A database of spectral surface reflectivity in the range 335–772 nm derived from 5.5 years of GOME observations



[1] A global database of Lambert-equivalent reflectivity (LER) of the Earth's surface has been constructed by analyzing observations of the reflectivity at the top of the atmosphere made by the Global Ozone Monitoring Experiment (GOME). Since its launch on board the ERS-2 satellite in April 1995, the GOME instrument has been measuring spectra of the Earth between 237 and 794 nm, with a spectral resolution between 0.2 and 0.4 nm and a spatial resolution between 40 × 80 and 40 × 320 km2. The LER database covers eleven 1-nm-wide wavelength bins centered at 335, 380, 416, 440, 463, 494.5, 555, 610, 670, 758, and 772 nm, which were selected for various retrieval applications. The database has a spatial resolution of 1° × 1°, is made for each month of the year, and pertains to the period June 1995–December 2000. Typical spectra of various surface types are presented. Attention is paid to instrument degradation and residual cloud contamination. We have found satisfactory agreement between our database at 380 nm and the Total Ozone Mapping Spectrometer (TOMS) LER database at 340–380 nm, with negligible average difference and a standard deviation of 0.013. The database presented here can be used to improve retrievals of trace gases, clouds and aerosols from GOME, Scanning Imaging Absorption Spectrometer or Atmospheric Cartography (SCIAMACHY), Ozone Monitoring Instrument (OMI), and GOME-2.

1. Introduction

[2] The Global Ozone Monitoring Experiment (GOME) is a 4-channel grating spectrometer, operating in the wavelength range 237–794 nm with a spectral resolution of 0.2–0.4 nm [Burrows et al., 1999]. In April 1995, GOME was launched on board the European ERS-2 satellite into a polar orbit at 790 km altitude with an equator crossing-time of 1030 (local time) for the descending node. By employing a scanning mirror, GOME observes the Earth in near-nadir direction (±30° from the subsatellite track), corresponding to a 960 km swath width and global coverage every three days. During each scan, three spectra are measured, pertaining to areas of 40 × 320 km2 (along × across-track direction). An additional spectrum is recorded during the back-scan, with a ground pixel size of 40 × 960 km2. For about 10% of the time, the swath width is reduced to 240 km, and all pixel sizes are four times smaller. Once per day the Sun is observed over a diffuser plate for radiometric calibration. Spectral calibration is performed by using an internal Pt–Cr–Ne lamp. From the GOME spectra, trace-gas column densities have been derived of O3, NO2, BrO, and several other gases, as well as profiles of O3, and properties of clouds and aerosols.

[3] Knowledge of spectral surface reflectance is important for trace gas retrieval in the UV-visible range, particularly if trace gas densities peak in the troposphere. For example, the sensitivity of the top of atmosphere radiance to boundary layer NO2 is critically dependent on surface reflectance [Martin et al., 2002]. Under polluted conditions boundary layer NO2 column densities can be as large as those of the stratosphere, and hence accurate knowledge of surface reflectance is needed for retrieval of NO2 column densities. Also, the sensitivity of the top of atmosphere radiance to boundary layer ozone varies up to more than a factor of 4 depending on surface reflectance [Hudson et al., 1995; P. Valks, private communication, 2002]. Knowledge of the spectral surface reflectance is needed also to derive properties of aerosols and clouds. Veefkind et al. [2000] estimate that an error in the surface reflectance of 0.01 gives an aerosol optical thickness retrieval error of 0.1 in the UV. Koelemeijer et al. [2001] show that a surface reflectance error of 0.02 gives rise to errors in cloud top pressures retrieved from oxygen A band measurements ranging from zero to up to more than 150 hPa, depending strongly on cloud fractional coverage and surface reflectance itself.

[4] Current global albedo data sets are based on broadband sensors such as the AVHRR [e.g., Csiszar and Gutman, 1999, and references therein]. These databases are important for energy balance studies and GCM modeling, but their use for retrieval of atmospheric properties is hampered because of their limited spectral information. In the future, multiyear global data sets with better spectral resolution may become available from new sensors such as MODIS and MISR [Vermote et al., 1997; Liang et al., 1999]. However, these latter sensors do not cover all wavelengths considered in this paper. The aim of this paper is to provide a database of the spectral reflectivity of the Earth's surface which can be used to improve retrieval of column densities of trace gases and properties of aerosols and clouds from measurements made by GOME and its successors, such as SCIAMACHY, launched on board Envisat on 1 March 2002, Ozone Monitoring Instrument (OMI), to be launched on board EOS-AURA in 2004, and the GOME-2 instruments on board the METOP series, the first of which is scheduled for launch in 2005. A similar database, but restricted to the range 340–380 nm, was derived by Herman and Celarier [1997] (hereinafter referred to as HC97) from Nimbus-7 Total Ozone Mapping Spectrometer (TOMS) data. Herman and Celarier have shown that the LER contains valuable information in itself, such as information on phytoplankton concentrations in oceans and features in coastal waters, vegetation and clouds [Herman et al., 2001a, 2001b]. The database described in this paper may be regarded as an extension of the database of Herman and Celarier to visible and near-infrared wavelengths.

