• Alton, P., North P. , Kaduk J. , and Los S. , 2005: Radiative transfer modeling of direct and diffuse sunlight in a Siberian pine forest. J. Geophys. Res., 110 , D23209. doi:10.1029/2005JD006060.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Asner, G., 1998: Biophysical and biochemical sources of variability in canopy reflectance. Remote Sens. Environ., 64 , 234253.

  • Barnsley, M., Hobson P. , Hyman A. , Lucht W. , Muller J. , and Strahler A. , 2000: Characterizing the spatial variability of broadband albedo in a semidesert environment for MODIS validation. Remote Sens. Environ., 74 , 5868.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Betts, A., and Ball J. , 1997: Albedo over the boreal forest. J. Geophys. Res., 102 , 2890128909.

  • Culf, A., Fisch G. , and Hodnett M. , 1995: The albedo of Amazonian forest and ranch land. J. Climate, 8 , 15441554.

  • Dunderdale, M., Muller J-P. , and Cox P. , 1999: Sensitivity of the Hadley Centre Climate Model to different Earth observation and cartographically derived land surface data-sets. Proc. Int. Conf. and Workshops on Ocean Color, Land Surfaces, Radiation and Clouds, Aerosols (ALPS ’99): The Contribution of POLDER and New Generation Spaceborne Sensors to Global Change Studies, Vol. 1–6; Meribel, France, Centre National d’Etudes Spatiales, 1–4.

    • Search Google Scholar
    • Export Citation
  • Gao, F., Schaaf C. , Strahler A. , Roesch A. , Lucht W. , and Dickinson R. , 2005: MODIS bidirectional reflectance distribution function and albedo Climate Modeling Grid products and the variability of albedo for major global vegetation types. J. Geophys. Res., 110 , D01104. doi:10.1029/2004JD005190.

    • Search Google Scholar
    • Export Citation
  • Grant, I., Prata A. , and Cechet R. , 2000: The impact of the diurnal variation of albedo on the remote sensing of the daily mean albedo of grassland. J. Appl. Meteor., 39 , 231244.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Huete, A., Didan K. , Miura T. , Rodriguez E. , Gao X. , and Ferreira L. , 2002: Overview of the radiometric and biophysical performance of the MODIS vegetation indices. Remote Sens. Environ., 83 , 195213.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Idso, S., Jackson R. , Reginato R. , Kimball B. , and Nakayama F. , 1975: The dependence of bare soil albedo on soil water content. J. Appl. Meteor., 14 , 109113.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Jin, Y., Schaaf C. , Woodcock C. , Gao F. , Li X. , Strahler A. , Lucht W. , and Liang S. , 2003: Consistency of MODIS surface bidirectional reflectance distribution function and albedo retrievals: 2. Validation. J. Geophys. Res., 108 , 4159. doi:10.1029/2002JD002804.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Johns, T., and Coauthors, 2006: The new Hadley Centre Climate Model (HadGEM1): Evaluation of coupled simulations. J. Climate, 19 , 13271353.

  • Lawrence, P., and Chase T. , 2007: Representing a new MODIS consistent land surface in the Community Land Model (CLM 3.0). J. Geophys. Res., 112 , G01023. doi:10.1029/2006JG000168.

    • Search Google Scholar
    • Export Citation
  • Lewis, P., and Barnsley M. , 1994: Influence of the sky radiance distribution on various formulations of the earth surface albedo. Proc. Sixth ISPRS Int. Symp. on Physical Measurements of Signatures in Remote Sensing, Val d’Isère, France, Centre National d’Etudes Spatiales, 707–715.

    • Search Google Scholar
    • Export Citation
  • Liang, X-Z., and Coauthors, 2005: Development of land surface albedo parameterization based on Moderate Resolution Imaging Spectroradiometer (MODIS) data. J. Geophys. Res., 110 , D11107. doi:10.1029/2004JD005579.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Lofgren, B., 1995: Surface albedo–climate feedback simulated using two-way coupling. J. Climate, 8 , 25432562.

  • Moody, E., King M. , Platnick S. , Schaaf C. , and Gao F. , 2005: Spatially complete global spectral surface albedos: Value-added datasets derived from Terra MODIS land products. IEEE Trans. Geosci. Remote Sens., 43 , 144158.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Myneni, R., and Williams D. , 1994: On the relationship between FAPAR and NDVI. Remote Sens. Environ., 49 , 200211.

  • North, P. R. J., 2002: Estimation of fAPAR, LAI, and vegetation fractional cover from ATSR-2 imagery. Remote Sens. Environ., 80 , 114121.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Pinker, R., and Laszlo I. , 1992: Modeling surface solar irradiance for satellite applications on a global scale. J. Appl. Meteor., 31 , 194211.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Pinty, B., and Coauthors, 2007: Retrieving surface parameters for climate models from Moderate Resolution Imaging Spectroradiometer (MODIS)-Multiangle Imaging Spectroradiometer (MISR) albedo products. J. Geophys. Res., 112 , D10116. doi:10.1029/2006JD008105.

    • Search Google Scholar
    • Export Citation
  • Press, W., Flannery B. , Teukolsky S. , and Vetterling W. , 1992: Numerical Recipes in FORTRAN 77: The Art of Scientific Computing. 2nd ed. FORTRAN Numerical Recipes Series, Vol. 1, Cambridge University Press, 992 pp.

    • Search Google Scholar
    • Export Citation
  • Roesch, A., Schaaf C. , and Gao F. , 2004: Use of Moderate-Resolution Imaging Spectroradiometer bidirectional reflectance distribution function products to enhance simulated surface albedos. J. Geophys. Res., 109 , D12105. doi:10.1029/2004JD004552.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Ross, J., 1981: The Radiation Regime and Architecture of Plant Stands. Tasks for Vegetarian Science Series, Vol. 3, Kluwer Academic, 391 pp.

    • Search Google Scholar
    • Export Citation
  • Salomon, J., Schaaf C. , Strahler A. , Gao F. , and Jin Y. , 2006: Validation of the MODIS bidirectional reflectance distribution function and albedo retrievals using combined observations from the Aqua and Terra platforms. IEEE Trans. Geosci. Remote Sens., 44 , 15551565.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Schaaf, C., and Coauthors, 2002: First operational BRDF, albedo nadir reflectance products from MODIS. Remote Sens. Environ., 83 , 135148.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Sellers, P., 1985: Canopy reflectance, photosynthesis and transpiration. Int. J. Remote Sens., 6 , 13351372.

  • Sud, Y., and Fennessy M. , 1982: A study of the influence of surface albedo on July circulation in semi-arid regions using the GLAS GCM. Int. J. Climatol., 2 , 105125.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Susaki, J., Yasuoka Y. , Kajiwara K. , Honda Y. , and Hara K. , 2007: Validation of MODIS albedo products of paddy fields in Japan. IEEE Trans. Geosci. Remote Sens., 45 , 206217.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Tsvetsinskaya, E., Schaaf C. , Gao F. , Strahler A. , Dickinson R. , Zeng X. , and Lucht W. , 2002: Relating MODIS-derived surface albedo to soils and rock types over Northern Africa and the Arabian peninsula. Geophys. Res. Lett., 29 , 1353. doi:10.1029/2001GL014096.

