GEOPHYSICAL RESEARCH LETTERS, VOL. 29, NO. 9, 1353, 10.1029/2001GL014096, 2002

Relating MODIS-derived surface albedo to soils and rock types over Northern Africa and the Arabian peninsula Elena A. Tsvetsinskaya,1 C. B. Schaaf,1 F. Gao,1 A. H. Strahler,1 R. E. Dickinson,2 X. Zeng,3 and W. Lucht4 Received 14 September 2001; revised 30 November 2001; accepted 20 December 2001; published 15 May 2002.

[1] We use the MODerate resolution Imaging Spectroradiometer (MODIS) aboard the Terra spacecraft to derive surface albedo for the arid areas of Northern Africa and the Arabian peninsula. Albedo in seven MODIS spectral bands for land and three broad bands (for shortwave, near infrared, and visible portions of the spectrum) is produced. Surface albedo is derived from MODIS observations during a sixteen-day period and is analyzed at 1 km spatial resolution. MODIS data show considerable spatial variability of surface albedo in the study region that is related to soil and geological characteristics of the surface. For example, solar shortwave white-sky albedo varies by a factor of about 2.5 from the darkest volcanic terrains to the brightest sand sheets. Vegetation contribution to surface reflectance is essentially negligible since we only considered pixels with under ten percent fractional canopy cover. Few, if any, coupled landatmosphere global or regional models capture this observed spatial variability in surface reflectance or albedo. Here we suggest a scheme that relates soil groups (based on the United Nations Food and Agriculture Organization, FAO, soil classification) and rock types (based on the United States Geological Survey, USGS, geological maps) to MODIS derived surface albedo statistics. This approach is a first step towards the incorporation of the observed spatial variability in surface reflective properties into climate models. INDEX TERMS: 3322 Meteorology and Atmospheric Dynamics: Land/ atmosphere interactions; 3337 Meteorology and Atmospheric Dynamics: Numerical modeling and data assimilation; 3359 Meteorology and Atmospheric Dynamics: Radiative processes; 3360 Meteorology and Atmospheric Dynamics: Remote sensing

1. Introduction [2] Surface albedo represents the fraction of incoming solar radiation reflected back to the atmosphere and space. As such, it is a basic geophysical property of land surfaces and is needed for any study involving the linking of terrestrial surfaces to the absorption of solar energy. In particular, surface albedo is a required component for models of global and regional climate [e.g., Dickinson, 1992; Knorr et al., 2001], or for numerical weather prediction [e.g. Viterbo and Betts, 1999]. By determining in part the absorption of solar energy, it helps determine the sensible, latent, and soil heat fluxes, and consequently the thermal and moisture stratification of the atmospheric boundary layer. 1 Department of Geography and Center for Remote Sensing, Boston University, Boston, Massachusetts, USA. 2 Georgia Institute of Technology, Atlanta, Georgia, USA. 3 Institute of Atmospheric Physics, The University of Arizona, Tucson, Arizona, USA. 4 Potsdam Institute for Climate Impact Research, Potsdam, Germany.

Copyright 2002 by the American Geophysical Union. 0094-8276/02/2001GL014096$05.00

