A method was investigated to estimate broadband surface shortwave albedo from the narrowband reflectances obtained by the Advanced Very High Resolution Radiometers (AVHRRs) on board the polar orbiting satellites. Field experiments were conducted to measure simultaneously multispectral narrowband reflectances and broadband albedo over various vegetation and soil surfaces. These data were combined to examine the behavior of narrowband-to-broadband (NTB) conversion factors for different surfaces. Many previous studies have used constant NTB conversion factors for the AVHRR data. The results from this investigation indicate that the optimal NTB conversion factors for AVHRR channels 1 and 2 have a strong dependence on the amount of green vegetation within the field of view. Two conversion factors, f1 and f2, were established to quantify, respectively, 1) the relationship between the reflectance in the narrow red wave band and the total reflectance within the whole visible subregion (0.3–0.685 μm) and 2) the relationship between the reflectance in the narrow near-infrared wave band and the total reflectance within the whole near-infrared subregion (0.685–2.8 μm). Values of f1 and f2, calculated from field data, correlated well with the normalized difference vegetation index (NDVI), largely because of the unique characteristics of light absorption and scattering within the red and near-infrared wave bands by green leaves. The f1–NDVI and f2–NDVI relationships developed from this study were used to infer empirical coefficients needed to estimate surface albedo from AVHRR data. The surface albedo values calculated by the new method agreed with ground-based measurements within a root-mean-square error of 0.02, which is better than other methods that use constant empirical coefficients. Testing with albedo measurements made by unmanned aerospace vehicles during a field campaign also indicates that the new method provides an improved estimate of surface albedo.
The spatial representativeness of ground-based point measurements of surface albedo becomes an important issue when such measurements are compared with satellite observations in an attempt to infer atmospheric radiation absorption and the radiation budget because surface heterogeneity can lead to point measurements misrepresenting the surface area covered by the satellite field of view. Reliable methods are needed to estimate surface albedo over large areas. Such methods are also needed for parameterization of the surface radiation budget for atmospheric models. The narrowband reflectance data from high-resolution remote sensing measurements by satellites, such as Landsat and the National Oceanic and Atmospheric Administration (NOAA) Advanced Very High Resolution Radiometer (AVHRR) may have a potential to provide area-representative estimates of surface albedo, provided that 1) a reliable method for converting narrowband reflectances to broadband albedo values can be established (e.g., Brest and Goward 1987; Dubayah 1992; Valiente et al. 1995;Russell et al. 1997) and 2) the effect of bidirectional reflectance can be treated properly (e.g., Gao 1993; Gao and Lesht 1997). The advantages of using the high-resolution narrowband data are that 1) they are available at desired fine spatial resolution with sufficient temporal and spatial coverage and 2) atmospheric effects on remote sensing measurements for individual narrow bands can be corrected with reasonable confidence. The disadvantages are that uncertainties can be induced in 1) the conversion from satellite-derived narrowband reflectances to broadband surface albedo and 2) the conversion from the satellite-derived directional reflectances to spherically integrated surface reflectances or albedo. Although the effect of bidirectional reflectance is also important for estimating hemispheric surface albedo from directional narrowband satellite data, this study is focused on the narrowband-to-broadband (NTB) conversion by using the data from the parallel field measurements of nadir, narrowband reflectances, and broadband albedo to develop the improved NTB relationships necessary for estimating surface albedo from AVHRR data.
The narrowband reflectances in the visible (VIS) and near-infrared (NIR) spectra, such as provided by AVHRR channels 1 and 2, are used most frequently to characterize land surface conditions; they have been used to estimate surface albedo in only a limited number of studies. Vegetation has very distinct optical characteristics in these two spectral wave bands. Thus, one of the problems is that the NTB conversions for the AVHRR VIS and NIR wave bands may not be stable because the amount of green vegetation tends to have a much stronger influence on the reflectances within these two wave bands than on the reflectances within other wave bands of the solar spectrum. For example, green leaves absorb most incident radiation in the red band but reflect most incident radiation in the NIR band. In contrast, soil has a much smaller reflective difference between the two spectral regions. A simulation study by Vulis and Cess (1989) indicates that NTB conversion coefficients are larger for vegetated surfaces than for ocean or desert. In many studies, different NTB conversion coefficients have been suggested for the use of data from AVHRR channels 1 and 2, as summarized in Table 1. Some of these studies addressed the conversion for measurements at the top of atmosphere (e.g., Wydick et al. 1987; Hucek and Jakobowitz 1995) and some of them addressed the conversion for ground-level reflectances (e.g., Russell et al. 1997; Saunders 1990; Brest and Goward 1987). The present investigation also addressed the ground-level conversion. The uncertainty in determining these coefficients makes it difficult to use the high-resolution AVHRR data to derive the spatial variation of surface albedo. The two hypotheses to be tested here are that 1) the NTB coefficient for AVHRR channel 1 increases with increased amounts of green vegetation because green leaves tend to have relatively stronger absorption (weak reflection) in this band and 2) the NTB coefficient for AVHRR channel 2 decreases with increased amounts of green leaves because green leaves tend to have relatively stronger scattering or reflection in this band. Thus, a reliable algorithm for NTB conversion suitable for major land-use types may need to include the description of vegetative conditions.