[5] The paper is structured as follows. In section 2, the contents of the database are described. The LER retrieval method is described in section 3. Examples of geographical, temporal and spectral variations of the derived LER values are presented in section 4. In section 5 error sources are discussed. We compare our results at UV wavelengths with TOMS derived LER values in section 6. We end with concluding remarks in section 7.

2. Contents of the Database

[6] Generally, the reflection of light by a homogeneous surface can be described by its bidirectional reflectance distribution function (BRDF). In retrievals of column densities of trace gases and properties of aerosols and clouds from spectrometer measurements, however, nearly always a Lambertian surface is assumed. The values in our LER database can be used for such retrievals. Following Bhartia et al. [1993] and Herman and Celarier [1997], we define the Lambert equivalent reflectivity (LER) as the value of the Lambertian spectral surface albedo for which the modeled and measured reflectivity at the top of the atmosphere are equal, assuming a Rayleigh scattering atmosphere above a Lambertian surface in the radiative transfer model. We recognize that non-Lambertian effects are present in the derived LER value. However, by employing this LER value with a Lambertian surface in trace-gas retrievals from instruments with an observation geometry similar to GOME, non-Lambertian effects are, to first order, implicitly accounted for. The same reasoning holds for the contribution by background aerosols, for which no corrections are made (see also HC97).

[7] Our database consists of minimum LER values which occurred in the period 27 June 1995–31 December 2000. The minimum LER values are stored as a function of wavelength λ, month of the year, and geographic location. Table 1 lists the wavelengths that were selected in the database and their possible application for retrieval. The wavelengths were arithmetically averaged over 1-nm wide bins. The wavelengths were selected to be outside strong narrow absorption bands of atmospheric gasses, and outside strong solar Fraunhofer lines. Hence, we avoid as much as possible the necessity to correct for the presence of strong absorbers in the Earth's atmosphere, and for the filling in of the Fraunhofer lines and strong atmospheric lines caused by rotational Raman scattering. The only strong absorber that has been corrected for is ozone, whose column density is known from the GOME level 2 data product [Deutsches Zentrum für Luft- und Raumfahrt (DLR), 1994].

Table 1. Selected Wavelengths in the LER Database
λ, nmRetrieval Application
335.0O3 (Huggins band)
463.0O2–O2 (477 nm band)
670.0cloud detection
758.0O2 (A band)
772.0O2 (A band)

3. LER Retrieval Method

[8] In the retrieval we make use of a radiative transfer model to calculate a look-up table of top of atmosphere reflectivities, which is described in section 3.1. Section 3.2 describes the retrieval approach, and section 3.3 describes several corrections that were applied to the database.

3.1. Look-Up Table Generation

[9] We employ a polarized doubling-adding radiative transfer model [De Haan et al., 1987; Stammes, 2001] to simulate the reflectivity at the top of the atmosphere, which is defined by

equation image

Here, I is the radiance reflected by the Earth (in Wm−2nm−1sr−1), E0 is the incident solar irradiance at the top of atmosphere perpendicular to the solar beam (in Wm−2nm−1), and μ0 is the cosine of the solar zenith angle.

[10] In the model, a multilayered Rayleigh-scattering atmosphere is assumed, bounded below by a Lambertian surface. Furthermore absorption by ozone is taken into account. We used the midlatitude summer pressure, temperature and ozone profiles from Anderson et al. [1986]. We multiplied the ozone profile with a constant factor to take into account variations of the ozone column density. Also, the surface pressure was varied to take into account effects of the Earth's topography.