[3] We focus our study on the largest desert belt in the world that encompasses Northern Africa and the Arabian peninsula. Much of the region is as reflective as almost any surface on earth (with the exception of fresh snow and ice). Such highly reflective surfaces coupled to the overlying atmosphere can be major energy sinks relative to more highly absorbing neighboring regions. Such gradients in forcing lead to regional circulations that can suppress precipitation and can be strongly dependent on the degree of surface darkening by overlying vegetation [Charney, 1975; Claussen and Gayler, 1997]. [4] Climate models commonly specify separate albedos for vegetation and soils and determine total surface albedo as a mixture of the two. Many such models have based their soil albedos on Wilson and Henderson-Sellers [1985] who used an earlier version of FAO [1994] to classify world soils at 1 degree resolution as light, medium, and dark. Global climate models [e.g., Bonan, 1996] have typically averaged these over the larger model grid-squares and stretched them to a scale of 1 (light) to 8 (dark). High albedo values typical for the study region have prompted the creation of an additional land/soil class of bright or hot desert in such classification schemes, different from the other desert regions of the world [Bonan, 1996; Henderson-Sellers, 1990]. Obviously, with increasingly sparse vegetation, the soil albedo becomes increasingly dominant in determining the total surface albedo. Hence, determining arid region ‘‘end-members’’ is especially important. Not only mean values but spatial variability in albedo is important for the climate of Northern Africa [Lofgren, 1995]. Although lower resolution global climate models typically have meshes several hundred km on a side, the current widespread application of regional climate models, and development of higher resolution versions of the global climate models either for land or for the atmosphere as well require data sets at 50 km or finer resolution. [5] Considerable spatial variability in surface albedo over Northern Africa and the Arabian peninsula has been observed by the MODerate resolution Imaging Spectroradiometer (MODIS), as well as by other space borne spectroradiometers [Strugnell et al., 2001; Pinty et al., 2000]. Recently attempts have been made to introduce this spatial variability to land surface models [Wei et al., 2001; Knorr et al., 2001]. In this paper, we relate the MODIS observed spatial variability in surface albedo to specific soil and geological features in the study region.

2. Materials and Methods [6] The MODIS albedo algorithm [Lucht et al., 2000; Schaaf et al., 2002] couples multidate, cloud-free, atmospherically-corrected surface reflectance with a semiempirical, kernel-driven Bidirectional Reflectance Distribution Function (BRDF) model to determine the anisotropy of global land surfaces at 1 km spatial resolution every 16 days. The resulting BRDFs are used to produce directional hemispherical albedo (hereafter referred to as black-sky albedo) and bihemispherical albedo (hereafter referred to as whitesky albedo) for seven MODIS spectral bands for land and three broadbands (visible, near infrared, and total shortwave). Note that black-sky albedo represents the direct beam contribution while white-sky albedo represents the entire diffuse portion and these two

67 - 1

67 - 2 TSVETSINSKAYA ET AL.: RELATING MODIS-DERIVED SURFACE ALBEDO TO SOILS AND ROCK TYPES

Figure 1. White-sky surface albedo at 1 km spatial resolution. The image is a true-color RGB composite of MODIS band 1 (red, with a linear stretch applied from 0 to 0.5), band 4 (green, with a linear stretch applied from 0 to 0.4), and band 3 (blue, with a linear stretch applied from 0 to 0.4). Lambert Azimuthal Equal Area projection. extremes can be combined as a function of the diffuse skylight for a representation of an actual albedo such as measured by field instruments [Lucht et al., 2000]. The MODIS Albedo/BRDF product has been operational since April 2000, and has been released to the public and assigned a provisional status since October 2000. Prelaunch error statistics were estimated to be under 10% [Lucht, 1998] and initial postlaunch field evaluations under clear sky and stable surface conditions support these estimates [Liang et al., 2002]. The results presented here focus on the first reprocessing of a 16-day period starting on October 31, 2000 (Day of Year 305). This period falls into the most desert dust-free window of the year in Northern Africa and the Arabian peninsula, which is from November through January [Herman et al., 1997]. The other 16-day periods examined were those of November 16, December 2, and December 18, 2000. [7] Soil information for the study region comes from the United Nations Food and Agriculture Organization, FAO, digitized soil map of the world [FAO, 1994], which is a global soils data set compiled at the scale of 1:5,000,000. We used soil groups rather than individual soil types in the final analysis to allow for a more useful generalization and identification of the clearest trends in the data. Wilson and Henderson-Sellers [1985] used an earlier version of the same soils data set to derive global maps of soil color and texture that are currently used in many land surface models [Bonan, 1996; Henderson-Sellers, 1990]. To complement the soils information, we used an auxiliary data set of the United States Geological Survey, USGS, digitized geological maps of Africa and the Arabian peninsula [USGS, 1997, 1998], which were originally compiled at the scales of 1:12,500,000 and 1:4,500,000, respectively. Geological units of the maps are based on the age of rock and in some cases the type of rock or surface sediment (e.g., volcanics, intrusives, eolian, fluvial). [8] The three data sets (MODIS albedo, FAO soils, and USGS geological maps) were all reprojected to Lambert Azimuthal Equal Area projection, with the center of projection at 20N 20E, and regridded if necessary to the spatial resolution of 1 km. A mask was imposed, based on Friedl et al. [2002], such that only pixels classified as barren (i.e., with under 10 percent fractional canopy cover) were used for the analysis. This barren mask is consistent with Zeng et al. [2000].