Ground-based simultaneous measurements of narrowband reflectances and broadband albedo
Multispectral reflectance (MSR) and broadband albedo were measured simultaneously at a total of 18 sites of solar and infrared radiation observation stations (SIROS) that were located over grassland within the Southern Great Plains (SGP) Cloud and Radiation Testbed (CART) site in Oklahoma and Kansas (Stokes and Schwartz 1994) during the summer of 1996 and over various agricultural surfaces near DeKalb, Illinois, during the summer of 1997, as summarized in Table 2. Overall data represent measurements under sunny conditions, with only a few data samples containing some cloud effects. A multispectral radiometer (MSR87) was used to measure the surface reflectances. The MSR87 instrument measures downwelling and upwelling radiation fluxes with narrowband interference filters for eight narrow wavelength bands, shown in Table 3, with the bands centered at 0.46, 0.51, 0.56, 0.61, 0.66, 0.71, 0.76, and 0.81 μm, respectively. The MSR87 bands 4 (0.585–0.635 μm) and 5 (0.635–0.685 μm) approximately cover the wavelength range of AVHRR channel 1 (0.58–0.68 μm), and MSR87 bands 7 (0.735–0.785 μm) and 8 (0.785–0.835 μm) are close to the wavelength range covered by AVHRR channel 2 (0.725–1.10 μm). Because the MSR channels are much narrower than AVHRR channels and do not completely fill the AVHRR wavelength ranges, particularly for AVHRR channel 2, it is difficult to obtain an equivalent AVHRR reflectance from the MSR measurements by integrating reflectances within the wavelength range for each of the AVHRR channels according to AVHRR response functions. Instead, the average of measured reflectances for MSR bands 4 and 5 was used to represent the equivalent reflectance for AVHRR channel 1, and the average of measured reflectances for MSR bands 7 and 8 was used to represent the equivalent reflectance for AVHRR channel 2. Although such average reflectances may not represent the “true” reflectances that the AVHRR would measure, they provide a good approximation for the reflectance in the red and NIR wave bands. The normalized difference vegetation index (NDVI), which will be used in the present method, was estimated using the difference between the measured reflectances in NIR and VIS bands normalized by their sum. The upward-facing sensor was covered with an opal glass cosine diffuser to measure solar irradiance, and the downward-facing sensor was always operated at the nadir view position with a fixed field-of-view (FOV) angle of 48°. At the SIROS sites, the upwelling and downwelling fluxes within 0.3–2.8 μm were measured with downward- and upward-facing pyranometers, respectively. The albedo value was calculated as the ratio of the upwelling to the downwelling fluxes. At the Illinois site, the surface albedo was measured with an albedometer, which outputs surface albedo values directly. The MSR87 instrument was either held by hand for short-time operation or was mounted on a tripod for diurnal observations, with the sensor normally about 2 m above the ground. At the SIROS sites, the pyranometers were mounted at about 10 m above the ground, so MSR measurements were taken at multiple points in the immediate vicinity to obtain average values for the pyranometer footprints. At the Illinois site, the MSR87 instrument and the albedometer were collocated at the same height. Most of the measurements at this site were made over a fairly uniform surface, and thus the difference in FOV between the albedometer and the MSR87 instrument should have a minimal effect on the comparison between albedo and MSR data.
