Modeling shortwave radiative fluxes from satellites



[1] During the last two decades, significant progress has been made in assessing the Earth Radiation Balance from satellite observations. Yet, satellite based estimates differ from each other and from those provided by numerical models. Major issues are related to quality of satellite observations, such as the frequent changes in satellite observing systems, degradation of sensors, restricted spectral intervals and viewing geometry of sensors, and changes in the quality of atmospheric inputs that drive the inference schemes. To reduce differences among the satellite based estimates requires, among others, updates to inference schemes so that most recent auxiliary information can be fully utilized. This paper reports on improvements introduced to a methodology developed at the University of Maryland to estimate shortwave (SW) radiative fluxes within the atmosphere system from satellite observations, the implementation of the approach with newly available auxiliary information, evaluation of the downwelling SW flux against ground observations, and comparison with independent satellite methods and numerical models. Specifically, introduced are: new Narrow to Broadband (N/B) transformations and new Angular Distribution Models (ADM) for clear and cloudy sky that incorporate most recent land use classifications; improved aerosol treatment; separation of clouds by phase; improved sun-earth geometry; and implementation at 0.5° spatial resolution at 3-hourly intervals integrated to daily and monthly time scales. When compared to an earlier version of the model as implemented at 2.5° at global scale and against observations from the globally distributed Baseline Surface Radiation Network (BSRN) stations for a period of six years (at monthly time scale), the bias was reduced from 8.6 (4.6%) to −0.5 (0.3%) W/m2, the standard deviation from 16.6 (8.9%) to 14.5 (7.8%) W/m2while the correlation remained high at 0.98 in both cases. Evaluation was also done over oceanic sites as available from the Pilot Research Moored Array in the Tropical Atlantic (PIRATA) moorings and from the Tropical Atmosphere Ocean/Triangle Trans-Ocean Buoy Network (TAO/TRITON) moorings in the tropical Pacific Ocean. Overall, results over oceans were not as good as over land for all the satellite retrievals compared in this study.

1. Introduction

[2] The incoming shortwave radiation (SW) from the sun that reaches the Earth's surface determines the exchange of energy between land, oceans and the atmosphere and consequently, controls the hydrologic cycle. Environmental satellites are considered as useful tools for providing information on surface radiative fluxes at various temporal and spatial scales, allowing improvement in estimation of terrestrial water and energy storage and oceanic heat flux [Trenberth et al., 2011]. The research plan of the World Climate Research Programme (WCRP) [World Meteorological Organization (WMO), 1985] serves as an interface with national and international programs with respect to energy and water cycle climate research requirements. Information on the various SW components of atmospheric and surface radiative fluxes has been available for more than two decades [Wielicki et al., 1996]. As stated in numerous National and International Climate Research Program plans (i.e., the NASA Earth Science Research Strategy) the need is for long-term, consistent, and calibrated data and products that are valid across multiple missions and satellite sensors.

[3] Methods spanning a wide range of complexity have been developed to derive surface radiative fluxes from satellite observations. Many attempts have been made to estimate surface shortwave (SW) fluxes at both regional and global scales [Pinker and Ewing, 1985; Ramanathan, 1986; Raschke et al., 1991; Pinker and Laszlo, 1992; Li and Leighton, 1993; Stephens et al., 1994; Zhang et al., 1995, 2004; Gupta et al., 1999; Mueller et al., 2004; Rigollier et al., 2004; Hinkelman et al., 2009; Wang and Pinker, 2009; Hatzianastassiou et al., 2007; Vardavas and Taylor, 2007]. The need for a systematic evaluation of such fluxes has been recognized and in 2004 the World Climate Research Program (WCRP) has established the Radiative Flux Assessment (RFA) Working Group to address these issues ( Available long-term global scale estimates are based primarily on inputs from the International Satellite Cloud Climatology Project (ISCCP) (e.g., C1, D1, D2, and DX versions) [Rossow and Schiffer, 1991, 1999]. For instance, global information on radiative fluxes known as ISCCP-FD has been produced at the Goddard Institute for Space Studies (GISS) at three hourly intervals on a 280 km equal-area global grid [Zhang et al., 2004]. Utilized is a radiative transfer model from the Goddard Institute for Space Studies (GISS) General Circulation Model (GCM) with the atmospheric properties primarily from the TIROS Operational Vertical Sounding (TOVS) data. The GEWEX/SRB product at NASA LaRC at 1° resolution [Hinkelman et al., 2009] and the University of Maryland (UMD)/Shortwave Radiation budget (SRB) product at 2.5° resolution [Pinker et al., 2005] use TOA radiances, transform them into broadband albedos to infer atmospheric transmissivity which in turn, determines the amount of radiation that reaches the surface. The atmospheric products used in the UMD/SRB model are also from TOVS while GEWEX/SRB uses inputs from the Global Modeling and Assimilation Office (GMAO) [Norris and da Silva, 2007]. Each model treats surface properties independently. The resulting radiative fluxes are archived at the Langley Radiative Flux Assessment Archive ( Information on radiative fluxes is also available under the Clouds and the Earth's Radiant Energy System (CERES) activity [Wielicki et al., 1996]. A 1° product is described at: ( [Loeb et al., 2005; Young et al., 1998]. CERES data from the RFA website are currently available for the period March 2000 through October 2005. Products on surface fluxes are also available from more recent satellite observations such as the Moderate Resolution Imaging Spectroradiometer (MODIS) [Wang and Pinker, 2009], and at the Top of the Atmosphere (TOA) from the Scanner for Radiation Budget (ScaRaB) [Duvel et al., 2001] and from the Geostationary Earth Radiation Budget (GERB) instruments [Harries et al., 2005]. Data from European Geostationary satellites, specifically, from the Meteosat Secon Generation (MSG) series are available from the Satellite Applications Facilities (SAFs) ( However, some of these are at regional scale only and/or for a shorter time period than those based on ISCCP. Information resulting from the various ISCCP products is widely used [e.g., Rodriguez-Puebla et al., 2008; Trenberth et al., 2009].

[4] Early inference schemes have been reviewed in a series of publications starting with Schmetz [1989, 1991, 1993] who critically evaluates sensitivities to input parameters, as well as physical principles of the methodologies. This was followed with papers by Pinker et al. [1995] and Whitlock et al. [1995]. A summary of current and future satellite observations of relevance for radiation budget research is presented in Wielicki et al. [1995]. Additional relevant work is reported on in Charlock and Alberta [1996], Wielicki et al. [1998], and Wang and Pinker [2009].

[5] In section 2 presented is the updated methodology; in section 3described is the implementation of the new methodology at global scale, at 3-hourly intervals, and at 0.5° resolution as driven with the ISCCP DX data; results are discussed insection 4; and a summary is presented in section 5.

2. Methodology

2.1. Brief Summary of Earlier Version of UMD/SRB Model

[6] We utilize a Look-Up Table (LUT) approach of matching radiative transfer computations and satellite observations to infer the desired radiative fluxes at the Top of the Atmosphere (TOA) and at the surface. The LUTs contain discrete values of atmospheric transmissivity and reflectivity in five broadband intervals as a function of solar zenith angle, amount of water vapor and ozone, aerosol single scattering albedo, asymmetry factor, optical depth and cloud optical depth. The LUTs are computed for a plane-parallel, vertically inhomogeneous, scattering and absorbing atmosphere as originally described inPinker and Ewing [1985] and updated in Pinker and Laszlo [1992]. The model accounts for (1) absorption by ozone and water vapor; (2) Rayleigh scattering; (3) multiple scattering and absorption by aerosols and cloud droplets; and (4) multiple reflection between the atmosphere and surface. It has five or six vertical layers, depending on the aerosol profile considered and whether a cloud is present. Multiple scattering is treated using the delta-Eddington approximation [Joseph et al., 1976]. Water vapor absorption is accounted for in the 0.7–4.0 μm spectral interval. Ozone is accounted for in 0.2–0.4 μm (UV) and in 0.5–0.6 μm (VIS) spectral intervals. The amount of ozone and water vapor are taken from the Standard Atmospheres (tropical, midlatitude summer and winter, sub-arctic summer and winter) [Kneizys et al., 1980]. Rayleigh scattering follows Penndorf [1957] and Wiscombe et al. [1984]. Cloud optical properties are prescribed according to the parameterization of Stephens [1979] and Stephens et al. [1984].

