious spaceborne sensors, paper presented at Workshop on Inter‐Comparison of Large Scale Optical and Infrared Sensors, ESA ESTEC,Noordwijk, Netherlands.

Pinty, B., A. Lattanzio, J. V. Martonchik, M. M. Verstraete, N. Gobron, M.Taberner, J.‐L. Widlowski, R. E. Dickinson, and Y. Goverts (2005),Coupling diffuse sky radiation and surface albedo, J. Atmos. Sci., 62(7), 2580–2591.

Schaaf, C. B., et al. (2002), First operational BRDF, albedo and nadirreflectance products from MODIS, Remote Sens. Environ., 83, 135–148.

Stroeve, J., J. E. Box, F. Gao, S. Liang, A. Nolin, and C. Schaaf (2005),Accuracy assessment of the MODIS 16‐day albedo product for snow:Comparisons with Greenland in situ measurements, Remote Sens. Envi-ron., 94, 46–60, doi:10.1016/j.rse.2004.09.001.

Thome, K., S. Biggar, and H. J. Choi (2004), Vicarious calibration ofTERRA ASTER, MISR and MODIS, Proc. SPIE Int. Soc. Opt. Eng.,5542, 290–299.

Xiong, X., N. Che, W. L. Barnes, and V. V. Salomonson (2005), Five‐years of Terra MODIS on‐orbit spectral characterization, Proc. SPIEInt. Soc. Opt. Eng., 5882, 58820R, doi:10.1117/12.614090.

N. Gobron, B. Pinty, M. Taberner, M. M. Verstraete, and J.‐L.Widlowski,Global Environment Monitoring Unit, Institute for Environment andSustainability, DG Joint Research Centre, European Commission, TP 440,via E. Fermi, I‐21020 Ispra (VA), Italy. (bernard.pinty@jrc.ec.europa.eu)Y. Govaerts, EUMETSAT, Am Kavalleriesand 31, D‐64295 Darmstadt,

Germany.S. Liang, Department of Geography, University of Maryland, 2181

LeFrak Hall, College Park, MD 20742, USA.

TABERNER ET AL.: MISR‐MODIS SURFACE ALBEDO COMPARISON D05101D05101

13 of 13