[11] To describe the dependence of R on the surface albedo As, we make use of the formula [Chandrasekhar, 1960, section 72; Van de Hulst, 1980, section 4.5.4]

equation image

where the first term describes the path radiance and the second term describes the radiance reflected by the surface. The function t describes the total atmospheric transmission and s* is the spherical albedo of the atmosphere for illumination from below. The dependence of the path radiance on the azimuth difference between Sun and satellite, ϕ − ϕ0, can be written as

equation image

A lookup table of am(μ, μ0), t(μ) and s* was calculated as a function of surface pressure Ps, ozone column density Ω and wavelength λ. These calculations were performed at the center of each 1-nm wide wavelength bin. By interpolation in this lookup table, the reflectivity can be found as a function of μ, μ0, ϕ − ϕ0, Ps, As, Ω, for each value of λ.

3.2. Retrieval Approach

[12] Reflectivities at the top of the atmosphere as measured by GOME are obtained from the raw signals using version 2.0 of the extraction software GDP01_EX as provided by the Deutsches Zentrum für Luft und Raumfahrt (DLR) with all calibration options applied. The raw signals were provided and processed by DLR with the same version of the level 0–1 processor (reprocessing stage 02). The calibration options include adjustments for leakage current, stray light, focal plane assembly noise, detector pixel-to-pixel variability, correction for polarization sensitivity of the instrument, correction for asymmetric azimuth dependence of the diffuser, and absolute radiometric calibration. Reflectivities are arithmetically averaged over a 1-nm wide wavelength window centered at the wavelengths listed in Table 1.

[13] The Lambert-equivalent reflectivity for each of the 11 wavelengths, denoted by RL(λ), is found for each GOME pixel as the value of the surface albedo needed to match the measured reflectivity at the top of the atmosphere, using the look-up table described in section 3.1 and equations (2) and (3). The values of μ, μ0, ϕ − ϕ0 and Ω are taken from the GOME level 2 data product. The surface pressure is deduced from the ETOPO-5 topography database [Haxby et al., 1983] assuming the midlatitude summer pressure profile. To take into account variations of solar and viewing angles over the GOME ground pixel, we made simulations for two Sun-satellite geometries for each GOME pixel, based on a two-point Gaussian quadrature. We analyzed all approximately 50 million GOME pixels acquired between 27 June 1995 and 31 December 2000, but we discarded observations with solar zenith angles larger than 80°. Also, we discarded observations made in nadir-static mode, for which the scanning mirror is fixed in nadir position leading to a swath of only a few kilometers wide.

[14] All RL(λ) spectra were then sorted into grid-cells of 1° × 1° and per month of the year, yielding ∼60 LER spectra per grid-cell per month. For each grid cell and month, the minimum value of RL is searched at 670 nm. We used 670 nm because at that wavelength the contrast between cloud-free and cloudy pixels is large for most surface types. This value, as well as the corresponding values at the other 10 wavelengths, are then stored in an intermediate results database. These monthly minimum LER (MLER) values are denoted by RL,min(λ). Our approach ensures that the MLER values in the database are derived from the same GOME spectrum for each grid cell.

3.3. Correction Steps

[15] The intermediate results were postprocessed by applying three correction steps. The first correction step is applied only to ocean grid cells, the second and third step are applied to both land and ocean grid cells. The first correction step consists of reducing effects of clouds, and makes use of the small spatial variability of the oceanic reflectivity at 772 nm. Land surfaces exhibit a rather large spatial variability for all wavelengths considered in our paper, hence this correction has been applied to ocean grid cells only. To this end, we separated ocean grid cells into three groups, based on the MLER value at 772 nm, RL,min(772). The first group with RL,min(772) < 0.05 is assumed not to be affected much by residual cloud contamination, and is referred to as “clear”. The second group with 0.05 < RL,min(772) < 0.5 has a high probability of residual cloud contamination, and is referred to as “cloudy”. The third group with RL,min(772) > 0.5 is assumed to be (partly) covered by ice shelves, because of the persistent high reflectivity. We verified that the results are not sensitive to this particular choice of 0.5. Cloudy ocean grid cells have been replaced by a distance weighted average of clear ocean grid cells in a 5° × 5° neighborhood around the cloudy grid cell, if present. Grid cells with RL,min(772) > 0.5 were not corrected, to retain the boundary between open ocean and ocean covered with ice-shelves. Using this procedure, 14.5% of all grid cells were corrected for cloud contamination.