[7] In the UMD/SRB model the shortwave (SW) radiative fluxes are inferred in 5 broadband intervals (0.2–0.4, 0.4–0.5, 0.5–0.6, 0.6–0.7, and 0.7–4.0 μm). The radiation in the spectral interval of (0.4–0.7 μm) is referred to as Photosynthetically Active Radiation (PAR). The radiation in the spectral interval of (0.7–4.0 μm) is known as the Near-Infra-Red radiation (NIR). These components are of interest due to the differences in their biological impacts and their distinct interaction with vegetation and water bodies. Stored information includes SW, PAR, and NIR surface radiative fluxes, both upwelling and downwelling. These are provided as total and diffuse fluxes at the surface and upwelling fluxes at the TOA for clear and all sky conditions. The 1992 version was partially updated and implemented with ISCCP D1 inputs (v3.3). The following improvements have been introduced: elevation correction as reflected in new LUT; updated sun-earth geometry; and new aerosol treatment. Results have been submitted to theRadiative Flux Assessment evaluation archive at NASA Langley research Center ( Since these features are also part of the version used in this study (not described previously in open literature) they will be described in more detail in section 2.2.

2.2. The Updated Inference Scheme (v3.3.3)

2.2.1. Treatment of Aerosols

[8] In the original version of the UMD/SRB model, variable aerosol optical depth was accounted for while only preset values of single-scattering albedo (SSA) (ω), and asymmetry factor (ASYM) (g) were used. The aerosol module in the inference scheme was modified so that new aerosol information can be utilized by expanding the data library of clear-sky atmospheric optical functions to include variations of aerosol SSA and ASYM parameter [Liu and Pinker, 2008]. The new LUTs include aerosol single-scattering albedo (ω) and asymmetry factor (g) as two new dimensions allowing to represent their variability. For a clear atmosphere with aerosol loading, the transmission and reflection tables are computed for 5 single-scattering albedo values and 4 asymmetry factor values for two standard aerosol models, namely CONT-I for land and MAR-I for ocean and snow or ice covered surface (provided byWMO [1986]).

[9] For the SSA, the specified values at 0.55 μm are 0.200, 0.500, 0.750, 0.893/0.989, 1.000, where ω0.55 = 0.893 is for continental aerosol and 0.989 is for maritime aerosol. For the ASYM factor, the specified g0.55 values are 0.300, 0.500, 0.637/0.745, 0.950, where g0.55 = 0.637 for continental aerosol and 0.745 for maritime aerosol. Incorporating the aerosol scattering and absorption properties into the transmission and reflection tables allows more detailed treatment of aerosols and requires input information to specify aerosol properties. Aerosol information described in Liu et al. [2005], Liu et al. [2008] and Liu and Pinker [2008] is used in this study. The data give monthly mean, regionally resolved aerosol information on a 2° × 2.5° grid. Parameters include extinction coefficient, single scattering albedo and asymmetry factor for the five wavelength bands listed above.

[10] Liu et al. [2005]aerosol information is derived by combining data from independent sources that complement each other in their capabilities so that a global characterization of monthly mean clear-sky daytime aerosol optical depth can be achieved. The global scale estimates of aerosol optical depth at 0.55μm as based on the spatial and temporal variation patterns of model and satellite data and regulated by the Aerosol Robotic NETwork (AERONET) measurements [Holben et al., 1998] as described in Liu et al. [2005] have been supplemented with information on the large scale distribution of the single scattering albedo and the asymmetry parameter. Data from the Global Ozone Chemistry Aerosol Radiation and Transport (GOCART) model, Moderate Resolution Imaging Spectroradiometer (MODIS) retrievals, and AERONET measurements/retrievals are used. The single scattering albedo is generated by extending ω at 0.55 μm from GOCART to the entire shortwave (SW) spectrum using spectral dependence derived from available AERONET retrievals. The asymmetry parameter over the solar spectrum is derived from the MODIS Angström wavelength exponent, utilizing an empirical relationship based on AERONET almucantar observations. The normalized extinction coefficient is estimated from the MODIS Ångström wavelength exponent. Leading empirical orthogonal functions (EOFs) were used to represent the significant variation signals from model and satellite results; the EOFs were fitted to the ground observations to propagate the AERONET information to global scale. The methodology was implemented with a 2-year time record when collocated data from all three sources were available. Sensitivity tests forω and g were performed to assess effects on surface downward SW fluxes. For an assumed global average τ0.55 of 0.15, a perturbation of 0.05 and 0.1 in ω and g results in aerosol intensive optical properties with estimates based on the GOCART model about 2.0 and 1.5 Wm−2 flux changes, respectively. Details can be found in Liu et al. [2008]. In principle, once such relevant information becomes available from independent sources, it is possible to replace the currently used values.

2.2.2. New Land Use Information

[11] For the ISCCP DX version at resolution of 0.5°, a compatible land surface classification is needed. The International Geosphere-Biosphere Project (IGBP) as described inHansen et al. [2000] is at 1/6 degree resolution and is adopted for this study. The IGBP global classification data include 18 vegetation and land use types as listed in Table 1 (reduced for this study to 12). One 0.5° grid box contains nine IGBP pixels. The surface type for the grid box is determined by the most frequent surface type that occurred in the box. These surface types are also used for the UMD/SRB ISCCP D1 implementation of v3.3.3.

Table 1. The International Geosphere-Biosphere Project (IGBP) Surface Classification Map and the Corresponding 12 Surface Types Used for Clear Sky AMD Simulations
IGBP 18 TypesADM 12 TypesIGBP 18 TypesADM 12 Types
Evergreen Needle ForestNeedleleaf ForestGrasslandGrassland
Decid Needle ForestWetlands
Evergreen Broad ForestBroadleaf ForestTundra
Decid Broad ForestCropsCropland
Mixed Broad ForestMixed ForestCrop/Mosaic
Closed ShrubsClosed ShrubSnow/IceSnow/Ice
Open ShrubsOpen ShrubBarren/DesertBarren/Desert
Woody SavannasWoody SavannasWaterOcean

2.2.3. Treatment of Water and Ice Clouds

[12] In the UMD/SRB v3.3 model, all clouds are assumed to be water clouds. The new v3.3.3 version distinguishes between water and ice clouds. For each cloud type, the algorithm requires as input the visible cloud radiance and the corresponding cloud fractional area. Fluxes are calculated for each type separately. The all-sky flux is obtained as fractional area average of the scene-dependent fluxes.Fu's [1996] scheme is used to parameterize the ice cloud optics in the shortwave region. The scheme is based on in situ measured ice crystal size distribution, refractive indices and an improved geometric ray tracing algorithm [Warren, 1984]. Ice particle shapes are assumed to be randomly oriented hexagonal crystal. The six-band version of the Fu's parameterization is used in this study. Fu's single scattering properties for spectral interval 0.25 to 0.70μm are used directly for the UV and VIS bands. For the UMD/SRB NIR band, values are taken as the solar irradiance weighted average of the 5 bands in range from 0.70 to 4.00 μm from Table 3 of Fu [1996]. The ice cloud look-up tables are computed for a midlatitude summer profile with ice clouds located in between 10.0 to 10.2 km altitude. A mean effective diameter of 30μm is assumed for the ice clouds. The look-up tables are computed for 9 optical depths of 0.01, 0.05, 0.1, 0.25, 0.5, 1.0, 2.0, 5.0, and 10.0, and 2 aerosol conditions (maritime and continental).