    • Search Google Scholar
    • Export Citation
  • Wang, Z., Zeng X. , Barlage M. , Dickinson R. , Gao F. , and Schaaf C. , 2004: Using MODIS BRDF and albedo data to evaluate global model land surface albedo. J. Hydrometeor., 5 , 314.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Wilson, M., and Henderson-Sellers A. , 1985: A global archive of land cover and soils data for use in general circulation climate models. Int. J. Climatol., 5 , 119143.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Zeng, X., Dickinson R. , Walker A. , Shaikh M. , DeFries R. , and Qi J. , 2000: Derivation and evaluation of global 1-km fractional vegetation cover data for land modeling. J. Appl. Meteor., 39 , 826839.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • View in gallery

    Relationship between MODIS broadband albedos and NDVI for a location in the Sahel. NDVI is computed from MODIS white sky albedos in narrowband channels 1 and 2 [Eq. (2)]: (a) visible, (b) near-infrared, and (c) shortwave.

  • View in gallery

    Frequency distributions of NDVI over “pure” IGBP cells corresponding to >90% of a single IGBP class for 14 different IGBP classes. NDVI is computed from MODIS white sky albedos in narrowband channels 1 and 2 [Eq. (2)] for the entire period from 2002 to 2006. The data are binned into NDVI intervals of 0.01.

  • View in gallery

    Global total shortwave white sky soil background (a) albedos and (b) associated error map. Uncertainty corresponds to the 95% confidence interval derived from mean values over bare soil and derived from the linear regression model of NDVI against surface albedo over partially vegetated cells.

  • View in gallery

    Differences between revised total shortwave white sky soil background albedos and existing JULES soil albedos.

  • View in gallery

    Differences between total shortwave white sky land albedo computed using revised soil background and PFT albedos, and existing JULES soil and PFT albedos: (a) January and (b) July. Other information on vegetation fraction, LAI, and snow depth, taken from the HadGEM1 climate model, is also used to derive the land albedo.

  • View in gallery

    Differences between monthly average MODIS shortwave albedo, white sky surface albedo, and land albedo computed using revised and existing soil background and PFT albedos: MODIS minus revised (a) January and (b) July; MODIS minus existing (c) January and (d) July.

  • View in gallery

    Difference between reconstructed total shortwave white sky land albedo using minimum and maximum estimates of soil background albedo for (a) January and (b) July.

  • View in gallery

    Time series plots for a model grid cell in the Sahel showing the seasonal variation in (a) land surface albedo reconstructed using revised and existing soil background and PFT albedos, and MODIS albedo, and (b) total LAI for the grid cell.

  • View in gallery

    Zonal mean total shortwave white sky land albedo for (a) January and (b) July. Mean values derived from the model are produced using revised and existing values for soil background and PFT albedos.

All Time Past Year Past 30 Days
Abstract Views 0 0 0
Full Text Views 438 406 18
PDF Downloads 351 328 15

New Vegetation Albedo Parameters and Global Fields of Soil Background Albedo Derived from MODIS for Use in a Climate Model

View More View Less
  • 1 Department of Geography, Swansea University, Swansea, United Kingdom
  • | 2 The Abbey, Swansea University, Swansea, United Kingdom
  • | 3 Centre for Ecology and Hydrology, Wallingford, Oxfordshire, United Kingdom
  • | 4 Department of Geography, Swansea University, Swansea, United Kingdom
© Get Permissions
Full access

Abstract

New values are derived for snow-free albedo of five plant functional types (PFTs) and the soil/litter substrate from data acquired by the Moderate Resolution Imaging Spectroradiometer (MODIS) sensors on board Terra and Aqua. The derived albedo values are used to provide and test an improved specification of surface albedo for the land surface scheme known as the Joint U.K. Land Environment Simulator (JULES) that forms part of the Hadley Centre Global Environmental Model (HadGEM) climate model. The International Geosphere–Biosphere Programme (IGBP) global land cover map is used in combination with the MODIS albedo to estimate the albedo of each cover type in the IGBP classification scheme, from which the albedo values of the JULES PFTs are computed. The albedo of the soil/litter substrate, referred to as the soil background albedo, is derived from partially vegetated regions using a method that separates the vegetation contribution to the albedo signal from that of the soil/litter substrate. The global fields of soil background albedo produced using this method exhibit more realistic spatial variations than the soil albedo map usually employed in conjunction with the JULES model. The revised total shortwave albedo values of the PFTs are up to 8% higher than those in the existing HadGEM scheme. To evaluate the influence of these differences upon surface albedo in the climate model, differences are computed globally between mean monthly land surface albedo, modeled using the existing and revised albedo values, and MODIS data. Incorporating the revised albedo values into the model reduces the global rmse for snow-free July land surface albedo from 0.051 to 0.024, representing a marked improvement on the existing parameterization.

Corresponding author address: William M. F. Grey, Met Office, FitzRoy Road, Exeter, EX1 3PB, United Kingdom. Email: william.grey@metoffice.gov.uk

Abstract

New values are derived for snow-free albedo of five plant functional types (PFTs) and the soil/litter substrate from data acquired by the Moderate Resolution Imaging Spectroradiometer (MODIS) sensors on board Terra and Aqua. The derived albedo values are used to provide and test an improved specification of surface albedo for the land surface scheme known as the Joint U.K. Land Environment Simulator (JULES) that forms part of the Hadley Centre Global Environmental Model (HadGEM) climate model. The International Geosphere–Biosphere Programme (IGBP) global land cover map is used in combination with the MODIS albedo to estimate the albedo of each cover type in the IGBP classification scheme, from which the albedo values of the JULES PFTs are computed. The albedo of the soil/litter substrate, referred to as the soil background albedo, is derived from partially vegetated regions using a method that separates the vegetation contribution to the albedo signal from that of the soil/litter substrate. The global fields of soil background albedo produced using this method exhibit more realistic spatial variations than the soil albedo map usually employed in conjunction with the JULES model. The revised total shortwave albedo values of the PFTs are up to 8% higher than those in the existing HadGEM scheme. To evaluate the influence of these differences upon surface albedo in the climate model, differences are computed globally between mean monthly land surface albedo, modeled using the existing and revised albedo values, and MODIS data. Incorporating the revised albedo values into the model reduces the global rmse for snow-free July land surface albedo from 0.051 to 0.024, representing a marked improvement on the existing parameterization.

Corresponding author address: William M. F. Grey, Met Office, FitzRoy Road, Exeter, EX1 3PB, United Kingdom. Email: william.grey@metoffice.gov.uk

1. Introduction

Surface albedo, the ratio of the total outgoing to total incoming solar radiation at the earth’s surface, often constrained to wavelengths in the range of 0.3–3 μm, is one of the most important factors controlling the amount of energy available to drive daytime surface exchange processes. Thus, surface albedo determines the total amount of energy available for evaporation, sensible heat flux, gas exchange and, through these, more generally earth’s climate system (Lofgren 1995; Sud and Fennessy 1982). The albedo of earth’s land surface is determined locally by inter alia, variations in the following properties: (i) leaf optical properties (reflectivity and transmissivity), leaf area index (LAI), and leaf angle distribution (LAD) of the vegetation cover; (ii) the physical and chemical properties of the substrate (e.g., soil, litter, and snow) and its moisture content; (iii) the illumination geometry, particularly the solar zenith angle; and (iv) the composition of the atmosphere, which affects the relative proportions of direct and diffuse solar radiation incident on the earth’s surface (Grant et al. 2000; Asner 1998; Sellers 1985; Idso et al. 1975). The affect of these factors can be seen clearly in satellite sensor data, which demonstrate that the albedo of the land surface is both spatially variable and highly dynamic (diurnally and seasonally).