[16] The second correction step consists of filling in grid cells with missing data. We distinguish three regions with missing data. The first region is a narrow region over the Himalayas, caused by a data storage constraint of the ERS-2 data recorder; this region is filled in by MLER values from adjacent grid cells at the same latitude. The second region is the polar region which is observed only during part of the year. Grid cells with missing data during a certain month are filled in by data of the month closest in time where data are available. The third region is the polar region which is never observed (i.e., very close to the poles); these grid-cells are filled in by copying data from lower latitude polar regions with similar surface characteristics (the Beaufort sea centered at 82N, 140W for the North pole and an area of the Antarctic centered at 80S, 50E for the South pole). In total, 19.5% of the grid cells were filled in. Since the large majority of these grid cells is located in the polar regions, the surface area of the Earth pertaining to these grid cells is only 7.1%.

[17] The third correction step consists of identifying and removing outliers by considering the month-to-month variation of RL,min(772). Most land surfaces have high RL,min(772) values. This property has been exploited to identify and remove a few outliers to low values. For land grid cells, if RL,min(772) in a particular month is lower by 0.1 than the RL,min(772) values of the previous and next month, it is replaced by the minimum RL,min(772) value of the previous and next month. Ocean shows low RL,min(772) values for all months, and this property is used to detect and remove outliers to high values. For ocean grid cells, if RL,min(772) in a particular month is higher by 0.05 than the RL,min(772) values of the previous and next month, it is replaced by the maximum RL,min(772) value of the previous and next month. Both corrections concern less than 1% of all grid cells. The outliers above ocean may be due to residual cloud contamination. For the remaining 65% of the grid cells no corrections were applied.

[18] In addition to the database with monthly minimum values, a database was made with annual minimum values, containing the minimum MLER value that occurred in each of the twelve months for each grid cell and wavelength. In this database of annual minimum LER values it is no longer the case that the values are derived from the same GOME spectrum, as minima at different wavelength generally occur in different months. Both the monthly minimum and the annual minimum databases are accompanied by a database of flags, which keeps track of corrections that have been applied to each grid cell.

4. Example of Results

[19] Global maps of MLER values at 440 nm and 758 nm derived from GOME are shown in Figures 1 and 2. At 440 nm, much large-scale structure can be observed over the ocean, which can be related to variations in concentrations of phytoplankton, yellow substance, and detritus. For example, high MLER values over ocean correlate with low concentrations of phytoplankton, as derived by SeaWiFS [Gregg, 2002]. A persistent band in the equatorial Pacific shows low MLER values almost year-round, caused by upwelling of nutrient rich water at the equator leading to large concentrations of phytoplankton near the surface. Similar features can be observed in the UV (see also section 6); a detailed description of such features in the UV is given in HC97 and [Herman et al., 2001a]. At 758 nm, the ocean is much darker and more homogeneous than at short wavelengths. This is because at 758 nm water itself is the dominant absorber, whereas below 500 nm absorption by pure water is small, and the dominant absorbers are substances like phytoplankton and yellow substance which may vary considerably in space and time [Schwartz, 2001]. The MLER values of land surfaces at 440 nm are high for snow covered areas and deserts and are low for densely vegetated areas, because of absorption by chlorophyll and carotenes. At 758 nm, vegetated area is highly reflective because absorption by leaf-pigments is small and scattering occurs mainly at cell-walls through the change of refractive index between cells and air cavities between cells [Bowker et al., 1985]. MLER values at 758 nm show much seasonal variation, which correlates with the amount and health of vegetation.

Figure 1.

Seasonal variation of MLER values derived from GOME data at 440 nm. The values in the legend have been multiplied by 100.

Figure 2.

Same as Figure 1 but at 758 nm.

[20] For a few surface types spatial averages and standard deviations of monthly minimum LER spectra were calculated, as shown in Figure 3. Surface type was deduced from the Matthews land usage database [Matthews, 1983]. Spectra are shown for January, April, July, and October. Table 2 lists the surface types and latitude/longitude ranges of the spectra considered in Figure 3. We note that the values at 335 nm are probably too high by ∼0.02 or more, see section 5.1. Spectra which were filled in because of missing data are not considered in this figure.

Figure 3.

Average monthly minimum LER spectra and their standard deviation for various surface types specified in Table 2. The values at 335 nm may be affected by degradation, see section 5.1.