[13] The ISCCP DX data do not provide directly fractional radiance for water and ice clouds. We generate this information in the process of re-gridding the DX data. The needed cloud type information is determined from cloud top temperature. Accordingly, for a cloudy pixel, if no top ice cloud temperature is reported or if both the VIS adjusted cloud top temperature and the VIS adjusted ice cloud top temperature are larger than 260 K (count 74 in the raw DX data), the cloud is water cloud; otherwise, it is an ice cloud (to be consistent with the ISCCP scheme of cloud typing). Cloud optical depth as provided by ISCCP is used as a first guess to determine the narrow-to-broad band conversion coefficients and to select the corresponding ADM model. If the first guess ice cloud optical depth is greater than 3.5, it is treated as water cloud.

[14] The impact of the distinction between the cloud phases on surface SW radiative fluxes as evaluated against ground observations from BSRN is illustrated in Figure 1 (top row). Distinguishing water and ice clouds in the new scheme reduces the bias with respect to BSRN observations by 6.2 W/m2 or 3.3% (1.8 W/m2 or 1% for RMS). Comparing these improvements to those achieved due to the new ADMs (improve bias by 1% but increase the STD by 0.3%), or to those due to the new gridding method (no effect on bias and improved RMS by 0.6%) it seems that incorporation of distinct cloud phases has the dominant effect on the results.

Figure 1.

(top row) Impact of separating water and ice clouds. Ice clouds are separated from water clouds (top left). If ice cloud optical depth is greater than 3.5, the cloud is reclassified as water cloud. All clouds are water clouds; different colors correspond to different seasons (top right). Evaluation is done for the BSRN stations within the GOES-East satellite coverage from 1/2000 to 12/2005 at monthly scale. (middle row) Evaluation of the impact of the new ADMs on surface SW fluxes. Test was done with all the seven satellites used in the ISCCP DX data. The evaluation is done with monthly mean data for the period 1/2000–12/2005 against 26 BSRN stations within latitudes from 65°S to 65°N. (bottom row) Different methods of gridding the ISCCP DX data to 0.5° equal-angles grid.

2.2.4. Narrow-to-Broadband Transformations and Anisotropic Corrections

[15] The broadband TOA albedo is the starting point for the retrievals of the surface radiative fluxes, namely, it is used to infer the transmission-reflection functions. The broadband albedo is obtained by first transforming the narrowband bidirectional reflectance into a broadband bidirectional reflectance using a linear transformation between the two. Subsequently, the bidirectional reflectance is converted into a broadband albedo using an anisotropic correction factor. The narrow-to-broadband conversion factors used in the early version of the algorithm are based on the visible channel of AVHRR on NOAA-7, which is the original absolute reference channel selected by ISCCP before 1997 [Brest et al., 1997]. Later, the normalizing reference for all the geostationary satellites used in the ISCCP data is changed to AVHRR on NOAA-9. The anisotropic corrections were improved by introducing synthesized Angular Distribution Models (ADMs) based on the newly developed CERES ADMs and on theoretically simulated ADMs [Niu and Pinker, 2011]. Included in the new conversion factors is dependency on satellite-solar geometry and cloud optical depth; and dependence on surface type. The original 4 surface types (water, vegetation, desert, snow/ice) were expanded to 12 types from the IGBP inventory for clear condition. The 12 new types are (1) Water; (2) Needleleaf Forest; (3) Broadleaf Forest; (4) Mixed Forest; (5) Woody Savannas; (6) Savannas; (7) Closed Shrub; (8) Open shrub; (9) Grasslands; (10) Croplands; (11) Bare Ground; (12) Snow and Ice (Table 1).

[16] CERES ADMs are available for 8 scene types as shown in Table 2. Due to satellite orbit limitations, under-sampling may occur for some view geometry and surface type combinations. To get ADMs for all 12 scene types, theoretical simulations are also introduced. The final ADMs are a combination of the empirical CERES ADMs and the theoretically simulated ones.Table 2shows the 8 clear-sky surface types available in the CERES AMDs. Also shown in the table are the 12 surface types used in the simulations. The CERES ADMs and the corresponding simulated ones are combined based on the number of samples [Niu and Pinker, 2011].

Table 2. Scene Classification Used in CERES ADMs and Their Mapping to the IGBP Surface Typesa
CERES Clear-Sky Surface TypesIGBP Surface Types
Bright DesertBare Ground
Dark Desert
Permanent SnowBright 
Fresh SnowSF > 99%, Bright 
SF > 99%, Dark 
75% < SF < 99% 
50% < SF < 75% 
25% < SF < 50%Snow and Ice
1% < SF < 25% 
SF < 1% 
Sea IceIF > 99%, Bright 
IF > 99%, Dark 
75% < IF < 99% 
50% < IF < 75% 
25% < IF < 50% 
1% < IF < 25% 
IF < 1% 
 Broadleaf Forest;
High-Mod Shrub/TreeMixed Forest;
 Woody Savannas;
 Closed Shrub;
Low-Mod Shrub/TreeOpen Shrub;

[17] The CERES ADMs use a detailed classification scheme for cloudy conditions, as shown in Table 3. The classification is based on Cloud Optical Depth (COD) as well as on cloud phase (water cloud, ice cloud), and also on the underlying surface type (ocean, tree/shrub, desert, and snow/ice). There are 88 classes in CERES ADMs for cloudy sky. The ISCCP DX data provide estimation of the cloud optical depth and information for determining the cloud phase; this allows the algorithm to utilize the detailed CERES ADMs (not possible with the ISCCP D1 data). The impact of using the new ADMs is illustrated in Figure 1 (middle row).

Table 3. Cloud Classification in CERES ADMsa
SceneCloud Optical Depth (COD)Cloud Phase
Ocean14 types: 0.01–1.0; 1.0–2.5; 2.5–5.0; 5.0–7.5; 7.5–10.0; 10.0–12.5; 12.5–15.0; 15.0–17.5; 17.5–20.0; 20.0–25.0; 25.0–30.0; 30.0–40.0; 40.0–50.0; COD > 50.0Water Cloud
Ice Cloud
Low-Mod Tree/Shrub6 types 0.01–2.5; 2.5–6.0; 6.0–10.0; 10.0–18.0; 18.0–40.0; COD > 40.0Water Cloud
Mod-High Tree/ShrubIce Cloud
Permanent SnowBright2 types: COD > 10.0 COD < 10.0 
Fresh SnowBrightNone
Sea IceBright 

2.2.5. Remapping ISCCP DX Data

[18] The ISCCP DX data are sampled at pixel level with spacing of about 30 km. We have re-mapped them to equal-area grids of 0.5° resolution using a weighted average method. The value for each equal-area grid point is calculated as the weighted average of all pixels that fall within 100 km around the grid center. The weighting function is of Gaussian shape with a full width at half maximum (FWHM) of 50 km, which is about 0.5° at the equator. This FWHM value is defined as the effective resolution of the re-mapped data. The ISCCP D1 data are obtained as an average of pixels that fall within a grid box and are provided at a resolution of 280 km equal-area grid. Experiments conducted with these two gridding methods using GOES-East satellite observations are illustrated for 1/2000 to 12/2005 inFigure 1 (bottom row). The weighted averaging is applied to all relevant input parameters such as cloud optical depth, cloud fraction, and satellite radiances. As seen from Figure 1 (bottom row) when results are evaluated against the Baseline Surface Radiation Network (BSRN) stations, the new gridding method (SRB/UMD_DX) (v3.3.3_sp) lowers the standard deviation from 14.5 W/m2 (7.8%) to 13.4 W/m2 (7.2%), though the bias increases somewhat from 1.42 W/m2 (0.8%) to 1.45 W/m2 (0.8%).