All SIROS measurements were over grassland, but most of the measurements in Illinois were over more diversified surface types, as shown in Table 2. The narrowband measurements without simultaneous albedo measurements available are also listed in Table 2 because they were used to infer NTB conversion functions as described in section 3. The albedo measurements, in fact, were only used to check the validity of the developed functions. The total data, however, do not include other important surface types such as forest, water, and snow. Thus, results from this study may be limited to the range of surface conditions represented, although methods developed here may be extendable to other surface types when appropriate data become available.
Measurements of surface albedo from unmanned aerospace vehicles and AVHRR measurements of spectral reflectances
The upwelling and downwelling radiation fluxes within the shortwave region were also measured by the shortwave broadband radiometers on board unmanned aerospace vehicles (UAVs) flying at altitudes between 800 and 1800 m within the SGP CART site during the Atmospheric Radiation Measurement (ARM) program’s Enhanced Shortwave Experiment in October 1995. The data used here were obtained on 11 October 1995 during UAV horizontal flights along four tracks over a path of about 180 km from the central facility to the western edge of the CART site between midmorning and noon (Gao et al. 1998a). The upwelling and downwelling fluxes from the UAV data were measured by upward- and downward-facing pyranometers at an approximately level position adjusted during flight. They represent spherical integrated radiances weighted by cosine angle distribution. The surface albedo was estimated by the ratio of the upwelling to downwelling fluxes along the flight path. Atmospheric effects, such as light scattering by aerosols between the ground and the UAV flight altitudes (800–1800 m) may have influenced the measured albedo, but spatial variations in surface albedo appear to be much greater than the atmospheric effects, as further discussed in section 4.
The AVHRR spectral reflectance data used for comparison with UAV results were obtained from the ARM data archive, and the measurements were taken by the NOAA-14 satellite at 1300 CST 11 October 1995 when solar zenith angle was near 54°. Atmospheric corrections to the AVHRR data were made by using the LOWTRAN-7 model and in situ measurements of atmospheric temperature and humidity (Gao et al. 1998b). The corrected AVHRR 1-km-resolution data then were extracted for the surface areas under the UAV flight path. The surface reflectances in channels 1 and 2 were used to calculate the surface albedo with the method described in section 3, and the calculated albedo values were compared with the corresponding UAV-measured values, as discussed in section 4.
Method for conversion from AVHRR narrowband reflectances to albedo
The surface albedo (δ), defined as the fraction of the solar irradiance reflected back to the sky from the earth’s surface, can be expressed as
where ρ(λ) and S(λ) represent the spherical reflectance and the downwelling solar irradiance, respectively, at the wavelength λ. Wavelength integration from λ1 = 0.3 μm to λ2 = 2.8 μm covers most of the solar spectrum. To use the reflectances within a few narrow wave bands observed by AVHRR channel 1 (λ = 0.58–0.68 μm) and channel 2 (λ = 0.725–1.10 μm) to estimate the total reflectance within the whole solar spectrum, an improved understanding is required of the relationship between the reflection characteristics of the narrow bands and that of the broad band for various surface types. In this study, the solar spectrum is divided into two subregions: the VIS subregion (0.3–0.685 μm) and the NIR subregion (0.685–2.8 μm); and the reflectance relationships between AVHRR channel 1 and the VIS subregion and between AVHRR channel 2 and the NIR subregion are examined. The wavelength of 0.685 μm was chosen to separate VIS and NIR to be consistent with the waveband arrangement used by the MSR instrument that provides the data for this analysis. With the definition of the VIS and NIR spectral subregions, Eq. (1) can be converted into the following discrete form, in which the first term in the numerator is the reflected radiation for the VIS subregions and the second term is the reflected radiation for the NIR subregion:
Here the VIS and NIR spectral subregions are divided at the wave band m′ for the numerical integration, and m denotes the total number of wave bands used in the numerical integration. We assume that the VIS- and NIR-reflected radiation represented by each term in the numerator of Eq. (2) can be parameterized by a corresponding narrowband reflected radiance accompanied by appropriate conversion factors f1 and f2, respectively, in the following form:
Here the subscripts υ and n represent the narrow wave bands in the VIS and NIR channels, respectively. Now the problem of estimating δ from ρυ and ρn yields to determining the new conversion factors f1 and f2; the fractions of the downwelling solar radiation within the specified narrow bands are independent of surface conditions. Comparing Eqs. (3) with (2) indicates that the values of f1 and f2 are determined by the following equations:
The physical meaning of the conversion factors f1 and f2 is obvious from (4a) and (4b). They quantify the ratio of the reflected radiances within the narrow bands to the reflected radiances for the VIS and NIR broad bands, or the degree of the representativeness of the narrowband reflectance for the broadband reflectance. If we replace ρυ and ρn with the reflectances derived from AVHRR channels 1 and 2, as shown in section 4, the values of f1 and f2 can be related to the empirical coefficients shown in Table 1 for estimating the surface albedo from AVHRR data, as further discussed in section 4.