Table 2. Selected Surface Types for Figure 3
Matthews TypeDescriptionLatitude RangeLongitude Range
  1. a

    The symbol – indicates that no limitation was set to the latitude or longitude ranges. n/a stands for not applicable.

n/aAtlantic Ocean0N–60N20W–50W
1Tropical evergreen rain forests
9–11Deciduous forests
31Snow/ice (Greenland)60N–70N30W–50W

[21] The average spectra of the Atlantic Ocean show a rather monotonic decease of MLER with wavelength, for wavelengths shorter than 600 nm. For wavelengths between 600 and 800 nm, the MLER values are almost constant. Standard deviations are slightly larger for λ ≤ 440 nm compared with longer wavelengths. Tropical evergreen rain forests show a strong increase in reflectivity between 670 and 758 nm, known as the red-edge. In the visible, a local maximum occurs near 555 nm (green), because of smaller absorption by leaf pigments in the green compared to the blue and red. Little seasonal variation is observed. The average spectrum of deciduous forests in July resembles that of tropical rain-forests. The largest difference between RL,min(670) and RL,min(758) occurs in July, when vegetation is optimally developed. In the other months shown, MLER values are lower in the near-infrared compared to July. The spectra of grasslands show a more gradual increase with wavelength than that of tropical evergreen rain-forests. The reason is that spectral mixing between the soil and grass spectra occurs, caused by the relatively high transmission of the canopy of most grasslands compared to that of forests. For deciduous forests and grasslands, standard deviations are higher in January compared with the other months, probably due to snow coverage. Indeed, the standard deviation has a local maximum in the UV, where MLER values are low for vegetation but high for snow. The spectra of deserts show strongly increasing MLER values with wavelength with very little seasonal variation. The spectra of Greenland show decreasing MLER values with wavelength, and are highest in spring, when fresh-fallen snow is still present. In summer and fall, snow is older and has lower reflectivity. In January, data are missing because of the too large solar zenith angle. The spectra shown here correspond well with typical spectra given in Bowker et al. [1985]. It is beyond the scope of this paper to discuss the spectral variations for all surface types in detail.

5. Discussion of Error Sources

[22] Errors in the derived LER values are mainly caused by errors in radiometric calibration of GOME and errors caused by residual cloud contamination, which are discussed in this section.

5.1. Radiometric Calibration

5.1.1. Sensitivity to Radiometric Calibration

[23] It is clear from equation (2) that the accuracy of the derived LER depends on the absolute calibration of the GOME reflectivity measurements R. The sensitivity of the retrieved LER to R is given by the derivative dRL/dR and can easily be obtained from equation (2). Figure 4 shows dRL/dR for different wavelengths, for θ0 = 60°. Triangles indicate points on the curves for which the surface albedo equals 0, 0.25, 0.50, 0.75, and 1. In this figure, the sensitivity is largest at 335 nm for RL = 0 and amounts to 2.57. Then, an error of ±0.01 in R gives rise to an error of ±0.0257 in the retrieved LER. For θ0 = 30° and 75°, the corresponding sensitivities are 2.07 and 3.39, respectively. Hence, radiometric calibration errors are amplified in the retrieved MLER, especially for short wavelengths and large solar zenith angles. Calibration issues are therefore addressed in sections 5.1.2 and 5.1.3.

Figure 4.

Derivative of the retrieved LER value to the reflectivity at the TOA, for 335, 380, 440, and 758 nm (see equation (2)). Nadir view, solar zenith angle 60°, clear-sky midlatitude summer atmosphere, surface elevation 0 km. Triangle marks indicate points on the curves for which the surface albedo is 0, 0.25, 0.50, 0.75, and 1.

5.1.2. Degradation

[24] Solar irradiance measurements are performed by GOME every day. Since our retrieval is based on reflectivities, any degradation in optical components which affects the solar and Earth observation mode similarly cancels in the reflectivity. Degradation of the GOME solar irradiances in the UV and blue is monitored by comparing with SOLSTICE data [Peeters et al., 1996]. A degradation correction option can be applied in the DLR extraction software, but this has no effect on the reflectivities because degradation is assumed to affect the solar irradiance and Earth radiance measurements similarly. However, recent studies on GOME reflectivity degradation in the UV [Tanzi et al., 2001; Van der A, 2001] have indicated that it is likely that the solar irradiance measurements generally degrade more rapidly than the Earth's radiance measurements, leading to an artificial increase of reflectivity with time. The observed degradation is wavelength-dependent and varies with time. Significant degradation of the GOME reflectivity at 335 nm started after mid-1999, with only little degradation before this date. Our study is based on the operationally available GOME reflectivity product and hence does not account for differences in degradation rate of observations of the Sun and Earth.