[19] The remapped data may contain missing cells, especially over sub-polar regions where both geostationary and polar-orbiting satellites have poor coverage. To fill missing data for a grid cell, the following was done. First, filling from observations for the same cell closest in time on the same day; if no daytime observations are available, filling from a cell closet to the missing cell in the same latitude band within 500 km radius, and of the same surface type on the same day; if no observations are available in the same latitude band for the current day, filling is done with monthly mean values.

2.3. Algorithm Implementation

2.3.1. Retrieval of Surface Albedo

[20] In principle, information on surface albedo and optical properties of aerosols and clouds can be supplied to the inference scheme from independent sources and the scheme can be run in a forward mode. In the case of the ISCCP data, experiments have indicated that the best results on surface radiative fluxes are obtained when partial aerosol and cloud information is derived from the same satellite observations used for retrieving the surface fluxes, under appropriate assumptions. The clear sky signal received by the satellite in addition to information on the surface also includes atmospheric effects. In order to separate these two influences there is a need to make certain assumptions. For example, for cases of clear sky, utilized can be “clear sky composites” which represent an average clear sky condition assuming that a monthly average aerosol optical depth will represent well the monthly conditions for each location. Global monthly mean aerosol optical depth at 0.55 μm; normalized extinction coefficients, SSA and ASYM averaged over five spectral intervals (0.2–0.4, 0.4–0.5, 0.5–0.6, 0.6–0.7 and 0.7–4.0 μm) are used as available from Liu et al. [2005], Liu et al. [2008], and Liu and Pinker [2008] in an initial step to retrieve the surface albedo from the clear sky composite. It gives monthly global scale distribution of aerosol single scattering albedo and asymmetry factor as well as aerosol optical depth. Once this assumption is made, an average surface albedo can be derived knowing all the other parameters that influence the satellite signal (i.e., water vapor content, ozone). This albedo is assumed to be constant for a period of one month and can be used in the retrieval of aerosol optical “index” from each instantaneous clear sky pixel. Similarly, using the same surface albedo one can use the instantaneous cloudy pixel radiances to retrieve a cloud optical depth. Having now estimates of these additional variable parameters, one can retrieve the corresponding surface fluxes from the LUTs. These steps will be described in more detail in what follows. In principle, once relevant information (e.g., surface albedo or instantaneous aerosol optical depth) becomes available from independent sources, it is possible to replace the currently used values.

2.3.2. Retrieval of Aerosol and/or Cloud Optical Depth

[21] Instantaneous clear sky conditions are used for aerosol optical “index” retrievals from match-ups with the LUT computations. The following quantities are assumed to be known when deriving the aerosol (or cloud) optical depth: sun-satellite geometry (satellite zenith angle, solar zenith angle and relative azimuth angle between satellite and the sun), columnar amounts of ozone and water vapor, broadband TOA albedos for clear sky, surface albedo, and single-scattering albedo and asymmetry factor.

[22] To derive the aerosol or cloud optical depth one needs to search through the optical depth space to find a value that gives the appropriate transmissivity and reflectivity values to match the observed TOA albedo to the computed one. This is possible because the assumption is made that all the other quantities are known except the optical depth. The matching of optical depth is determined by a linear interpolation scheme. The optical depth derived this way is an “effective optical depth,” since the value lumps together not only aerosol or cloud optical depth variation, but also unknown variations from other atmospheric or surface components that could have impact on the TOA reflectivity and are not properly accounted for in the auxiliary input data.

2.3.3. Computing Radiative Fluxes

[23] Once the inverse problem of determining the surface albedo, aerosol and cloud optical depth is solved and the atmospheric state is fully known, the corresponding transmission-reflection functions needed for computing the radiative flux at surface and TOA are determined. Given these values the following radiative fluxes are computed for both clear and cloudy conditions: surface downwelling and upwelling flux, TOA upwelling flux, both total and diffuse. All fluxes are instantaneous spectral values given in 5 broadband spectral intervals.

[24] The derived instantaneous fluxes are first scaled by the 3-hourly average of the cosine solar zenith angle to get the mean fluxes for the 3-hourly time interval, and then integrated numerically for the daylight hours to get a 24-h daily average. Because of the finite number of observations available per day, the total daily flux obtained from numerically integrating the instantaneous fluxes is potentially inaccurate. Therefore, the daily total fluxes are adjusted by the ratio of the TOA incoming flux as obtained by an analytical integration to that computed from the numerical integration.

3. Model Implementation With ISCCP DX Data

3.1. Background

[25] The ISCCP DX data are provided satellite by satellite. There are 5 geostationary satellites (4 before 1998) and 2 polar orbiting satellites that provide information to the ISCCP DX database. Ideally, five geostationary satellites should be available simultaneously for a complete global coverage up to about 55°N and 55°S; polar orbiting satellites are used above these latitudes. Before July 1998 only 3–4 such satellites were available. Meteosat-5 has been moved over the Indian Ocean only in 1998 to remove a geostationary gap over that region. We compute the radiative fluxes for each satellite domain independently. The quality of the computed radiative fluxes is not equal within the geostationary domain; the outer pixels with a larger viewing zenith angle cover larger footprints and tend to impact the estimated radiative fluxes. There is a need for a merging scheme to combine information from the multisources to produce complete and homogeneous information on radiative fluxes at global scale. For geostationary satellites, the merging inference scheme uses the monthly mean cosine of the satellite view angle as a weighting factor to obtain a weighted average from various satellites. Fluxes from polar orbiting satellite are assigned a fixed weighting factor; values depend on the latitude of the grid point. If pole-ward of 55°, fluxes from all polar satellite are given the same weighting factor; this makes the merged fluxes in polar region a simple average of all available polar satellite fluxes. At lower latitudes, fluxes are a weighted average as derived from geostationary satellites. In the gap regions between geostationary satellites, missing values are filled with observations from polar orbiting satellite. In this study, we are interested in radiative fluxes which are hemispherical integral quantities and as such it is possible to average fluxes that come from two different satellites (e.g., in areas of overlap between satellites). The scheme we use is a combination of the scheme used when producing the ISCCP D1 version of satellite observations [Zhang et al., 2004], and one developed for this study. Specifically, in the ISCCP D1 scheme, for every location, a hierarchy of preferred satellite observation is specified. At any time only one satellite is selected for each location. In our scheme, for every location, no observation is discarded but merged into the final product by taking a weighted average. The weighting factor is the cosine of the satellite viewing angle for geostationary satellites and a constant value for polar orbiting satellites. The preferred satellite is given the full value of the viewing cosine angle, while the value for the non-preferred one is scaled down by a factor of 0.5, as follows:

display math

where μprf and Fprf are the viewing angle cosine and the radiative flux for the preferred satellite, respectively; ui and Fi are the values for the ith non-preferred satellite available at the same location, respectively. An example of products from geostationary satellites only, from polar orbiting satellite only, and from merged data from both geostationary and polar orbiting are shown inFigure 2.

Figure 2.

(top) Monthly mean SW fluxes for April 2004 as derived with UMD/SRB v3.3.3 using the ISCCP DX data from five geostationary satellites and gridded at 0.5° resolution. (middle) Monthly mean SW fluxes for April 2004 as derived with v3.3.3 using the ISCCP DX data from the AVHRR on polar orbiting satellites and gridded at 0.5° resolution. (bottom) As above, merged from geostationary and polar orbiting satellites.