3. Results and Discussion [9] Spectral white-sky surface albedo for the study region is shown in Figure 1. The image is an RGB (Red-Green-Blue) truecolor composite of MODIS red (band 1, 620-670 nm), green (band 4, 545 – 565 nm), and blue (band 3, 459 – 479 nm) bands, with

linear stretches applied to each band. Lighter color (yellow) corresponds to the highest albedo values, i.e. the most reflective surfaces. Albedo is shown for the entire North Africa and the Arabian peninsula, prior to application of the barren land cover mask. Pixels shown in white in Figure 1 indicate areas of no or poor quality MODIS data and were excluded from the analysis. Note that black-sky albedo at local solar noon (not shown here) is essentially identical to the white-sky albedo (Figure 1) in terms of spatial pattern. Black-sky albedo is generally somewhat lower than white-sky albedo, but the difference between the two is uniformly well under 5%. Furthermore note that although not shown here, when both white-sky and black-sky albedo were aggregated to the coarser spatial resolutions typical for regional and global climate models (i.e., 50 km and 250 km, respectively), there was still substantial spatial variability in albedo over the study region. For example, white-sky albedo values in the total shortwave broadband at 250 km resolution ranged from 0.184 to 0.477. [10] Figure 2, in contrast to Figure 1, shows only the pixels classified as barren (i.e., with fractional canopy cover under 10 percent). In order to minimize the vegetation signal, only those pixels were used in our analysis. Considerable spatial variability of albedo in the study region can be clearly seen (Figure 1) and is associated with geological and soil variability (Figure 2). The most reflective areas in Figure 1 generally coincide with areas of sand dune fields and sand sheets (see Figure 3a for albedo histogram and Table 1 for albedo statistics). The next most reflective soil types are regosols and arenosols, both soils with no diagnostic pedogenic horizons (or none other than an ochric A horizon), and salinized soils (salt and solonetz). After examining histograms of salt and solonetz, which were essentially identical, we combined them into a single category (Figure 3b, Table 1). Yermosols, the most abundant soil group in the study region, are close to salt and solonetz in mean albedo, but have a somewhat wider spread around the mean (not shown here). This wider spread and greater standard deviation (Table 1) are related to the various degrees of soil salinization throughout the study region and the particular combinations of the stony/sandy/clayey fractions in any given area. Desert pavement of varying thickness typically forms on the surface of yermosols. [11] Several soil groups have somewhat lower albedo, and a generally higher coefficient of variation, compared to the aforementioned soil groups. Among them, lithosols (Figure 3c) and rock both have clearly bimodal albedo histograms. After overlaying albedo and soils data, for lithosols and rock, with the USGS geological maps [USGS, 1997, 1998], we could clearly see that the lower mode in both cases corresponds to quaternary and older volcanics and intrusives and the higher mode corresponds to other types of exposed bedrock. We split these modes into