[25] To assess how degradation of reflectivity affects the derived LER values, we analyzed a time series of LER values of the Lybian desert (22N, 28.5E), a very stable site which has been used for calibrations of several satellite instruments in the past [e.g., Smith, 1997]. In this time series, we discarded nadir-static and backscan observations. First we consider the ratio of RL(416)/RL(610) as a function of time, shown in Figure 5. The majority of points is symmetrically distributed around the median value at 0.306 and lie between 0.27 and 0.34. A number of points which exceed 0.34 are probably partly cloud contaminated and were therefore not considered to study degradation. Figure 6 shows derived LER values for the Lybian desert as a function of time. For wavelengths longer than 335 nm, degradation can be well described by a linear fit RL(t) = a + bt, where t is the number of days since 1 January 1995. The coefficients a and b, their errors ϵa and ϵb, and the absolute and relative degradation, Δabs and Δrel, are listed in Table 3. The absolute degradation is calculated as the difference between the average RL value of the last 100 days and the first 100 days of the 5.5 year period considered. Δrel is the absolute deviation divided by the average RL value of the first 100 days and is expressed in %/yr. Degradation is significant at the 95% confidence level for all wavelengths except 416 nm.

Figure 5.

Time series of RL(416)/RL(610) for the Lybian desert (22N, 28.5E). Observations with a ratio exceeding 0.34 (indicated by horizontal line) are probably cloud contaminated and therefore discarded.

Figure 6.

Time series of LER values of the Lybian desert (22N, 28.5E).

Table 3. Linear Fit of LER Degradation
λ, nmabϵaϵbΔabsΔrel, %/yr
  1. a

    Coefficients for the linear fit RL(t) = a + bt and their errors (ϵa and ϵb), the absolute degradation Δabs in 5.5 years, relative degradation Δrel (in %/yr) as derived from observations of the Lybian desert.


[26] At 335 nm rapid increase in degradation rate starts after mid-1999, consistent with findings of Van der A [2001]. At 380 nm degradation starts near the end of 2000. Degradation at wavelengths longer than 400 nm, corresponding to GOME channels 3 and 4, is much smaller than in the UV and has a sign opposite to that of channel 2 (313–405 nm), yielding slightly decreasing LER values with time. We conclude that absolute errors in the derived LER values due to degradation can be up to 0.01–0.02 for all wavelengths considered except 335 nm. At 335 nm this error can be up to 0.07. Since the Lybian desert is among the brightest snow-free sites in the world, this is a maximum estimate and absolute errors are likely to be smaller for most other areas.

5.1.3. Absolute Calibration

[27] In the beginning of the ERS-2 mission, reflectivities measured simultaneously by GOME and ATSR-2 at 555 nm and 660 nm have been compared and agreed to within 4.0% and 2.2%, respectively (GOME being higher than ATSR-2) [Koelemeijer et al., 1998]. Since only relatively minor degradation of LER values has been observed over the GOME lifetime for wavelengths outside the UV, we assume that absolute calibration is still sufficiently accurate for visible and near-infrared wavelengths.

[28] In a few case studies GOME UV measurements were compared with radiative transfer calculations, using measured ozone profiles from lidars or sondes as input and assuming a spectrally independent surface albedo derived at 380 nm [Van der A, 2001; Stammes, 2001].

[29] These studies indicate that TOA reflectivity values measured by GOME are typically up to 2–5% too high at 335 nm as compared to 380 nm, for measurements before mid-1999. Therefore, besides the error due to degradation, we may expect that our results at 335 nm are too high by about 0.02–0.04, depending on surface reflectivity and solar zenith angle.

5.2. Residual Cloud Contamination

[30] To get a qualitative impression of residual cloud contamination in the MLER data, color composite images are shown in Figure 7 using MLER values at 440 nm (blue), 555 nm (green) and 670 nm (red). Permanently snow covered areas and seasonally snow covered areas, e.g., high latitude landmasses in the Northern Hemisphere in winter, show up in white and can be clearly distinguished. Seasonally persistent clouds are most frequent above equatorial landmasses, associated with the intertropical convergence zone (ITCZ). For such regions, often covered by rain forests, trace gas and aerosol retrievals will be seriously hampered by clouds, particularly if most of the component of interest is below the cloud. Clouds in the ITCZ are more bound to certain longitudinal positions above land than above ocean, and hence the chance to obtain cloud free measurements in the ITCZ region for a certain location is larger over ocean than over land. Hence, land areas are affected more strongly by residual cloud contamination than ocean areas. This is reinforced by the fact that it is easier to detect and correct for residual clouds over ocean than over land. Generally land surfaces are much brighter at 758 nm than at 440 nm, so the relative contribution of residual clouds is smaller at 758 nm than at 440 nm. In the annual minimum data, residual clouds are almost entirely removed (not shown). Grid cells which are likely to be contaminated by snow or residual clouds are flagged. This flag is set for ocean grid cells with RL,min(772) > 0.05 and for land grid cells with RL,min(380) > 0.2. If the MLER database is used for retrieval of atmospheric quantities, one may eventually choose to replace cloud contaminated MLER values by that of the month closest in time for which the residual cloud contamination flag is not set.