3.2. Inputs to the Algorithm

[26] The UMD/SRB model uses as input satellite observation of TOA reflectivity combined with various ancillary data to derive shortwave radiative fluxes at both boundaries of the atmosphere. In Table 4all required input quantities as well as their primary function and source are listed. Some of the needed parameters are part of the ISCCP DX database. Others, such as ozone and water vapor, need to be imported from independent sources. The satellites used to generate the ISCCP data sets have between two to five spectral channels that are relevant for inferring shortwave fluxes and for detecting clouds. They have different spectral intervals and as such, sense different amounts of energy. Moreover, the filter functions of each sensor differ in their response curve. For the ISCCP DX data, the homogeneity is achieved by normalization of observations from geostationary satellites by a polar orbiter. Originally, NOAA-7 was used and more recently, replaced by NOAA-9.

Table 4. Input Parameters, Their Primary Functions and Sources
Input ParameterFunctionSource
Clear-sky TOA radianceTo derive aerosol optical depthISCCP DX
Cloudy-sky TOA radiance (water and ice)To derive cloud optical depthISCCP DX
Clear-sky composite radianceTo derive surface albedoISCCP DX
Number of clear pixelsTo compute scene fractional areaISCCP DX
Number of cloudy pixels (water and ice)To compute scene fraction areaISCCP DX
Amount of water vaporInput to the radiative model (LUT)ISCCP D1
Amount of OzoneInput to the radiative model (LUT)ISCCP D1
Surface pressureInput to the radiative model (LUT)ISCCP D1
Snow coverTo weight snow albedoISCCP DX
Solar zenith angleInput to the radiative model (LUT) and to determine anisotropic correction factorsISCCP DX
Satellite zenith angleInput to the radiative model (LUT) and to determine anisotropic correction factorsISCCP DX
Relative azimuth angleInput to the radiative model (LUT) and to determine anisotropic correction factorsISCCP DX
Latitude and longitudeTo identify pixel characteristicsISCCP DX
Aerosol climatologyInput to the radiative model (LUT) and to compute surface albedoLiu et al. [2005]
Anisotropic correctionBidirectional reflection correctionCERES ADM
N/B transformationNarrow to broadband transformation 
Surface albedo modelFirst guess to retrieval surface albedo.B., W. and M.a
Surface typeInput to model and surface albedo.IGBP/UMD
Climatology of wv and O3Used when satellite measurement missingKneizys et al. [1980]

[27] The observations made by each instrument at their original resolution are re-sampled at 30 km each 3 h. The sampling is such that at each time interval a different area is sampled to achieve best possible time and space representation. Before using these observations to drive the inference scheme, there is a need to re-grid the data first to a resolution that sufficiently represents each grid cell as was explained insection 2.2.5.

4. Results and Evaluation

4.1. Evaluation Approach

[28] Obtaining accurate estimates of the radiation budget at global scale has been a major objective of numerous WCRP and national programs [Intergovernmental Panel on Climate Change, 2007]. Early information was derived from very limited observations of surface radiative fluxes [e.g., Budyko, 1956] or from simple models [e. g., London, 1952]. Currently, estimates are obtained from numerical models and from satellite observations. Comprehensive summaries on what is known are provided by [Kiehl and Trenberth, 1997; Trenberth et al., 2011]. In order to obtain a meaningful comparison between different approaches, the length of the period for which data are available from all sources restricts the time span of the comparison. As more information becomes available, it is possible to extend the comparisons for longer time periods. The primary objectives of the evaluation effort of this study are: provide information on the performance of the improved methodology (UMD/SRB DX) to derive surface SW radiation; compare these results to independent satellites information; and to results from numerical models. Additional SW related parameters such as SW net at the TOA, surface albedo, TOA albedo, SW absorbed in the atmosphere, will be also compared to each other when feasible (direct observations for these additional parameters are not available). Since specific regions may be of special interest, the results are presented independently at global scale, for land and oceans, and stratified by latitude. Results are provided in Tables 5a5d. The numbers in parenthesis give the % of the range to the average value from all the models that were compared. A summary for the range of the various results as observed from satellites is presented in Table 6. The range of model results for each parameter is presented in Table 7.

Table 5a. Global Averages of SW Radiation Budget Quantities Derived From Satellite Observations and Re-analysis Products, Stratified by Latitudinal Beltsa
  • a

    Albedo values are computed as the ratio of globally averaged mean up and down fluxes. The time period is from October 2000 to October 2005. The % of the range to the average as computed from all available data for a particular parameter is presented in Table 6.

  • b

    Values are from Trenberth et al. [2009] (TFK09) Table 2. Averaging period is from March 2000 to May 2004.

  • c

    Values are from TFK09 Table 1, which gives the global mean values averaged over the ERBE period (Feb 1985 to Apr 1989).

UMD/SRB DX v3.3.3190.8164.114.0236.230.372.1
UMD/SRB D1 v3.3.3195.2168.413.7239.629.371.2
UMD/SRB D1 v3.3195.5172.911.6   
GEWEX v3.0188.3166.611.5   
ISCCP-FD (This paper)188.3165.612.1236.430.970.8
ISCCP-FD (00–04,TFK09)b188.5165.712.1236.530.870.8
ERBE FT08 (85–89, TFK09)c   234.431.3 
ISCCP-FD (85–89,TFK09)c188.9164.912.7235.931.071
UMD/SRB DX v3.3.3201.3179.910.6255.429.175.5
UMD/SRB D1 v3.3.3205.6184.310.4259.228.074.9
UMD/SRB D1 v3.3210.4191.09.2   
GEWEX v3.0201.9183.69.1   
UMD/SRB DX v3.3.3235.0214.08.9302.525.688.5
UMD/SRB D1 v3.3.3239.1217.79.0305.924.788.2
UMD/SRB D1 v3.3249.2228.48.3   
GEWEX v3.0239.7219.88.3   
High Lat (S) 60°S-90°S
UMD/SRB DX v3.3.3106.337.564.777.861.540.3
UMD/SRB D1 v3.3.3111.438.465.575.562.637.1
UMD/SRB D1 v3.3105.649.952.7   
GEWEX v3.0103.650.151.6   
High Lat (N) 60°N-90°N
UMD/SRB DX v3.3.395.750.147.693.452.843.3
UMD/SRB D1 v3.3.3100.555.
UMD/SRB D1 v3.392.760.834.4   
GEWEX v3.096.464.033.6   
Table 5b. Averages Over Land of SW Radiation Budget Quantities Derived From Satellite Observations and Re-analysis Products, Stratified by Latitudinal Beltsa
  • a

    Albedo values are computed as the ratio of globally averaged mean up and down fluxes. The time period is from October 2000 to October 2005. The % of the range to the average as computed from all available data for a particular parameter is presented in Table 6.

  • b

    Values are from Trenberth et al. [2009] (TFK09) Table 2. Averaging period is from March 2000 to May 2004.

  • c

    Values are from TFK09 Table 1, which gives the global mean values averaged over the ERBE period (Feb 1985 to Apr 1989). The atmospheric absorption values in TFK09 are adjusted values which do not equal the difference between TOA net and surface net shown in the same table.