Many of the current generation of global climate models contain uncertainties or biases in their description of land surface albedo (Roesch et al. 2004; Wang et al. 2004). Some climate models specify the albedo of the land surface based on a number of in situ measurements, which have been assembled for a sample set of land cover types. These data are typically drawn from a range of studies reported in science literature, for which the individual measurements of albedo were made using different instrumentation, at different spatial locations, and at various dates and times. Climate models may also include the option to specify the albedo of the land surface using a mathematical model of the interaction of solar radiation with the vegetation canopy and the soil/litter substrate. Once again, though, the albedo parameters for the vegetation types and the soil/litter substrate are often selected from science literature. We contend that there are uncertainties and biases associated with each of these approaches, which may produce significant errors in the representation of the surface energy balance. Our aim here, therefore, is to reduce this source of uncertainty in current climate models by improving the parameterization of land surface albedo.

To this end, several recent studies have demonstrated how satellite sensor data can be combined with surface radiation models to produce improved estimates of land surface albedo for use in climate models (Pinty et al. 2007; Lawrence and Chase 2007; Liang et al. 2005). Our aim is to enhance the existing parameterization of land surface albedo in the Joint UK Land Environment Simulator (JULES) model, a version of which prescribes the land surface scheme in the Hadley Centre Global Environmental Model (HadGEM; Johns et al. 2006). JULES characterizes the land surface using a spatial tiling scheme in which individual grid cells comprise a set of discrete, homogeneous surface tiles, each representing a particular land cover type. The land cover types include five categories of vegetation, referred to as plant functional types (PFTs), urban areas, water, bare soil, and snow and ice (Table 1). In the existing JULES model, a fixed value of broadband, snow-free albedo derived from science literature is assigned to each surface type. Thus, the albedo value of a vegetated tile is simply a function of the PFT and its LAI, in which the latter introduces a seasonal signal into the resulting albedo values. The soil background albedo is derived from information on soil type and composition provided by a global vegetation and soils dataset produced at 1° resolution (Wilson and Henderson-Sellers 1985); variations in soil moisture content are not explicitly taken into account. It is known that the existing global soil albedo dataset gives a poor representation of actual spatial variations in soil albedo within soil classes. This deficiency is particularly evident over desert regions and is thought to lead to biases in the simulated regional climate. The albedo values for the constituent tiles, weighted by the fractional cover of the corresponding cover type in the grid cell, are summed to compute the albedo of the grid cell as a whole.

The JULES model also prescribes the surface albedo of snow-covered tiles under conditions of maximum and minimum vegetation cover and computes, separately, near-infrared and visible albedos under direct and diffuse illumination conditions using a two-stream approximation. In this paper, we focus on the derivation of snow-free albedo values and use snow-free data from the Moderate Resolution Imaging Spectroradiometer (MODIS) sensors on board Terra and Aqua to establish optimal albedo values for each of the JULES surface types under conditions of pure cover. We also generate a high-resolution global map of soil–litter albedo, which we refer to as the “soil background albedo.” The map of soil background albedo is intended to be compatible with other surface radiation models, which use different land cover types, cover fractions, and vegetation albedo values.

Sections 3a and 3b describe how we produce the new albedo datasets and estimate uncertainties in their values. We compute model grid albedos of heterogeneous surfaces using the new albedo values and global information on the proportions of pure PFTs and LAI taken from the HadGEM model. Differences between the revised and existing prescriptions of model grid albedo are shown in section 4c(1). Through comparison with MODIS albedo actually sensed by the satellite, we evaluate the merits and limitations of the new albedo parameters in section 4c(2). The sensitivity of the reconstructed land surface albedo values to errors in the estimated soil background albedo are examined in section 4c(3), whereas section 4c(4) considers the regional effects of the revised albedo data on climate model forcing.

2. Data

In this study, we use the combined MODIS/Terra and MODIS/Aqua global albedo product (MCD43C1) and the MODIS/Terra land cover classification product (MOD12C1). These data are taken from collection 4 of the MODIS data processing system. The MCD43C1 data are provided in climate modeling grid (CMG) format, a latitude and longitude coordinate system produced at a spatial resolution of 0.05°, which is derived from the original 1-km spatial resolution data (Gao et al. 2005). The MCD43C1 data product contains spectral albedo values for channels 1–7 of the MODIS instrument, in addition to three broadband albedo values corresponding to the visible (0.3–0.7 μm), near-infrared (0.7–5.0 μm) and total shortwave (0.3–5.0 μm) spectral regions. These data are produced at nominal 16-day intervals, resulting in 91 usable images over the period from 9 October 2002 to 19 December 2006.

It is important to note that the MCD43C1 datasets include two different measures of the surface albedo, commonly known as the “black sky” albedo (αbs) and the “white sky” albedo (αws). The former can be thought of as the albedo that would be produced under direct solar irradiation conditions—that is, no diffuse illumination; the latter represents the albedo that would be produced under diffuse illumination conditions—that is, completely isotropic incident radiation. One consequence of this is that the values of αbs depend upon the solar zenith angle, whereas those of αws do not. Note that the values of αbs in the MCD43C1 dataset are produced for direct illumination at local solar noon. The albedo under normal conditions of mixed direct and diffuse solar irradiance, α(θ,λ), can be estimated from a linear combination of the αbs and αws values weighted by the relative proportion of direct and diffuse irradiance as follows:
i1525-7541-10-1-183-e1
where S[θ, τ (λ)] is the fraction of diffuse skylight, θ is the solar zenith angle, τ is the atmospheric optical depth, and λ is the wavelength (Lewis and Barnsley 1994).

The MCD43C1 product also includes comprehensive ancillary data on the quality of the albedo values. This information is used here to select the CMG cells to be retained for further analysis. More specifically, the CMG cells that are selected meet the following three criteria: i) they are associated with 0% snow cover, meaning that 100% of the 1-km albedo data contributing to the CMG cell are derived using satellite observations classed as snow free from the 16-day period; ii) the albedo values for the majority of the 1-km pixels used to calculate the albedo of the 0.05° CMG grid cell are derived using the full bidirectional reflectance distribution function (BRDF) model inversion method (cf. the magnitude inversion method; Schaaf et al. 2002); and iii) more than 75% of the 1-km resolution pixels falling within the CMG cell contribute to the CMG cell albedo value.