Figure 7.

Color composite constructed with MLER values at 440 nm (blue), 555 nm (green) and 670 nm (red).

6. Comparison With TOMS Data

[31] We compared our results with monthly and annual minimum LER values derived from Nimbus-7 TOMS data by Herman and Celarier [1997].The TOMS climatology is based on TOMS measurements at 340 and 380 nm. Generally not much difference was found between the 340 and 380 nm values, and therefore the TOMS database was assumed to be valid for the wavelength region 340–380 nm. Recent reanalysis of the TOMS data however suggest larger differences between 340 and 380 nm LER values, and the TOMS data in HC97 are more representative for 380 nm (E. Celarier, private communication, 2001). Given this, and given the problems with absolute calibration and degradation of GOME at 335 nm, we compare the TOMS data with our GOME results at 380 nm. We note that the TOMS database is constructed from observations in the period 1 November 1978–6 May 1993 (14.5 years), whereas the GOME data concern the period 27 June 1995–31 December 2000 (5.5 years).

[32] Figure 8 shows histograms of annual minimum LER values from GOME and TOMS and their difference, for latitudes between 60S and 60N. The GOME and TOMS histograms are similar in shape, but GOME data are slightly higher than TOMS by 0.005. However, this bias of GOME with respect to TOMS is small and is less than the intrinsic quantization step of TOMS LER values, which is 0.01.

Figure 8.

Histograms of annual minimum LER values from GOME (380 nm) and TOMS (340–380 nm) (left) and their difference (right), for latitudes between 60S and 60N.

[33] Global maps of annual minimum LER values at 380 nm of GOME and TOMS and their difference are presented in Figure 9. Generally, the maps agree very well. Both in GOME and TOMS data, vegetated areas are significantly darker than ocean, whereas deserts can be brighter than the ocean. Minimum LER values of ocean at 380 nm show similar features as at 440 nm. Land surfaces are generally darker in the UV as compared to 440 nm, particularly for sparsely vegetated areas and deserts. The difference map GOME-TOMS shows random noise, except for the ocean near the Antarctic, and for the Arabian Peninsula. The low annual minimum LER values in the TOMS climatology of the ocean south of 60S are probably an artifact caused by high solar zenith angles in the months May–July. In fact, this region exhibits high pixel-to-pixel variability caused by rough ocean and high cloud coverage for most of the year [HC97]. The low values for some desert areas may be due to uplifting of large amounts of dust, which lower the reflectivity [Herman et al., 1997; Torres et al., 1998]. The annual minimum difference GOME-TOMS averaged over the Southern Hemisphere (0–60S) is 0.010. This is larger than that of the Northern Hemisphere (0–60N), for which it is 0.001. We have not found a proper explanation for this phenomenon.

Figure 9.

Annual minimum LER values at 380 nm derived from GOME (top) compared with TOMS (middle). The lowest map shows the difference GOME-TOMS. TOMS data are from the Herman and Celarier [1997] climatology and are based on measurements in the period 1978–1993. The GOME data are derived from measurements in the period 1995–2000. The values in the legend have been multiplied by 100.

[34] For each month, we calculated the average difference and standard deviation of the difference between MLER values derived from GOME and TOMS, for latitudes between 60S and 60N. Generally, a satisfactory agreement was found between GOME and TOMS derived MLER values, with an average difference between 0.007 and 0.023 and a standard deviation between 0.017 and 0.024. The annual minimum data showed average difference of 0.005 and a standard deviation of 0.013. Effects of residual cloud and aerosol contamination and of differences in snow/ice coverage are smaller in the annual minimum data than in the monthly data, leading to better agreement between GOME and TOMS for the annual minimum data. Probably, these effects are the main source of random differences between GOME and TOMS. Also, part of the random difference may be caused by differences in data analysis, e.g., the use of the LER at 670 nm to select the minimum LER at other wavelengths in the present work. Systematic differences can be caused by radiometric calibration errors in the GOME or TOMS data, differences in the solar and viewing angles for which the GOME and TOMS data were acquired, and changes in surface properties that have occurred between the periods considered in the TOMS and GOME data sets. The GOME monthly and annual minimum LER data are probably affected more by residual cloud contamination than the TOMS data, because the TOMS data set is based on more observations.