Calculating f1 and f2
To estimate f1 and f2 values from the MSR-based measurements described in section 2, the solar spectrum from 0.3 to 2.8 μm was divided into 11 wavebands (Table 3). Similar wave bands were used for the calculation of solar irradiance within the solar spectrum shown in Fig. 1, which was calculated with the atmospheric radiation transfer model LOWTRAN-7 by using radiosonde and aerosol data taken at the CART site (Gao et al. 1998b). The transmitted solar irradiance at the surface, averaged over the assigned wave bands, is also shown in Fig. 1. The division of wave bands within the VIS subregion in Table 3 was made to be consistent with the channel assignment of the MSR instrument, and the waveband division within the NIR subregion was determined on the basis of the characteristic spectra of leaf reflectance and soil reflectance and their relative importance to the total reflectance. Because of the lack of measurements beyond 0.835 μm and because the contribution of the reflectance from the point of the spectrum beyond 0.835 μm decreases sharply due to the decrease in the incident solar radiation, this region is represented by only two wave bands (bands 10 and 11). The reflectance for band 1 was assumed to be the same as band 2, as indicated by the data of surface reflectances from various vegetation and soil (Bowker et al. 1985) and the reflectances for the remaining eight bands were measured with the MSR instrument.
The reflectances for the spectral region of λ > 0.835 μm were not measured and thus had to be parameterized. Both vegetation and soil have distinct changes in reflectance around a wavelength of 1.3 μm, as indicated by the typical spectra of reflectances by a plant leaf and soil shown in Fig. 2, and thus the spectral region beyond λ = 0.835 μm was divided at 1.3 μm into bands 10 and 11. Vegetation and soil appear to have different spectral shapes in this region, and their reflectances were parameterized separately as ρc for vegetation and ρs for soil. The total reflectances from the vegetation–soil system were estimated from the appropriate combination of ρc and ρs.
Because the ρc value is nearly invariant between 0.7 and 1.3 μm but drops by about one-half between wavelengths of 1.3 and 2.8 μm, the assumptions of ρc10 = ρ9 and ρc11 = 0.5ρ9 were used. Because the ρs value increases almost linearly from bands 6 to 10 and flattens out after band 10, the extrapolations of ρs10 = ρ9 + (λ10 − λ9)(ρ9 − ρ6)/(λ9 − λ6) and ρs11 = ρs10 were used. The total reflectance from the soil and vegetation system was estimated from the ρc and ρs weighted by the fraction fc of vegetation cover, which was estimated with NDVI. The fc value was parameterized to increase linearly with NDVI in the form of fc = 1.333NDVI − 0.267, which was obtained by assuming fc = 0 for NDVI = 0.2 and fc = 1 for NDVI = 0.95.
f1–NDVI and f2–NDVI relationships
The MSR data gathered from the measurements over various surfaces shown in Table 2 were used to calculate the values of the conversion factors f1 and f2 by using the methods described in Eq. (4) and the integration over the wave bands defined in Table 3. A total of 358 data samples were used to obtain the results shown in Figs. 3a and 3b for f1 and f2, respectively. The f1 value (Fig. 3a) appears to correlate with NDVI (r2 = 0.76). The f1 value does not show clear dependence for the data with a relatively smaller NDVI (<0.45), representing bare soil and soil covered with a small amount of green vegetation, but it shows a clear trend of decreasing with increased NDVI values for vegetated surfaces. The decrease in the f1 value with increasing NDVI is consistent with the observation that green leaves have a very strong absorption in the red wave band so that the f1 value decreases with increasing vegetation density.