Enhanced, spatially complete albedo products are also available to users and can provide additional information over areas where MODIS satellite data are sparse as a result of cloud and seasonal snow cover. The work by Moody et al. (2005) describes how these datasets are produced using interpolation techniques based upon ecosystem-dependent, temporal albedo data. For our work, we have chosen to use the MCD43C1 product for reasons of availability (collection 5 was incomplete at the time of analysis) and because the product is produced from corrected radiances using a single BRDF model. Our description of errors in reconstructed model grid albedos is based, therefore, on a comparison with the original MODIS albedo product. Statistics based upon five years (2000–04) of filled land surface albedo, as well as the original filled products, are available for public download (available online at ftp://modis-atmos.gsfc.nasa.gov). We use these statistics for white sky shortwave albedo to derive global mean values by IGBP class for comparison with the IGBP albedo values derived using our methodology (section 4a).

The MOD12C1 product describes, for a given year, the fraction of each 0.05° CMG cell that is allocated to each of the 17 IGBP land cover classes. These data are used here in two ways: 1) to provide a land–sea mask with which to stratify the global albedo datasets, and 2) as an intermediary dataset to map the albedo values observed in the MODIS data onto the PFTs used by JULES.

To reconstruct mean monthly land surface albedo, globally, using equations from the JULES model (see section 3c), we need to account for variation in PFT LAI and fractional cover. This information is taken from standard ancillary files for HadGEM version 1 (Had GEM1) at N96 (1.25° latitude by 1.875° longitude) resolution. The fractional cover dataset is time invariant and originates from the IGBP land cover map using the mapping between IGBP and JULES PFTs (Table 2). The standard LAI dataset is generated using the International Satellite Land Surface Climatology Project (ISLSCP) datasets and provides monthly LAI for each of the JULES PFTs. A furtherHadGEM1 ancillary file, providing monthly snow depth, is used to derive a snow mask. Snow amount is a climate model product.

3. Methods

a. Albedo values of JULES plant functional types

The global, geographical distribution of the nine JULES PFTs is determined by means of a mapping function from 17 IGBP land cover classes (Table 2). This mapping describes each IGBP class in terms of a linear combination of the JULES surface types and has been developed to produce realistic variation in land surface properties and hence surface fluxes for climate model forcing (Dunderdale et al. 1999). Note that the mapping includes a PFT for “crops,” which does not normally feature in the JULES scheme, but is included in this paper for completeness.

The albedo values for the surface tiles in the JULES model are computed as a linear combination of the vegetation and soil–litter (substrate) albedo values weighted by their radiative fraction. When the radiative fraction of the vegetation component is unity, as in the case of a “closed” canopy, the tile albedo is equal to the vegetation albedo. Likewise, when the radiative fraction of the vegetation component is zero—that is, the soil is bare—the tile albedo is equal to the soil albedo.

The vegetation albedo for each PFT, distinct from that of the soil background, is estimated from the MODIS albedo data by identifying areas that approximate closed canopy conditions. The first stage in this process is to select CMG cells that are dominated by a single land cover type according to the MOD12C1 land cover classification data. Here, we select those CMG cells in which greater than 90% of the contributing 1-km pixels were assigned to the same land cover type. These so-called pure CMG cells are subsequently used to compute time series of the normalized difference vegetation index (NDVI) from the MODIS channel 1 (red; 0.62–0.67 μm) and channel 2 (near-IR; 0.841–0.876 μm) albedo values. The NDVI is computed as follows:
i1525-7541-10-1-183-e2
where α is the relevant albedo value. It has been demonstrated elsewhere that NDVI is closely related to the fraction of photosynthetically active radiation (FPAR) of green vegetation and, despite a saturation effect at high vegetation densities, has been used to quantify biophysical properties such as LAI (Sellers 1985). Theoretical studies have also shown that NDVI gives a near-linear relationship with both FPAR and fractional cover under a range of canopy structures (Myneni and Williams 1994; North 2002). We use both white sky and black sky albedo datasets to derive NDVI using Eq. (2) and identify cells with the highest NDVI values in the corresponding (white sky–black sky) broadband albedo datasets. The MODIS NDVI product is calculated slightly differently using atmospherically corrected bidirectional red and near-infrared surface reflectances (Huete et al. 2002). Despite the differences, we can assume that the highest values of NDVI observed for a given land cover type corresponds to its maximum LAI. Consequently, the NDVI time series are used to calculate frequency distributions of NDVI for each land cover type and a threshold value for NDVI is set at the 98th percentile. This threshold is assumed to correspond to the maximum radiative fraction of a given cover type consistent with the JULES parameterization of albedo at maximum LAI for that cover type. The NDVI threshold relating to each land cover type is used to select a spatial sample set of visible, near-IR, and shortwave broadband albedo values for that cover type from which mean values are derived.
To derive equivalent albedo values for JULES PFTs at maximum LAI, the mapping function (Table 2) is used to define a system of equations in the form Ax = b, where A is the mapping matrix, b is the vector of MODIS albedo values derived for IGBP vegetation types at maximum LAI, and x is the vector of unknown albedo values for the JULES surface types. In our analysis, permanent wetlands, which produce poor frequency distributions of NDVI [see section 4 (a)], and nonvegetation IGBP types are excluded from matrix A, giving a total of 12 basis functions (n = 12). The x vector is also reduced to six values (m = 6), corresponding to the five JULES vegetation PFTs in addition to crops. The unknown albedo values for the JULES surface types are found by solving the system Ax = b using singular value decomposition (SVD). Sampling errors in the MODIS CMG albedo values are included during the SVD procedure by normalizing the vector b and matrix A by the standard deviations in mean observations given in Table 3, following the method described by Press et al. (1992). Errors in the fitted x values, which include the effects of errors in the IGBP albedo values, are computed from the SVD matrices as follows:
i1525-7541-10-1-183-e3
where σ(xj) are the standard deviations of the fitted parameters, Vij are the elements of the m × m orthogonal array computed via SVD, and wi are the singular values computed via SVD. This method is used to solve for xj using IGBP albedo values for visible (vis), near-IR (nir), and shortwave bands (sw).

b. Global soil background albedo

A global dataset on the soil background albedo at 0.05° resolution is also derived from the MCD43C1 albedo product. This is required to parameterize JULES, to allow correct temporal variation of surface albedo with LAI within PFTs, and to provide an updated soil albedo field in the absence of vegetative cover. Currently, the Met Office uses information on soil characteristics to derive soil albedo (Wilson and Henderson-Sellers 1985).

For bare soil regions as defined by the IGBP classification, the mean albedo for the period covered by the 91 datasets used in this study is assigned as the soil background albedo. Over bare soil regions where the mean value is calculated, the uncertainty is given as twice the standard deviation to the 95% confidence level.

For partially vegetated cells with seasonal variation in surface albedo due to vegetation phenology, a method has been developed to retrieve the soil background albedo. The method assumes that a linear relation can be found between albedo and NDVI. We then use a linear regression model to compute soil background albedo as the albedo corresponding to zero green LAI.

The form of the relationship between green fractional cover and albedo depends upon wavelength and the nature of the surface. For example, surface albedo tends to be decreased by the presence of vegetation at visible wavelengths. Over surfaces where the soil is brighter than the overlying vegetation—typical of desert areas with sandy soils—an inverse relationship is also expected between NDVI and albedo at near-infrared wavelengths (Fig. 1). Conversely, near-IR albedo may be increased by the presence of vegetation over dark soils. In this paper, NDVI is calculated for each CMG cell from narrowband albedos using the complete albedo time series. The NDVI is then transformed using the natural log and regressed against surface albedo. This model is used to derive the surface albedo at the NDVI corresponding to zero vegetation cover.