TSVETSINSKAYA ET AL.: RELATING MODIS-DERIVED SURFACE ALBEDO TO SOILS AND ROCK TYPES

67 - 3

Figure 2. FAO digitized soils map [FAO, 1994], resampled on a 1 km grid. Lambert Azimuthal Equal Area projection. Pixels shown in white are either non-barren (i.e., fractional canopy cover from Friedl et al. [2002] is over 10%) or soil groups of under 2,500 pixels (i.e., gleysols, phaeozems, and kastanozems). separate categories in Table 1. At the same time, we combined lithosols and rock, since both soil groups identify areas of exposed bedrock (with no overlaying material in case of rock and minimal material, under 10 cm, in case of lithosols), and the means of the two modes were essentially identical between the two groups. Xerosols, fluvisols (soils associated with recent alluvial deposits), and solonchaks (highly saline soils) have the red band albedo between .306 and .327 (with higher standard deviation, between .070 and .083). Their histograms suggest one clear mode with a considerable spread around it, which could either be interpreted as a very weakly pronounced second mode or a consequence of artifacts of misregistration of the maps (i.e., MODIS albedo, FAO soils, and USGS geological maps), mislabelling of the barren pixels, errors in the soils and geological databases, and natural inherent heterogeneity associated with spatial differences in soil water content for the particular period of analysis. Such issues are common for all surfaces but may be more pronounced in these soils due to their smaller sample size. [12] The darkest areas in Figure 1 correspond to vertisols, cambisols, rendzinas, and luvisols. They are mostly located in and around the Atlas mountains, in the northwestern portion of the study region. Our analysis is limited to supporting modeling studies on spatial scales of 50 km or greater so soil groups under 2,500 pixels are not considered. The only region affected by this exclusion was the high altitude deserts of the Atlas mountains, where soils tend to have smaller sample size and are generally characterized with lower albedo values. [13] The analysis presented in Table 1 is limited to November 2000. As mentioned earlier, we did examine other temporal slices in the relatively desert dust-free period from November 2000 through January 2001 [Herman et al., 1997] and found that albedo

is very stable over the entire period. MODIS-derived white-sky surface albedo presented here is somewhat lower (although the difference is generally small) compared to the recent Meteosatderived broadband surface albedo of Knorr et al. [2001], who give values above 0.35 over most of the Saharan Africa, with one area above 0.500. Our values are generally somewhat higher compared to the recent AVHRR-derived broadband albedo of Strugnell et al. [2001] and Wei et al. [2001], who suggest albedo under 0.4 over the Saharan region. While these differences are probably due to the specifics of the narrow spectral band to broadband conversions employed by different authors, lack of atmospheric correction and poor georeferencing characteristics of the AVHRR sensor, choice of temporal windows selected for the analysis, and the particulars of spatial aggregation of the data, the rigor of the MODIS geolocation and calibration, and its 1 km spatial resolution lend credibility to our results. [14] Many applications of measured albedos require their linkage to underlying surface properties. Climate models, for example examine consequences of changing vegetation covers and possibly on longer time scales, changing soil surfaces. This paper demonstrates, at least for Northern Africa and the Arabian peninsula, that the MODIS data for arid surfaces can be efficiently summarized in terms of readily available global distributions of soil groups and geological land forms. Although this paper has largely shown this representation for spectral and broadband total white-sky (i.e diffuse) albedo, the conclusions hold for the only slightly different black-sky values (i.e. direct beam; exact values are available from the authors). This representation spans a wide range of surface albedos from the extremely bright sand dune surfaces with albedo of approximately 0.5 to the darker mode of lithosol and rock surfaces with albedo of about 0.2.

Figure 3. Histograms of white-sky albedo for MODIS band 1 (red, 620-670 nm), for selected FAO soil groups: (a) dunes/shifting sands, (b) salt and solonetz, (c) lithosols.

67 - 4 TSVETSINSKAYA ET AL.: RELATING MODIS-DERIVED SURFACE ALBEDO TO SOILS AND ROCK TYPES Table 1. Mean and standard deviation (in parentheses) for MODIS white-sky albedo, by FAO soil group. Albedo Category

FAO Soil Group (Pixel Number)

620 – 670 nm

300 – 700 nm

700 – 5,000 nm

300 – 5,000 nm

Very High

Dunes / Shifting Sands (1,748,498) Regosols (901,044) Arenosols (426,355) Salt (30,832) and Solonetz (1,476) Yermosols (5,258,148) Lithosols and Rock: Othera (2,568,322b) Xerosols (301,269) Fluvisols (157,104) Solonchaks (168,391) Vertisols (18,505) Cambisols (15,368) Rendzinas (4,637) Luvisols (9,327) Lithosols and Rock: Volcanics and Intrusives (178,347b)