Calculated f2 values show a stronger relationship to NDVI (Fig. 3b) (r2 = 0.98). The f2 value reaches a minimum near an NDVI of 0.4. When NDVI is smaller than 0.4 (soil and low vegetation cases), the f2 value decreases with increasing NDVI, indicating that the magnitude of the reflectance in the NIR band relative to the total reflectance from the NIR spectral subregion decreases for the soil with smaller NDVI. As NDVI increases, the importance of the vegetation contribution gradually increases. Above NDVI = 0.4, the f2 value increases rapidly. The increase in the f2 value can be attributed to the fact that green leaves absorb very little radiation in the NIR band and most of incident radiation is either transmitted or reflected, resulting in strong leaf scattering. The large leaf scattering coefficient, in turn, leads to a strong reflectance from the whole canopy through multiple scattering among the leaves within the canopy (Gao 1993). Thus, for vegetated surfaces with an NDVI greater than 0.4, the relative contribution of the reflectance within the NIR band to the total reflectance within the whole NIR subregion appears to increase with increasing NDVI or canopy density.
Because NDVI was used to parameterize ρ10 and ρ11 in the calculation of the f2 values, the f2–NDVI relationship shown in Fig. 3b could be influenced by the assumed ρ10–NDVI and ρ11–NDVI functions shown in Table 3. To examine this possibility, we recalculated the f2 values without using the NDVI to specify the ρ10 and ρ11 values. Figure 4 shows the f2 values calculated by assuming ρ10 = ρs10 and ρ11 = ρs11 (for an assumed soil case) and by assuming ρ10 = ρc10 and ρ11 = ρc11 (for an assumed canopy case). For the assumed soil case, the f2 value decreases with increasing NDVI, and for the vegetation case, the f2 value increases linearly with increasing NDVI. The f2–NDVI curve shown in Fig. 3b in fact reflects the combined effects of these two extreme cases weighted by the canopy density described by NDVI. Thus, although the f2 value changes with NDVI, the quantitative f2–NDVI relationship may be influenced, to some degree, by the empirical functions assumed for ρ10 and ρ11 shown in Table 3. Better parameterization of ρ10 and ρ11 may lead to further improvement in the parameterization of NTB conversion factors.
The finding that the values of the NTB conversion factors f1 and f2 change with vegetative conditions is important for understanding uncertainties involved with using satellite-derived narrowband reflectances to infer broadband surface albedo. Many previous studies employed different weighting coefficients to use the reflectance data from AVHRR channels 1 and 2, as summarized in Table 1. As described in the following paragraph, the second-order functions obtained here to describe the f1–NDVI and f2–NDVI relationships can be used to determine these empirical weighting coefficients.
The AVHRR reflectances from channels 1 and 2 are often employed to estimate surface albedo by using the following linear combination:
Here w1 and w2 are the fraction of solar radiation within the wave bands of AVHRR channels 1 and 2, respectively. The values of w1 and w2 were estimated to be 0.1464 and 0.1050, respectively, according to the solar radiation spectrum calculated by using the LOWTRAN-7 model. Figure 5 shows the c1–NDVI and c2–NDVI relationships obtained from the f1–NDVI and f2–NDVI relationships. The resulting c1 and c2 values change within a range of 0.32–0.52 and 0.40–0.84, respectively. For a small NDVI representing nonvegetated surfaces, the c1 and c2 values are close to 0.4 and 0.6, respectively;the c1 value reaches a minimum of about 0.32 and the c2 value reaches a maximum of about 0.84 for moderate NDVI values representing partially vegetated surfaces;and the c1 value increases to about 0.52 and the c2 value decreases to about 0.4 for large NDVI values representing densely vegetated surfaces.
Adding the NDVI dependence into the parameterization of c1 and c2 allows the empirical method of Eq. (5) to account for influences of vegetative conditions on the NTB conversion. For vegetated surfaces with an NDVI greater than about 0.4, because f1 decreases with increasing NDVI, c1 should increase with increasing NDVI to compensate for the larger reduction of reflectance in the red narrow band than those within the whole VIS subregion. Similarly, because f2 increases with increasing NDVI, c2 should decrease with increasing NDVI to offset the larger increase of reflectance in the NIR narrow band than those within the whole NIR subregion because of the strong in-canopy multiscattering associated with this specific narrow band.