The NDVI over bare soil surfaces (identified by the IGBP classification) is remarkably consistent across the world at 0.09 (±0.03 albedo) for both white sky and black sky albedo in the three broadband channels. This was tested against a set of MCD43C1 data from 2005 and 2006 using the MODIS land cover map to identify the bare soil regions. Assuming that the bare soil regions are globally representative then we apply the same assumption below canopies and for desert and compute soil background albedo for all grid cells as the albedo at NDVI = 0.09.

The regression analysis was performed over the 91 images to derive soil background albedo at NDVI of 0.09 for each land cell in the MCD43C1 dataset for which a minimum of seven high quality albedo retrievals were available. This criterion was found to be effective at removing the poorest quality and least stable model fits. The 95% confidence interval is calculated to account for uncertainties in the linear regression parameters and in the bare soil NDVI value. To compare the new soil background albedo dataset with standard JULES soil albedo, the 0.05° product is regridded to N96 (1.25° latitude by 1.875° longitude), a standard resolution used by the HadGEM climate models. The lower resolution product is produced by aggregating cells from the new soil background albedo and computing average gridbox albedo wherever a minimum of 10% of contributing cells are available. The threshold value of 10% was found to be effective in removing cells dominated by low-quality soil background albedo in densely forested areas.

c. Constructing JULES land albedo

The JULES model computes total grid cell albedo as a linear combination of the albedos of surface tiles located within the cell. The tile albedos are weighted according to their fractional cover, and the radiative fraction is used to compute the individual tile albedos. Radiative fraction, fr, accounts for incomplete cover and is taken to be the interception according to a Beer’s law–type expression as follows:
i1525-7541-10-1-183-e4
where kpar is the extinction coefficient assumed to take a value of 0.5 appropriate for canopies with spherical leaf angle distribution (Ross 1981). Thus, model-dependent land surface albedo, αland, is constructed from the revised PFT and soil background albedos in conjunction with global information on vegetation fractional cover and LAI using the following equations:
i1525-7541-10-1-183-e5
where Fj is the cover fraction for PFT j, αj is the tile albedo, and vegetated tiles αj is computed from
i1525-7541-10-1-183-e6
where αsoil is the soil albedo and αveg is the snow-free vegetation (PFT) albedo. The affects of snow upon surface albedos are not included in the analysis. Because the purpose of this analysis was to examine how the new satellite-derived parameters affect the surface albedo forcing data appropriate to a climate model, fractional cover and LAI information are taken from standard ancillary files for the HadGEM climate models.

4. Results

a. Albedos of JULES plant functional types

Frequency distributions of NDVI over “pure” IGBP land cover types are shown in Fig. 2. These plots are used to illustrate the considerable variation between distributions for different IGBP classes. Evergreen broadleaf forest (class 2), which is dominated by low-latitude tropical forest, shows the smallest variation in NDVI. Classes with a seasonal cycle show expected bimodal distributions. For most classes, the maximum NDVI values are expected to correspond approximately to 100% cover. This may not be true for shrub and grassland classes, and in particular, globally averaged fractional cover for open shrub derived from 1-km satellite data has been found to be only 39% (Zeng et al. 2000). The NDVI threshold at the 98th percentile, corresponding to maximum LAI for these cover types, may therefore incorporate partially vegetated cells.

Mean and standard deviation values for MODIS albedo aggregated over so-called pure IGBP cells for which NDVI is greater than the threshold value are given in Table 3. Albedo variability is greatest for land cover classes such as open shrub (class 7) and deciduous needleleaf forest (class 3) for which a variety of types, structures, cover fractions, and leaf areas may be typical. The method produces sensible mean albedo values for the different IGBP classes, which compare well with literature values for a few key types [e.g., diffuse shortwave albedos for Amazonian forest = 0.134 (Culf et al. 1995) and Boreal forest dominated by aspen = 0.15 and by conifer = 0.083 (Betts and Ball 1997)]. Also included in Table 3 are global mean white sky shortwave albedos by IGBP class taken from the Moody et al. (2005) filled albedo products. In general, the differences between the Moody et al. (2005) global mean albedo values and the albedos we derive for “pure” cover types are less than 2% albedo and well within the standard deviation. This is despite the Moody et al. (2005) data coming from a much greater sample set with a range of radiative fractions for each IGBP class. The greatest differences between the Moody et al. (2005) global mean albedo values and the albedos we derive for “pure” cover occur for ecosystem classes with sparse cover (e.g., open shrub, class 7) or a strong seasonal signal (e.g., deciduous broadleaf forest, class 4). Over evergreen broadleaf forest (class 2), which we expect to have the least variation in albedo, the difference is 0.7%. These results suggest that using the filled datasets to derive albedo for “pure” cover types would make little difference for some classes but could be important for the accurate representation of global ecosystem classes that show large variability.

Mean values for pure cover from Table 3 are used to derive albedos for the JULES PFTs using SVD, and the results are shown in Table 4. Fitted values of total shortwave albedo decrease by up to 8% in the new parameter set compared with standard model values (Table 1). For PFT-type needleleaf forest, the fitted shortwave albedo of 0.088 is at the low end of field observations for various species (Betts and Ball 1997), and may be a result of biasing toward a particular type of vegetation within the class. Variability shown in Table 4 is of the order 20%–40% relative albedo and highlights the limitations of using single albedo values to represent a small number of JULES PFTs.

Given that large errors are associated with the albedo values derived for IGBP class open shrub (Table 3), we replace albedo values for open shrub with those for closed shrub and examine the difference it makes to the resulting JULES PFT albedos that map onto the IGBP classes. Results for the white sky shortwave albedo are given in Table 4 and show that the replacement makes very little difference to the JULES PFT albedos. We have, therefore, included the albedos for open shrub in our analysis.

b. Global soil background albedo

The soil background albedo dataset contains six albedo values per grid cell—that is, black sky and white sky albedos in each of the broadband visible, near-IR, and total shortwave regions—and uncertainty for each term. The total shortwave white sky soil background albedo and corresponding error map are shown in Fig. 3. Arid regions—the Sahara Desert and Saudi Arabia, in particular—show the highest soil background albedo values. Uncertainties are smallest over regions characterized by a strong seasonal signal and thus high variability of NDVI. The largest uncertainties occur over areas of dense evergreen vegetation where our method cannot be used to obtain reliable estimates for soil background albedo. These areas are mainly in the tropics (Amazonia, central Africa, and South East Asia), which also coincide with reduced numbers of good quality retrievals due to cloud cover, and boreal regions (Siberia and North America) where snow and cloud both reduce the number of good-quality snow-free retrievals. For example, the soil albedo map in Fig. 3 suggests that we tend to underestimate soil background albedo over large areas of Eurasia where soil background albedos are below 0.05. For many of these regions, however, the JULES model does not need such accurate estimates of soil background albedo because the vegetation radiative fraction fr is always large.