UMD/SRB DX v3.3.3187.5135.627.7208.035.772.4
UMD/SRB D1 v3.3.3194.0141.527.1212.534.171.0
UMD/SRB D1 v3.3192.6144.824.8   
GEWEX v3.0181.8139.223.4   
ISCCP-FD (This paper)185.7143.922.5213.334.469.4
ISCCP-FD (00–04,TFK09)b188.8148.721.2219.333.770.6
ERBE FT08 (85–89, TFK09)c   212.135.8 
ISCCP-FD (85–89,TFK09)c190.1147.222.6217.134.469.9
UMD/SRB DX v3.3.3201.4158.821.2238.033.379.2
UMD/SRB D1 v3.3.3207.8164.920.6243.031.678.1
UMD/SRB D1 v3.3212.8169.120.5   
GEWEX v3.0199.9162.019.0   
UMD/SRB DX v3.3.3222.9185.416.8280.131.094.7
UMD/SRB D1 v3.3.3228.6188.717.5283.230.294.5
UMD/SRB D1 v3.3244.5195.520.0   
GEWEX v3.0225.6184.418.3   
High Lat (S) 60°S-90°S
UMD/SRB DX v3.3.3135.716.388.051.872.135.3
UMD/SRB D1 v3.3.3141.116.988.048.673.731.7
UMD/SRB D1 v3.3133.832.975.4   
GEWEX v3.0126.630.376.1   
High Lat (N) 60°N-90°N
UMD/SRB DX v3.3.3103.753.048.998.052.245.0
UMD/SRB D1 v3.3.3110.561.244.6104.749.343.5
UMD/SRB D1 v3.3102.069.232.2   
GEWEX v3.0103.171.730.5   
Table 5c. Averages Over Oceans of SW Radiation Budget Quantities Derived From Satellite Observations and Re-analysis Products, Stratified by Latitudinal Beltsa
  • a

    Albedo values are computed as the ratio of globally averaged mean up and down fluxes. The time period is from October 2000 to October 2005. The % of the range to the average as computed from all available data for a particular parameter is presented in Table 6.

  • b

    Values are from Trenberth et al. [2009] (TFK09) Table 2. Averaging period is from March 2000 to May 2004.

  • c

    Values are from TFK09 Table 1, which gives the global mean values averaged over the ERBE period (Feb 1985 to Apr 1989).The atmospheric absorption values in TFK09 are adjusted values which do not equal the difference between TOA net and surface net shown in the same table.

UMD/SRB DX v3.3.3191.9175.78.4247.530.371.8
UMD/SRB D1 v3.3.3195.4178.28.8249.429.371.2
UMD/SRB D1 v3.3196.6183.56.7   
GEWEX v3.0190.7177.07.2   
ISCCP-FD (This paper)189.3173.88.2245.130.971.3
ISCCP-FD (00–04,TFK09)b188.3172.08.7242.829.870.8
ERBE FT08 (85–89, TFK09)c   242.429.8 
ISCCP-FD (85–89,TFK09)c188.5171.59.0243.029.771.4
UMD/SRB DX v3.3.3201.3187.66.8261.827.674.2
UMD/SRB D1 v3.3.3204.8190.76.9264.526.973.8
UMD/SRB D1 v3.3209.6198.35.4   
GEWEX v3.0202.6190.75.9   
UMD/SRB DX v3.3.3238.8223.16.6309.523.986.4
UMD/SRB D1 v3.3.3241.9225.66.7312.123.286.5
UMD/SRB D1 v3.3250.5237.45.2   
GEWEX v3.0243.6229.55.8   
High Lat (S) 60°S-90°S
UMD/SRB DX v3.3.386.152.638.996.454.943.8
UMD/SRB D1 v3.3.392.851.944.192.456.640.5
UMD/SRB D1 v3.387.960.631.1   
GEWEX v3.089.162.529.9   
High Lat (N) 60°N-90°N
UMD/SRB DX v3.3.387.847.246.288.853.441.6
UMD/SRB D1 v3.3.391.549.645.890.052.640.4
UMD/SRB D1 v3.384.453.237.0   
GEWEX v3.090.357.136.8   
Table 5d. SW Radiation Budget Quantities From Re-analysis Producta
  • a

    Values are taken from TFK09 and TFM11. Averaging periods are shown in parentheses. Notations: NRA-R1: NCEP reanalysis (known as NCEP/NCAR or R1 versions); NRA-R2: NCEP reanalysis (known as NCEP/DOE or R2 versions); CFSR: NCEP/CFSR. Values averaged from 2002 to 2008 (02–08) are fromTFM11. Values averaged from (00–04) and from (85–89) are from TFK09. ERA-40 (1990s) results are also fromTFM11. The % of the range to the average as computed from all available data for a particular parameter is presented in Table 7.

MERRA (02–08)19316912.42412973
NRA-R1 (02–08)20516022.02253464
NRA-R1 (00–04)205.6160.422.0224.834.264.4
NRA-R1 (85–89)207.1161.921.8226.333.864.4
NRA-R2 (02–08)18816114.42373176
ERA-40 (1990s)17815612.42383182
ERA-40 (85–89)178.9155.812.9236.531.080.7
ERA-I (02–08)18816412.82442980
JRA (02–08)19717212.72472875
JRA (00–04)195.4169.813.1244.527.974.7
JRA (85–89)194.5168.913.2243.928.175.0
CFSR (02–08)19817213.12482776
C20R (02–08)19216613.52432977
NRA (00–04)225.4155.131.2214.235.2)59.1
NRA (85–89)224.1155.230.8214.335.2)59.1
ERA-40 (85–89)177.2134.324.2220.333.3)86.0
JRA (00–04)207.4155.824.9227.730.6)71.9
JRA (85–89)206.4154.925.0227.130.8)72.2
NRA (00–04)198.5162.318.2228.633.966.3
NRA (85–89)201.0164.318.3230.633.366.3
ERA-40 (85–89)179.4163.58.9242.330.278.8
JRA (00–04)191.1174.98.5250.527.075.6
JRA (85–89)190.1173.98.5249.927.176.0
Table 6. Range of Values (in %) for Selected SW Components at the Surface, TOA and Within the Atmosphere, Averaged Globally and at Various Latitudinal Beltsa
  • a

    Estimated range of values is based on Tables 5a5c. The numbers give the % of the range to the average as computed from all available data for a particular parameter.

Table 7. Range of Values (in %) for Selected SW Components at the Surface, TOA and Within the Atmosphere, Averaged Globallya
  • a

    Estimated range of values is based on Table 5d. The numbers give the % of the range to the average as computed from all available data for a particular parameter.


[29] The UMD/SRB DX model version (v3.3.3) has been implemented at global scale at 0.5° resolution for the period of available data, namely, July 1983–June 2009. Used are merged data from geostationary and polar orbiting satellites. Satellite models selected for comparison are: ISCCP-FD, ISCCP/GEWEX and CERES; Numerical Models include NCEP Reanalysis and ERA-Interim. Previously compiled comprehensive comparisons [Kiehl and Trenberth, 1997; Trenberth et al., 2011] are also included, as well as results from UMD/SRB v3.3 as implemented with both ISCCP D1 and ISCCP DX (results from this version as implemented with ISCCP D1 have been submitted to the Radiative Flux Assessment comparison and are archived at NASA LaRC ( The Baseline Surface Radiation Network (BSRN) [Ohmura et al., 1998] provides the best available observations over land while the Pilot Research Moored Array in the Tropical Atlantic (PIRATA) and Tropical Atmosphere-Ocean (TAO) networks provide the best observations over oceans [McPhaden et al., 1998].

4.2. Evaluation Against the BSRN Network (Land)

[30] Over land we use ground observations from the BSRN network [Ohmura et al., 1998]. The BSRN network provides long-term stable high quality ground observation data from diverse climatological regions covering a latitude range from 80°N to 90°S. ( Since 1992 the BSRN stations provide data for monitoring variability of surface radiative fluxes, calibration of satellite-based measurements of radiative fluxes and evaluation of numerical model outputs. As of February, 2011 there are more than 50 stations contributing to the BSRN that provide measurements of shortwave global, diffuse and direct components of radiative fluxes. The BSRN network is expanding and adding new stations especially in remote land regions and over oceans [Wild, 2011].