[35] Figure 10 shows maps of the difference in MLER values derived by GOME at 335 nm and 380 nm. Even though interpretation of GOME data at 335 nm is hampered by calibration errors, these results suggest that appreciable wavelength dependence may occur depending on location. At some locations MLER values at 335 nm are distinctly lower than at 380 nm, indicating the presence of UV absorbing aerosols. Indeed, the geographical and temporal distribution of these areas correlates qualitatively well with absorbing aerosol index distributions derived from TOMS data [Herman et al., 1997]. For example, large negative values of RL,min(335) − RL,min(380) in tropical Africa and Brazil in the months June–September are probably caused by biomass burning (only July is shown here). Likewise, the occurrence of dust storms in the deserts of North Africa, the Arabian Peninsula, and Asia can be observed. Snow covered surfaces generally show higher MLER values at 335 nm compared to 380 nm. Such dependence on surface type and location is not expected if the differences are only caused by calibration errors.

Figure 10.

Difference RL,min(335) − RL,min(380) from GOME for January and July. The values in the legend have been multiplied by 100.

7. Concluding Remarks

[36] GOME measurements obtained in the period 1995–2000 have been analyzed to generate a global database of monthly minimum LER (MLER) values of the surface at eleven 1-nm wide wavelength bins in the range 335–772 nm. The database has a 1° × 1° resolution and is made for each month of the year.

[37] The spectral, geographical and temporal dependence of the derived MLER values was presented. The results agreed qualitatively well with other data sets. We have found a generally satisfactory agreement between our database at 380 nm with the TOMS minimum LER database at 340–380 nm. Notable exceptions are the Arabian Peninsula and the Ocean south of 60S. For the annual minimum data, the average difference between GOME and TOMS LER values was 0.005 and the standard deviation was 0.013. The monthly data have an average difference between 0.007 and 0.023, and a standard deviation of typically 0.02. The higher standard deviation values of the monthly data compared with the annual minimum data are probably caused by larger effects of snow/cloud contamination in the monthly data. Although the GOME data at 335 nm suffer from calibration errors, our results suggest that appreciable wavelength dependence of MLER values may occur in the UV, which can be related to the presence of UV absorbing aerosols (mineral dust and aerosols emitted by biomass burning).

[38] We have shown that the derived LER values are sensitive to radiometric calibration. Degradation of GOME reflectivities at the TOA affects the derived LER values, giving artificial increases or decreases of 0.5%–2% per year for all wavelengths except 335 nm (see Table 3). For the Lybian desert site, which is among the brightest snow-free sites available, this corresponds to absolute errors in the derived LER values of 0.01–0.02 in the 5.5 years considered for all wavelengths except 335 nm. At 335 nm, the errors are larger, and these LER values may be generally too high by 0.02 or more.

[39] The spectral MLER database can be a valuable aid in the retrieval of trace gases, clouds and aerosols from GOME, SCIAMACHY, OMI, and GOME-2. The wavelengths were specifically selected to be near absorption bands of trace gases, and cover wavelengths not covered by other currently available surface reflectance data sets. Presently, this database is used by various groups for retrieval of tropospheric trace gas column densities, clouds and aerosols from GOME, as well as cloud top pressures from the O2–O2 absorption band at 477 nm to be observed by OMI, and cloud top pressures derived from O2A band radiances measured by MERIS [e.g., Martin et al., 2002; Fischer et al., 2000].

[40] Our database can be improved when degradation correction of reflectivities is implemented in the GOME extraction software. This is particularly important at 335 nm. We note that these calibration problems at 335 nm do not directly affect ozone column density retrieval, since these are derived using the Differential Optical Absorption Spectroscopy method, for which absolute calibration is not important. Another improvement would be the use of a pseudo-spherical rather than a plane-parallel radiative transfer code.

[41] The database can be extended in the future by analyzing SCIAMACHY data to extend the wavelength coverage to 2380 nm, and by analyzing OMI and GOME-2 data to improve spatial resolution and reduce effects of residual cloud contamination. The database is available upon request from the authors.


[42] We kindly acknowledge the European Space Agency (ESA) and the Deutsches Zentrum für Luft- und Raumfahrt (DLR) for providing the GOME data. We would like to thank E. Celarier (NASA-GSFC) for his comments on the comparison between GOME and TOMS derived minimum LER values, as well as two anonymous reviewers for their valuable remarks. Part of this work was performed while R. Koelemeijer worked at KNMI, with financial support by NIVR (grant 11.901 KN).