The difference between the revised and existing JULES global soil background albedo is shown in Fig. 4. We compare the standard soil albedo dataset with the new diffuse white sky albedo product because this is independent of solar geometry. Large negative differences, corresponding to a reduction in soil background albedo in the new product, over the boreal zone, northern Europe, and Patagonia coincide with regions associated with larger uncertainties in the new product and/or dense vegetation. Other differences, particularly over desert regions of West Africa, parts of Asia, and Australia, are plausible and suggest that the existing model currently has poor representation of soil albedo variation. Other studies have also used satellite observations to show there is considerable spatial variability of surface albedo over desert regions (Tsvetsinskaya et al. 2002).

c. JULES land albedo

1) Comparisons with standard land albedo

Figure 5 shows January and July differences between land albedo computed using revised and existing JULES PFT and soil background albedos, accounting for differences in LAI taken from the standard HadGEM ancillary files. Regions where soil albedo cannot be derived from the new product are filled with values from the standard model dataset. Albedo data relate to snow-free shortwave white sky albedo. To be consistent with the climate model’s prescription of the land surface, we apply the HadGEM snow mask to the difference plot for January. No equivalent mask is used for the July difference map, when the fraction of snow-covered land pixels is at a minimum, to reduce loss of information due to a conservative snow mask. Crops are not included in the analysis.

Areas of positive and negative difference are very similar in the January and July difference maps. Seasonal differences between the January–July difference plots, as a result of the revised albedo values having differing effects under different LAIs, are expected to be more important at latitudes where the vegetation has a strong seasonal cycle. The greatest differences between revised and existing land albedos occur over desert and semidesert regions and may be largely attributed to the new soil background albedo. Shrub vegetation is also important in these regions, so this will also have an influence as a result of the decrease in shrub albedo in the revised PFT albedo set. PFT-type broadleaf forest is altered least in the revised parameter set and over tropical forest, where the contribution from the soil background to the total land albedo is negligible and the difference in July land albedo is close to zero. At mid-to-high latitudes, July land albedo differences are generally negative and exceed −0.025, which can be explained by the decrease in PFT albedos compared with standard albedo values (Tables 1 and 4).

2) Comparisons with MODIS albedo

We evaluate the revised and existing parameters through comparison of land albedo for January and July modeled using revised and existing albedo parameters, with monthly averaged MODIS surface albedo (Fig. 6). The differences in Fig. 6 may be attributed to not only poor representation of surface albedo but also to land cover types and LAI. These uncertainties are expected to be much greater than uncertainties in the MODIS data because we only use good quality retrievals that have sufficient information available for performing a full BRDF inversion (Schaaf et al. 2002). Missing values over land coincide with regions where good quality albedo estimates cannot be retrieved from the satellite data as a result of cloud cover and illumination conditions (high latitudes), and regions of permanent snow and ice.

Differences between MODIS albedo and land albedo derived using existing albedo parameters (Figs. 6c and 6d) are large over desert and semidesert regions and considerably reduced in the difference maps produced using the revised global soil background albedo (Figs. 6a and 6b). Regions where there is a significant improvement include western Australia, North and West Africa, the Red Sea, and central Asia. Over boreal forest, where the July land albedo modeled using existing PFT albedo parameters is too bright, incorporation of the revised parameters generally reduces differences compared with MODIS albedo. Improvements are also notable over regions where mosaics of forest, grassland, and savanna are typical over tropical and subtropical regions of central Africa and South America.

In the maps produced from the difference between MODIS albedo and land albedo derived using the revised albedo parameters (Figs. 6a and 6b), there is a large region of positive difference over the eastern United States. This anomaly does not appear in the existing July land albedo difference map or as a negative bias in the existing January land albedo difference map. The poor representation of this region may be explained by the complex albedo signal typical of deciduous and mixed forests, which is not captured by existing nor revised albedo parameters. Errors in the soil background albedo are less important (see Fig. 7). Some bias also remains over the bare soil of the Sahara Desert, suggesting deficiencies in the method we use to define soil background albedo, and in the January difference over desert and xeric shrublands of central Asia. Over the majority of the land surface, however, the albedo differences are less than ±0.025, which is comparable with the expected accuracy of the satellite data of the order 0.02 absolute albedo (Susaki et al. 2007; Salomon et al. 2006; Jin et al. 2003; Barnsley et al. 2000). Table 5 shows the global rmse for reconstructed snow-free land surface albedo, calculated from the difference between MODIS albedo actually sensed and land albedo derived using the revised and existing albedo parameters. Incorporating the revised albedo values into the model reduces the global rmse from 0.065 to 0.025 and from 0.051 to 0.024 for January and July, respectively. This suggests that the revised albedo parameters will produce a more realistic land surface albedo in the model than the existing parameters.

3) Sensitivity to errors in soil background albedo

Total shortwave white sky land surface albedo was reconstructed using maximum and minimum estimates for soil background albedo by adding and subtracting the derived uncertainty. The purpose of this analysis was to examine which areas of the world are sensitive to soil background estimates of albedo at different times of year. Figure 7 shows the difference between land surface albedo modeled using maximum and minimum estimates for soil background albedo for January and July. Missing values correspond to pixels where soil background albedo cannot be retrieved or pixels where the maximum–minimum soil background corresponds to negative values. As expected, differences are small over regions where LAI is high (e.g., boreal regions during summer and tropical regions). We conclude, therefore, that the model will also be relatively insensitive to errors in the soil background albedo over these regions. Conversely, large differences occur where LAI is low (arid and semiarid regions), and we infer a corresponding increase in model sensitivity to errors in the soil background albedo. Differences between the maximum and minimum estimates of albedo are generally within 10% albedo.

4) Implications for climate model forcing

We have shown that the greatest effect of the new albedo parameters upon reconstructed land surface albedo occurs in subtropical regions due to the influence of the improved global soil background albedo compared with the existing soil albedo map. To show the seasonal affect of the new parameters in this region, we generate time series plots of LAI and albedo for a model grid cell in the Sahel (Fig. 8). The model albedo reconstructed using the new datasets is much closer to the MODIS white sky shortwave albedo compared to model albedo reconstructed using the existing soil albedo map and PFT albedo values. Figure 8a also shows that poor representation of surface albedos, as in the reconstruction using the existing albedo parameters, can result in the seasonal cycle in albedo being lost when the vegetation and soil albedos are similar.

To examine how the improvements in the revised soil background and PFT albedos affect land surface albedo at a regional scale, modeled land surface albedo fields for January and July were averaged over 10° latitude bands. The results shown in Fig. 9 also include zonally averaged MODIS surface albedos.

At low latitudes, zonal mean land surface albedos using revised and existing albedo parameters are similar. For the subtropical latitude band 20°–30°N, modeled zonal albedos diverge and the zonal mean land surface albedo reconstructed using the existing albedo parameters is closer to zonal mean MODIS albedo. We conclude that this result is due to averaging over a large latitude band with considerable range in surface albedos because global analysis suggests that the new parameterization reduces errors in reconstructed land surface albedo (Table 5). At mid-to-high latitudes, there is a net decrease in zonal mean land surface albedo derived using the revised soil background and PFT albedos compared with existing values, and improved agreement with the zonal mean MODIS albedo. Associated with this decrease in surface albedo will be an increase in absorbed solar radiation. For a July mean insolation of 200 W m−2, appropriate for July 1982 around 50°–60°N and accounting for atmospheric absorption, daylength, and angle of incidence (Pinker and Laszlo 1992), a decrease in land surface albedo of 0.045 (Fig. 9b) equates to a positive radiative forcing of 9 Wm−2. This forcing represents an important uncertainty in the Met Office atmospheric model using the existing soil and PFT albedos and may be expected to affect simulations of climate.