[31] Monthly mean results from January 2000 to December 2005 as obtained from UMD/SRB v3.3 (as submitted to RFA) driven with ISCCP D1, v3.3.3 driven with ISCCP D1 and ISCCP DX both at 0.5° and 2.5° resolutions are compared against 26 BSRN stations (Figure 3) (polar stations are not included in this comparison). The UMD/SRB DX 2.5° results are obtained by up-scaling fluxes from the 0.5° data. The value for each 2.5° degree grid is the average of 5 × 5 values at 0.5°. When compared to v3.3, the v3.3.3 version is in better agreement with ground observations than results from ISCCP D1 as input (UMD/SRB D1 v3.3.3 and UMD/SRB D1 v3.3). It is believed that the improvement comes from the model updates such as the newly available CERES ADM, improved aerosol climatology and distinguishing between water and ice clouds, as described insection 2. Additional improvement can be seen in the UMD/SRB DX version. This may be due to both model updates and model input including data resolution (DX is gridded to 0.5°) and the new gridding and merging methods as described in section 2.2 and 3.1. After up-scaling from 0.5° to 2.5° (UMD/SRB DX 0.5° and UMD/SRB DX 2.5°), the uncertainty of the satellite results increased a little. This is in line with our expectation because of the scale disagreement between point observations and satellite footprint scale.

Figure 3.

Evaluation of UMD/SRB v3.3 and v3.3.3 against BSRN data from 26 globally distributed stations (polar stations are excluded) for 1/2000–12/2005. (top left) UMD/SRB DX v3.3.3; (top right) UMD/SRB DX v3.3.3 upgraded to 2.5°; (bottom left) UMD/SRB D1 v3.3.3; (bottom right) UMD/SRB D1 v3.3.

[32] In Figure 4comparison is made between UMD/SRB v3.3.3 at 0.5° resolution with similar results from ISCCP-FD (2.5°) and GEWEX/SRB (1.0°). The results are for the period 2000–2005. For the latitude zone between 65°S to 65°N, the bias of the satellite retrieved monthly mean downwelling SW flux ranges from −0.7 (0.4%) to 3.1 (1.7%) (W/m2), and the standard deviation is between 12.9 (6.9%) to 17.3 (9.3%) (W/m2). Products derived from the ISCCP observations have larger deviations for higher values of radiative fluxes than for the lower ones; they also tend to underestimate the downwelling fluxes while CERES overestimates them. In Figure 5 time series of bias and standard deviation (STD) are shown for several satellite products. As evident, the time series are quite consistent and do not indicate any spurious fluctuations. As seen in the time series of bias (Figure 5, bottom), starting from early 2002, the ISCCP based retrievals have a consistent and small bias which differs for the CERES bias when compared to the same BSRN ground observations, and keeping the retrieval algorithms unchanged. These features may be related to the uncertainty or changes in satellite calibration. The calibration of the polar orbiting satellites used in the ISCCP products underwent change in October, 2001 when NOAA-14 was replaced by NOAA-16. The previous linear calibration of the visible channel [Brest et al., 1997; Desormeaux et al., 1993] was changed to a bi-linear version ( As yet, the larger biases in the CERES data have not been fully understood. Evaluation of UMD/SRB DX results at daily time scale was also performed as illustrated in Figure 6 (left) that also shows the frequency distribution of the differences between satellite estimates and ground observations (Figure 6, right).

Figure 4.

Comparison between UMD/SRB v3.3.3 at 0.5° resolution with similar results from ISCCP-FD, GEWEX/SRB and CERES (CERES-SRBAVG-Terra-GEO-MOD_Ed02d from the RFA website). The results are for the period 2000–2005 and no outliers are removed.

Figure 5.

Time series of bias and standard deviation (STD) as evaluated against BSRN observations for 1/2000 to 12/2005. Values are averaged over 26 BSRN stations located within latitude from 65°S to 65°N. CERES data for 3/2000 to 10/2005 only were used.

Figure 6.

(left) Evaluation of UMD/SRB DX results at daily time scale against BSRN observations from 1/2003 to 12/2005. (right) Frequency distribution of the differences between satellite retrievals and ground observations. Evaluation was done for 28 stations between 60°S to 60°N.

4.3. Evaluation Against the PIRATA and TAO Networks (Ocean)

[33] In situ measurements from the Pilot Research Moored Array in the Tropical Atlantic (PIRATA) moorings in the tropical Atlantic [Bourlès et al., 2008], and the Tropical Atmosphere Ocean (TAO)/Triangle Trans-Ocean Buoy Network (TRITON) moorings in the tropical Pacific Ocean [McPhaden et al., 1998] are used for evaluation of the surface SW satellite estimates. Downwelling SW is measured with the Eppley Laboratory pyranometers that have nominal resolution of 0.4 W/m2 and relative accuracy of ±2% in the 0–1600 W/m2 interval in laboratory conditions [Cronin and McPhaden, 1997]. Measurement errors over oceans may come from sensor tilt associated with wind, ocean currents and installation. Aerosol build-up on the dome of the radiometers may be also a source of error. There is no general agreement on the accuracy of buoy-based radiometric measurements and the buoy measurements of surface radiative fluxes do not adhere to the rigorous standards of BSRN protocols. Evaluation over oceans (Figure 7) shows that the satellite retrievals overestimate the downwelling SW at all three regions (PIRATA, TAO and Nauru), which is not the case over land. Evaluation of SW fluxes as derived from the Moderate Resolution Imaging Spectroradiometer (MODIS) observations [King et al., 1992, 2003] against the same oceanic arrays are described in Pinker et al. [2009]; they show good agreement between the estimates and the observations. This would indicate possible dependence of the SW estimates on the quality of the input data.

Figure 7.

Comparison between UMD/SRB v3.3.3 at 0.5° resolution and similar results from ISCCP-FD and GEWEX/SRB against two buoy networks (PIRATA over the Atlantic, TAO over the Pacific) and the Nauru island station at monthly time scale from 1/2001 to 12/2005; no outliers are removed.

4.4. Comparison With Independent Estimates of Various SW Parameters

[34] A comparison of results from the various implementations of the UMD/SRB model with independent and widely used satellite estimates and selected estimates from numerical models has been conducted and is summarized in Tables 5a5d. Results are presented with same version as implemented with the ISCCP D1 data (UMD/SRB D1 v3.3.3) at 2.5° resolution; an earlier version of the model (submitted to the RFA archive) as implemented at 2.5° resolution (UMD/SRB D1 v3.3); UMD/MODIS results as implemented at 1° resolution and using methodology described in Wang and Pinker [2009]; CERES results as provided at the RFAarchive also at 2.5° resolution; GEWEX v3.0 and ISCCP-FD at 1° and 2.5° resolution, respectively, as available from the Atmospheric Science Data Center (ASDC) at NASA Langley Research Center; ISCCP-FD, MERRA [Rienecker et al., 2011], NRA [Kistler et al., 2001], ERA-I [Berrisford et al., 2009], JRA (02–08) [Onogi et al., 2007], CFSR [Saha et al., 2010], and C20R [Compo et al., 2011] are from Figure 10 of Trenberth et al. [2011](TFM11). Results labeled TFK09, KT97, ERBE FT08, ERA-40 and JRA (00–04) are fromKiehl and Trenberth [1997] and Trenberth et al. [2009]. Results are presented for surface downwelling and net SW, surface albedo, TOA net SW, TOA albedo, and absorbed SW in the atmospheric column. The data are stratified by latitudinal belts that include the entire globe (60°S-90°N), the regions between 60°S-60°N, 20°S-20°N, high latitudes 60°S-90°S, and 60°N-90°N.

[35] As evident from Table 5a for global averages, and as summarized in Table 6, the smallest differences between the various estimates are in the NETTOA values up to 60°S or 60°N and range between 1 and 2%. For latitudes above 60° differences increase to 8% for above 60°N and 13% for above 60°S. The differences are also relatively small for the ALBTOA between 4 and 8% till 60°N and 60°S, increasing at higher latitudes. The largest differences between satellite estimates are for surface albedo ranging between 18 and 20% up to 60°N and 60°S, increasing to 28% above 60°S and 35% above 60°N. A concise summary of the range of variability is presented in Table 6.