0.459 (0.053)

0.295 (0.044)

0.575 (0.053)

0.426 (0.043)

0.409 (0.078) 0.374 (0.073) 0.371 (0.057)

0.272 (0.056) 0.255 (0.052) 0.250 (0.044)

0.518 (0.092) 0.479 (0.083) 0.459 (0.063)

0.384 (0.068) 0.357 (0.064) 0.346 (0.048)

0.367 (0.073)

0.247 (0.048)

0.470 (0.089)

0.348 (0.065)

High

Moderate

Low

a

0.320 0.327 0.321 0.306 0.231 0.226 0.211 0.198

(0.096) (0.070) (0.083) (0.080) (0.043) (0.077) (0.079) (0.044)

0.187 (0.061)

a

0.217 0.213 0.216 0.225 0.163 0.159 0.152 0.133

(0.058) (0.041) (0.050) (0.057) (0.028) (0.051) (0.060) (0.029)

0.137 (0.039)

a

(0.118) (0.088) (0.100) (0.097) (0.057) (0.094) (0.097) (0.056)

0.302a (0.086) 0.312 (0.064) 0.306 (0.074) 0.284 (0.072) 0.220 (0.040) 0.227 (0.071) 0.204 (0.072) 0.196 (0.041)

0.223 (0.077)

0.174 (0.056)

0.406 0.423 0.412 0.370 0.293 0.314 0.272 0.266

Number of pixels in each group is shown in parentheses following the group name. MODIS band 1 (red, 620 – 670 nm), the visible broadband (300 – 700 nm), the near infrared broadband (700 – 5,000 nm), and the total shortwave broadband (300 – 5,000 nm) are shown. Note that only pixels classified by Friedl et al. [2002] as barren (i.e., under 10% fractional canopy cover) were considered. Note also that the black-sky albedo at local solar noon is generally similar to the white-sky albedo presented here (the mean of the black-sky albedo tends to be lower than the white-sky albedo values presented here, but the difference between the means is always well within 5%). a In the USGS [1997, 1998] geological maps, Precambian rocks were undivided and consisted of sedimentary, metamorphic, and igneous types, all in one category. Hence, Precambrian volcanics and intrusives (i.e., mostly the areas in the rift zone along the Red Sea), which have low albedo (similar to that of other volcanics and intrusives) were included in the ‘‘Other’’ category, substantially decreasing the mean albedo for the category and raising the standard deviation. b Note that we combined Lithosols (soil group of 2,464,161 pixels) with Rock (soil group 284,617 pixels). We then split the combined groups into two modes: Volcanics and Intrusives (consisting of 178,347 pixels) and Other (consisting of 2,568,322 pixels).

[15] Acknowledgments. This work was supported by grants from NASA’s EOS NRA-99-OES-04 and MODIS projects. We especially thank the anonymous reviewer for constructive and helpful comments. We thank Alessandro Baccini and Mutlu Ozdogan of Boston University for their great help with image registration.