5. Conclusions

The revised albedo products developed in our study have significantly improved the global representation of land surface albedo produced by the JULES surface scheme as it is used by the HadGEM climate model. Our approach is compatible with the radiation option in the JULES model that computes surface albedo as a linear combination of soil and vegetation albedos. The advantage of this simple radiation scheme is that is does not rely upon large numbers of radiation parameters that require sophisticated techniques to be retrieved simultaneously (Pinty et al. 2007; Liang et al. 2005). The approach does, however, make broad assumptions about the radiative transfer process and how it scales for canopies with different densities and, by using a very small number of PFT classes, does not account for regional variations in similar vegetation types. The method used to derive soil background albedo makes assumptions about the form of the relationship between NDVI and albedo, both of which are derived from the same satellite product, and that soil–litter albedo is constant. Despite these assumptions, large improvements have been made in surface albedo representation over desert and semidesert regions, and global biases in the new land surface albedo are generally less than ±0.025. Improvements have also been made over large vegetated areas, suggesting that the existing PFT albedo values previously used by the HadGEM1 model were not globally applicable. We expect future improvements will be made by directly using a land cover classification for the JULES surface types that includes more PFT classes, the global map of the Food and Agriculture Organization’s (FAO) soil types to refine the soil albedo algorithm parameters regionally, and the MODIS data to derive model parameters for more sophisticated radiation schemes (Alton et al. 2005).

Acknowledgments

CH and WG are supported by the UK Natural Environment Research Council through the Climate Land Surface systems Interaction Centre (CLASSIC). The MODIS data have been obtained from the EROS Data Center (EDC) Data Active Archive Center (DAAC). The authors thank Dr. Pier Luigi Vidale for supplying the HadGEM ancillary files. Global fields of soil background albedo are available from the authors on request. We are indebted to MB for his contribution to this work and to whom we dedicate this paper.

REFERENCES

  • Alton, P., North P. , Kaduk J. , and Los S. , 2005: Radiative transfer modeling of direct and diffuse sunlight in a Siberian pine forest. J. Geophys. Res., 110 , D23209. doi:10.1029/2005JD006060.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Asner, G., 1998: Biophysical and biochemical sources of variability in canopy reflectance. Remote Sens. Environ., 64 , 234253.

  • Barnsley, M., Hobson P. , Hyman A. , Lucht W. , Muller J. , and Strahler A. , 2000: Characterizing the spatial variability of broadband albedo in a semidesert environment for MODIS validation. Remote Sens. Environ., 74 , 5868.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Betts, A., and Ball J. , 1997: Albedo over the boreal forest. J. Geophys. Res., 102 , 2890128909.

  • Culf, A., Fisch G. , and Hodnett M. , 1995: The albedo of Amazonian forest and ranch land. J. Climate, 8 , 15441554.

  • Dunderdale, M., Muller J-P. , and Cox P. , 1999: Sensitivity of the Hadley Centre Climate Model to different Earth observation and cartographically derived land surface data-sets. Proc. Int. Conf. and Workshops on Ocean Color, Land Surfaces, Radiation and Clouds, Aerosols (ALPS ’99): The Contribution of POLDER and New Generation Spaceborne Sensors to Global Change Studies, Vol. 1–6; Meribel, France, Centre National d’Etudes Spatiales, 1–4.

    • Search Google Scholar
    • Export Citation
  • Gao, F., Schaaf C. , Strahler A. , Roesch A. , Lucht W. , and Dickinson R. , 2005: MODIS bidirectional reflectance distribution function and albedo Climate Modeling Grid products and the variability of albedo for major global vegetation types. J. Geophys. Res., 110 , D01104. doi:10.1029/2004JD005190.

    • Search Google Scholar
    • Export Citation
  • Grant, I., Prata A. , and Cechet R. , 2000: The impact of the diurnal variation of albedo on the remote sensing of the daily mean albedo of grassland. J. Appl. Meteor., 39 , 231244.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Huete, A., Didan K. , Miura T. , Rodriguez E. , Gao X. , and Ferreira L. , 2002: Overview of the radiometric and biophysical performance of the MODIS vegetation indices. Remote Sens. Environ., 83 , 195213.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Idso, S., Jackson R. , Reginato R. , Kimball B. , and Nakayama F. , 1975: The dependence of bare soil albedo on soil water content. J. Appl. Meteor., 14 , 109113.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Jin, Y., Schaaf C. , Woodcock C. , Gao F. , Li X. , Strahler A. , Lucht W. , and Liang S. , 2003: Consistency of MODIS surface bidirectional reflectance distribution function and albedo retrievals: 2. Validation. J. Geophys. Res., 108 , 4159. doi:10.1029/2002JD002804.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Johns, T., and Coauthors, 2006: The new Hadley Centre Climate Model (HadGEM1): Evaluation of coupled simulations. J. Climate, 19 , 13271353.

  • Lawrence, P., and Chase T. , 2007: Representing a new MODIS consistent land surface in the Community Land Model (CLM 3.0). J. Geophys. Res., 112 , G01023. doi:10.1029/2006JG000168.

    • Search Google Scholar
    • Export Citation
  • Lewis, P., and Barnsley M. , 1994: Influence of the sky radiance distribution on various formulations of the earth surface albedo. Proc. Sixth ISPRS Int. Symp. on Physical Measurements of Signatures in Remote Sensing, Val d’Isère, France, Centre National d’Etudes Spatiales, 707–715.

    • Search Google Scholar
    • Export Citation
  • Liang, X-Z., and Coauthors, 2005: Development of land surface albedo parameterization based on Moderate Resolution Imaging Spectroradiometer (MODIS) data. J. Geophys. Res., 110 , D11107. doi:10.1029/2004JD005579.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Lofgren, B., 1995: Surface albedo–climate feedback simulated using two-way coupling. J. Climate, 8 , 25432562.

  • Moody, E., King M. , Platnick S. , Schaaf C. , and Gao F. , 2005: Spatially complete global spectral surface albedos: Value-added datasets derived from Terra MODIS land products. IEEE Trans. Geosci. Remote Sens., 43 , 144158.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Myneni, R., and Williams D. , 1994: On the relationship between FAPAR and NDVI. Remote Sens. Environ., 49 , 200211.

  • North, P. R. J., 2002: Estimation of fAPAR, LAI, and vegetation fractional cover from ATSR-2 imagery. Remote Sens. Environ., 80 , 114121.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Pinker, R., and Laszlo I. , 1992: Modeling surface solar irradiance for satellite applications on a global scale. J. Appl. Meteor., 31 , 194211.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Pinty, B., and Coauthors, 2007: Retrieving surface parameters for climate models from Moderate Resolution Imaging Spectroradiometer (MODIS)-Multiangle Imaging Spectroradiometer (MISR) albedo products. J. Geophys. Res., 112 , D10116. doi:10.1029/2006JD008105.