[36] In Table 5b similar results to those of Table 5a are presented for land only. The range of values for the same categories as presented in Table 5a increased over land alone. It should be noted that the number of model results for land alone is smaller than for the entire globe. Of notable difference are the values in ALBSFC in the tropical belt and (most likely, due to large differences in the estimates of ALBSFC over deserts between models) and the NETSFC at high southern latitudes (most likely, due to difficulties to discriminate between clouds and snow/ice at these latitudes). In Table 5c similar results to those of Table 5a are presented for oceans only. Here too the largest differences are in ALBSFC possibly, for several reasons. Some of the models assign ocean surface albedos according to known climatologies that are not time dependent. Methodologies that derive the surface albedo over oceans face issues of possibly cloud contaminated pixels, issues of glint, and difficulties to detect clouds at low solar zenith angles. The largest difference is in the region between 60°S to 90°S. As have been already shown in Nussbaumer and Pinker [2012], in this region the differences in cloud amounts between ISCCP and MODIS are very large, possibly due to the fact that the numerous spectral channels of the MODIS instrument can improve cloud detection. Table 5d is similar to Table 5a for model results alone based on the work of Kiehl and Trenberth [1997], Trenberth et al. [2009], and Trenberth et al. [2011]. A summary of the range % of the mean for model results is presented in Table 7. As evident, the spread between model results is much higher than between satellite estimates.

[37] A concise comparison of several surface SW radiative flux components from selected sources used in Tables 5a5d is illustrated in Figure 8. All values are globally averaged over land and oceans for various periods as stated in the figure. Values for MERRA [Rienecker et al., 2011], NRA [Kistler et al., 2001], ERA-I [Berrisford et al., 2009], JRA [Onogi et al., 2007] (02–08), CFSR [Saha et al., 2010], and C20R [Compo et al., 2011] are from Figure 10 of TFM11. Values for TFK09, KT97, ERA-40 and JRA (00–04) are from Table 2 ofTFK09. Albedo values are computed following the TFK09 method. CERES data are at a resolution of 2.5° and were obtained from the Radiative Flux Assessment website (

Figure 8.

Comparison of surface SW radiative flux components from various sources as presented in Tables 5a5d. All values are globally averaged over land and oceans for periods as indicated on the figure: (00–05): from October 2000 to October 2005; (02–08): from 2002 to 2008; (00–04): from March 2000 to May 2004; (85–89): from February 1985 to April 1989. Values for MERRA, NRA, ERA-I, JRA (02–08), CFSR, C20R are from Figure 10 ofTFM11. Values for TFK09, KT97, ERA-40 and JRA (00–04) are from Table 2 ofTFK09. UMD/MODIS data are averaged for August 2002 to October 2005. Albedo values are computed following the TFK09 method. CERES data are at a resolution of 2.5° and were obtained from the Radiative Flux Assessment website (

[38] It should be noted that the quality of the ground observations has also an effect on the results of the evaluation. The operational limits of the ground observations have been discussed in Shi and Long [2002]. It is stated that the best estimate of accuracy for the direct component is 6.3 + 3.3, for the diffuse component it is 4.0 + 1.4 and for the upwelling it is 11.1 + 2.8 in (W/m2), respectively.

5. Summary

[39] The use of satellite observations in climate research has been steadily expanding. It has been demonstrated that the quality of the satellite based estimates of parameters such as radiative fluxes exceeds those available from numerical models when compared to ground observations. As yet, not enough is known on the uncertainties of the satellite products, in particular, on their regional dependencies. Moreover, as new type of information becomes available, the inference schemes need to be updated to accommodate the new information. For many climate issues, information is needed at global scale which poses a challenge since satellite observations at global scale and at high temporal resolution are not readily available. This study describes the evolution of an inference scheme for deriving SW flux components to enable the incorporation of more recent auxiliary information. Primarily, progress has been made on the availability of information on aerosols at global scale, cloud and surface properties, and on ADM models at the top of the atmosphere. Ongoing comparison activities among products have contributed to a better understanding of the impacts of computational approaches on the results and thus enabled to reduce errors from sources that have not been understood before. Examples include the various options for numerical integration in time and space as well as the impact of gap filling for missing values and gridding of the satellite observations. Briefly, the major changes of the model described here as compared to its previous versions include: new narrow-to-broad band conversion; new filling procedure for missing input data, specifically, the use of a Gaussian shape weighting function to grid the input data and to merge the output fluxes from various satellites based on their view zenith angle; clouds are separated into water and ice clouds (previous version assumes all clouds are water clouds); the latest most updated version (v3.3.3) takes into account new information on surface properties, and new angular distribution models (ADMs) based on CERES [Niu and Pinker, 2011]. The methodology is implemented with available satellite observations that as yet, were not optimally utilized.

[40] Comparison of various satellite products against land and ocean ground observations has been done at a monthly time scales and where possible, also on daily time scale. It was shown that over land sites there is a good agreement between the various satellite estimates. It was also shown that over oceanic sites all the satellite estimates based on the ISCCP observations had a positive bias that as yet is not fully understood.

[41] A comprehensive comparison was conducted between available satellite products and selected numerical re-analysis results as available from independent studies. The comparison was done at global scale covering the entire globe and stratified by latitudes to include 90°S-90°N, 60°S-60°N, 20°S-20°N, High Lat (S) 60°S-90°S, and High Lat (N) 60°N-90°N (Table 5a). Such comparison was repeated over land and oceans alone (Tables 5b and 5c). Comparison from various models is included in Table 5d. To obtain a measure for differences, a range of values for selected SW components at the surface, TOA and within the atmosphere based on Tables 5a5d was compiled as presented in Tables 6 and 7. The numbers give the % of the range to the average as computed from all available data for a particular parameter. As evident, the largest variability in most of the parameters is at high latitudes. A very large scatter is in the surface albedo values at these regions while most parameters agree much better in the equatorial belt. The net SW fluxes at the TOA as well as the absorbed fraction of the incident radiation tend to show smaller variability than other parameters, possibly, due to error cancellation. Furthermore, differences between various products may come from differences in instrumentation, calibration, model physics, model implementation, differences in utilizing input quantities and gridding methods. The new data described in this manuscript are at highest available resolution for global scale and at longest time scale possible; as such, they are suitable to support hydrologic and environmental research. Evaluations against best available ground measurements over land and oceans demonstrated improvement when compared to an earlier version of the model with better agreement over land than over oceans. While it was demonstrated that there is room for improvement in estimating SW fluxes from historical observations, the study also highlights the various difficulties for obtaining global view from numerous satellites that differ in their instrument specifications and mode of observations. Further investigations are needed to learn how to reduce uncertainties at global scale.


[42] This work was supported under NASA grant NNX08AN40A from the Science Mission Directorate-Division of Earth Science and benefited from support under NSF grant ATM0631685. Thanks are due to the NASA GES DISC Giovanni for the MODIS data, to the various MODIS teams that produced data used in this study, to the Baseline Surface Radiation Network for observations used in the evaluation, and to theRadiative Flux Assessment archive ( for several data sets used. We acknowledge the TAO Project Office of NOAA/PMEL for providing data from the Pilot Research Moored Array in the Tropical Atlantic (PIRATA) and from the Tropical Atmosphere Ocean/Triangle Trans-Ocean Buoy Network (TAO/TRITON) in the Tropical Pacific Ocean that were used in this study as well as the DOE/ARM Program for providing the data from the Nauru Island station. The very useful comments of Kevin Trenberth, two anonymous reviewers, and the Editor are greatly appreciated.