References Bonan, G. B., A Land Surface Model (LSM Version 1.0) for Ecological, Hydrological, and Atmospheric Studies: Technical Description and User’s Guide, NCAR Tech. Note, 417, 150, 1996. Charney, J. G., Dynamics of Deserts and Drought in the Sahel, Quart. J. Roy. Meteor. Soc., 101, 193 – 202, 1975. Claussen, M., and V. Gayler, The greening of the Sahara during the midHolocene: Results of an interactive atmosphere-biome model, Global Ecol. Biogeogr. Lett., 6, 369 – 377, 1997. Dickinson, R. E., Land Surface, In: Climate System Modeling, edited by K. E. Trenberth, Cambridge University Press, 149 – 171, 1992. FAO, Digitized Soil Map of the World (version 2.0) derived from the FAO/ UNESCO Soil Map of the World, Food and Agriculture Organization, Rome, Italy, 1994. Friedl, M. A., D. K. McIver, J. C. F. Hodges, X. Zhang, D. Muchoney, A. H. Strahler, C. E. Woodcock, S. Gopal, A. Schnieder, A. Cooper, A. Baccini, F. Gao, and C. Schaaf, Global Land Cover from MODIS: Algorithms and Early Results, Remote Sens. Environ., in press 2002. Henderson-Sellers, A., Predicting Generalized Ecosystem Groups with the NCAR CCM: First Steps Towards an Interactive Biosphere, J. Climate, 3, 917 – 940, 1990. Herman, J. R., P. K. Bhartia, O. Torres, C. Hsu, C. Seftor, and E. Celarier, Global Distribution of UV-Absorbing Aerosols from Nimbus 7/TOMS Data, J. Geophys. Res., 102(D14), 16,911 – 16,922, 1997. Knorr, W., K.-G. Schnitzler, and Y. Govaerts, The role of bright desert regions in shaping North African climate, Geophysical Research Letters, 28(18), 3489 – 3492, 2001. Liang, S., H. Fang, M. Chen, C. Shuey, C. Walthall, C. Daughtry, J. Morisette, C. Schaaf, and A. Strahler, Validating MODIS land surface reflectance and albedo products: Methods and preliminary results, Remote Sens. Environ., in press 2002. Lofgren, B. M., Surface Albedo Climate Feedback Simulated Using TwoWay Coupling, J. Climate, 8(10), 2543 – 2562, 1995.

Lucht, W., Expected retrieval accuracies of bidirectional reflectance and albedo from EOS-MODIS and MISR angular sampling, J. Geophys. Res., 103(D8), 8763 – 8778, 1998. Lucht, W., 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., 38(2), 977 – 998, 2000. Pinty, B., F. Roveda, M. M. Verstraete, N. Gobron, Y. Govaerts, J. V. Martonchik, D. J. Diner, and R. A. Kahn, Surface albedo retrieval from Meteosat-2. Applications, J. Geophys. Res., 105, 18,099 – 18,112, 2000. Schaaf, C. B., F. Gao, A. H. Strahler, and co-authors, First Operational BRDF, Albedo, and Nadir Reflectance Products from MODIS, Remote Sens. Environ., in press 2002. Strugnell, N. C., W. Lucht, and C. Schaaf, A global albedo data set derived from AVHRR data for use in climate simulations, Geophysical Research Letters, 28(1), 191 – 194, 2001. USGS, Maps Showing Geology, Oil and Gas Fields and Geological Provinces of Africa, USGS Open File Report 97-470A (CD-ROM ), 1997. USGS, Maps Showing Geology, Oil and Gas Fields, and Geologic Provinces of the Arabian Peninsula, USGS Open File Report 97-470B (CDROM ), 1998. Viterbo, P., and A. K. Betts, Impact on ECMWF forecasts of changes to the albedo of the boreal forests in the presence of snow, J. Geophys. Res., 104(D22), 27,803 – 27,810, 1999. Wei, X., A. N. Hahmann, R. E. Dickinson, Z.-L. Yang, X. Zeng, J. Schaudt, C. B. Schaaf, and N. Strugnell, Comparison of albedos computed by land surface models and their evaluation against remotely sensed data, J. Geophys. Res., 106(D18), 20,687 – 20,702, 2001. Wilson, M. F., and A. Henderson-Sellers, A Global Archive of Land Cover and Soils Data for Use in General Circulation Climate Models, J. Climatology, 5, 119 – 143, 1985. Zeng, X., R. E. Dickinson, A. Walker, M. Shaikh, R. S. DeFries, and J. Qi, Derivation and Evaluation of Global 1-km Fractional Vegetation Cover Data for Land Modeling, J. Appl. Meteor., 39, 826 – 839, 2000.

Elena A. Tsvetsinskaya, Department of Geography and Center for Remote Sensing, Boston University, 675 Commonwealth Avenue, Boston, MA, USA. ([email protected])