    • Search Google Scholar
    • Export Citation
  • Press, W., Flannery B. , Teukolsky S. , and Vetterling W. , 1992: Numerical Recipes in FORTRAN 77: The Art of Scientific Computing. 2nd ed. FORTRAN Numerical Recipes Series, Vol. 1, Cambridge University Press, 992 pp.

    • Search Google Scholar
    • Export Citation
  • Roesch, A., Schaaf C. , and Gao F. , 2004: Use of Moderate-Resolution Imaging Spectroradiometer bidirectional reflectance distribution function products to enhance simulated surface albedos. J. Geophys. Res., 109 , D12105. doi:10.1029/2004JD004552.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Ross, J., 1981: The Radiation Regime and Architecture of Plant Stands. Tasks for Vegetarian Science Series, Vol. 3, Kluwer Academic, 391 pp.

    • Search Google Scholar
    • Export Citation
  • Salomon, J., Schaaf C. , Strahler A. , Gao F. , and Jin Y. , 2006: Validation of the MODIS bidirectional reflectance distribution function and albedo retrievals using combined observations from the Aqua and Terra platforms. IEEE Trans. Geosci. Remote Sens., 44 , 15551565.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Schaaf, C., and Coauthors, 2002: First operational BRDF, albedo nadir reflectance products from MODIS. Remote Sens. Environ., 83 , 135148.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Sellers, P., 1985: Canopy reflectance, photosynthesis and transpiration. Int. J. Remote Sens., 6 , 13351372.

  • Sud, Y., and Fennessy M. , 1982: A study of the influence of surface albedo on July circulation in semi-arid regions using the GLAS GCM. Int. J. Climatol., 2 , 105125.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Susaki, J., Yasuoka Y. , Kajiwara K. , Honda Y. , and Hara K. , 2007: Validation of MODIS albedo products of paddy fields in Japan. IEEE Trans. Geosci. Remote Sens., 45 , 206217.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Tsvetsinskaya, E., Schaaf C. , Gao F. , Strahler A. , Dickinson R. , Zeng X. , and Lucht W. , 2002: Relating MODIS-derived surface albedo to soils and rock types over Northern Africa and the Arabian peninsula. Geophys. Res. Lett., 29 , 1353. doi:10.1029/2001GL014096.

    • Search Google Scholar
    • Export Citation
  • Wang, Z., Zeng X. , Barlage M. , Dickinson R. , Gao F. , and Schaaf C. , 2004: Using MODIS BRDF and albedo data to evaluate global model land surface albedo. J. Hydrometeor., 5 , 314.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Wilson, M., and Henderson-Sellers A. , 1985: A global archive of land cover and soils data for use in general circulation climate models. Int. J. Climatol., 5 , 119143.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Zeng, X., Dickinson R. , Walker A. , Shaikh M. , DeFries R. , and Qi J. , 2000: Derivation and evaluation of global 1-km fractional vegetation cover data for land modeling. J. Appl. Meteor., 39 , 826839.

    • Crossref
    • Search Google Scholar
    • Export Citation

Fig. 1.
Fig. 1.

Relationship between MODIS broadband albedos and NDVI for a location in the Sahel. NDVI is computed from MODIS white sky albedos in narrowband channels 1 and 2 [Eq. (2)]: (a) visible, (b) near-infrared, and (c) shortwave.

Citation: Journal of Hydrometeorology 10, 1; 10.1175/2008JHM1021.1

Fig. 2.
Fig. 2.

Frequency distributions of NDVI over “pure” IGBP cells corresponding to >90% of a single IGBP class for 14 different IGBP classes. NDVI is computed from MODIS white sky albedos in narrowband channels 1 and 2 [Eq. (2)] for the entire period from 2002 to 2006. The data are binned into NDVI intervals of 0.01.

Citation: Journal of Hydrometeorology 10, 1; 10.1175/2008JHM1021.1

Fig. 3.
Fig. 3.

Global total shortwave white sky soil background (a) albedos and (b) associated error map. Uncertainty corresponds to the 95% confidence interval derived from mean values over bare soil and derived from the linear regression model of NDVI against surface albedo over partially vegetated cells.

Citation: Journal of Hydrometeorology 10, 1; 10.1175/2008JHM1021.1

Fig. 4.
Fig. 4.

Differences between revised total shortwave white sky soil background albedos and existing JULES soil albedos.

Citation: Journal of Hydrometeorology 10, 1; 10.1175/2008JHM1021.1

Fig. 5.
Fig. 5.

Differences between total shortwave white sky land albedo computed using revised soil background and PFT albedos, and existing JULES soil and PFT albedos: (a) January and (b) July. Other information on vegetation fraction, LAI, and snow depth, taken from the HadGEM1 climate model, is also used to derive the land albedo.

Citation: Journal of Hydrometeorology 10, 1; 10.1175/2008JHM1021.1

Fig. 6.
Fig. 6.

Differences between monthly average MODIS shortwave albedo, white sky surface albedo, and land albedo computed using revised and existing soil background and PFT albedos: MODIS minus revised (a) January and (b) July; MODIS minus existing (c) January and (d) July.

Citation: Journal of Hydrometeorology 10, 1; 10.1175/2008JHM1021.1

Fig. 7.
Fig. 7.

Difference between reconstructed total shortwave white sky land albedo using minimum and maximum estimates of soil background albedo for (a) January and (b) July.

Citation: Journal of Hydrometeorology 10, 1; 10.1175/2008JHM1021.1

Fig. 8.
Fig. 8.

Time series plots for a model grid cell in the Sahel showing the seasonal variation in (a) land surface albedo reconstructed using revised and existing soil background and PFT albedos, and MODIS albedo, and (b) total LAI for the grid cell.

Citation: Journal of Hydrometeorology 10, 1; 10.1175/2008JHM1021.1

Fig. 9.
Fig. 9.

Zonal mean total shortwave white sky land albedo for (a) January and (b) July. Mean values derived from the model are produced using revised and existing values for soil background and PFT albedos.

Citation: Journal of Hydrometeorology 10, 1; 10.1175/2008JHM1021.1

Table 1.

JULES PFTs.

Table 1.
Table 2.

Mapping for IGBP to JULES surface types.

Table 2.
Table 3.

Mean values for MODIS albedo aggregated over pure IGBP cells where NDVI is greater than the 98th percentile NDVI taken from frequency distributions for each IGBP class. Values in parentheses indicate the standard deviation. Also included are global mean white sky albedos by IGBP class derived from filled land surface albedo products (Moody et al. 2005).

Table 3.
Table 4.

MODIS albedo optimized for JULES PFTs. Values in parentheses indicate the standard deviation in the fitted albedo values. Also shown are the JULES PFTs albedos for white sky shortwave albedo where the albedo for open shrub in the array of IGBP albedos used as input to the SVD procedure is replaced with that for closed shrub (labeled version 2).

Table 4.
Table 5.

Global rmse for reconstructed land surface albedo compared with actual MODIS albedo by month.

Table 5.
Save