Comparison of calculated albedo with ground-based measurements
As shown in Table 2, both multispectral reflectance and albedo were simultaneously measured over various surfaces during the field experiments conducted across the SGP CART site and in Illinois. At each of the sampling locations, the two instruments were closely spaced in order to observe the same surface. Because the pyranometers (at 10 m) and the albedometer (at 3 m), which measured spherical-integrated (including cosine effect) radiation fluxes, had different FOVs from those for the MSR instrument (which has a smaller FOV angle of 48°), the MSR measurements were taken at multiple points for each location to obtain appropriate average values over the local area covered by the approximate FOV of albedo instruments. The differences in the measured surface areas between these two sets of measurements were estimated to be minimal.
Figure 6 shows the spectral reflectances measured by the MSR instrument at various sites covered by different surface types. The relatively lower reflectance near λ = 0.65 μm and the sharp increase in the reflectance near λ = 0.8 μm are typical for surfaces covered by green vegetation, and the resulting large reflectance differences produce much larger NDVI values for vegetated surfaces than for nonvegetated surfaces. These distinct shapes of the reflection spectra for vegetation also can change the values of f1 and f2, as has been described in previous sections.
Figure 7 compares the albedo values calculated by different methods with corresponding measured albedo values. The present method, which uses the NDVI-dependent NTB conversion factors, appears to improve the estimation of surface albedo, compared with other empirical methods. Values of the root-mean-square error between the calculated and measured albedo are 0.045, 0.034, and 0.087 for the methods of Brest and Goward (1987), Saunders (1990), and Russell et al. (1997), respectively, using constant coefficients (Table 3), and that error value is 0.022 for the present method using the NDVI-dependent c1 and c2 as shown in Fig. 5. The improvement with the new method appears to be especially significant for the data with smaller albedo over some vegetative surfaces. Most of the data for vegetation cases were collected over uniform fully covered canopies. The behavior of the improved method over partially vegetated surfaces and over forest canopy needs to be further investigated. In addition, the data for bare soil and other nonvegetated surfaces are limited in the present studies. More data need to be gathered to test the consistency of the method on nonvegetated surfaces.
Comparison of calculated albedo with UAV-based measurements
The UAV-based measurements of surface albedo provide a unique opportunity to evaluate the worthiness of the spatial variation of estimated albedo based on AVHRR 1-km resolution data. Figure 8 shows the albedo values calculated from the reflectances of AVHRR channels 1 and 2 along the four UAV flight tracks with a similar path and the corresponding UAV-based measurements of surface albedo. The UAV flew four times over the same general flight path, although slight shifts may have occurred among the four tracks. The patterns of change in albedo along the first half of the path were quite similar among the four tracks, but the patterns were less consistent along the second half of the path.
The same AVHRR data extracted for the path were used to calculate the albedo for the different UAV flight tracks. The estimated albedo values from the independent AVHRR data have a spatial variation that closely matches the spatial variations observed by the UAV. This match confirms that the extracted satellite data had an FOV very close to that for the UAV and that the variations are caused largely by changes in surface conditions. The navigation error of AVHRR was about 1 km. The result shown in Fig. 8 indicates that this small error did not significantly influence the match between these two datasets in terms of their representation of relative spatial variations in surface reflectance. The magnitudes of the calculated albedo values, however, depart from the measured values for all four methods used. The methods of Brest and Goward (1987) and of Saunders (1990) appeared to significantly underestimate the surface albedo, compared with UAV-based measurements. This result is consistent with the results shown in Fig. 7, which indicate that these two methods underestimate ground-based measurements of surface albedo. The method of Russell et al. (1997) predicted quite closely the UAV-derived surface albedo for the first and second tracks, probably because that method generally produces relatively higher albedo values, as shown in Fig. 7, but overestimated the UAV measurements for the third and fourth tracks. The present method underestimated the UAV-measured surface albedo for the first and second tracks, but the difference decreases for the third and fourth tracks.
The UAV-derived albedo values were slightly higher than the estimates with the new method; this fact may be partially attributed to the effect of atmospheric scattering within the layer between the surface and the UAV flight altitudes. The atmospheric backscattering tends to increase the upwelling radiation flux and the UAV-derived albedo. Another possible cause may be the different solar zenith angles associated with UAV and AVHRR data. The solar zenith angle ranges for each UAV track were 65° for track 1, 60° for track 2, 56° for track 3, and 50° for track 4. The AVHRR measurements were taken near the local solar noon at about 1300 CST, and the UAV measurements for all four tracks were taken with relatively larger solar zenith angles during the earlier times, which tend to be associated with larger albedo values because of increased light scattering within the surface canopy and atmospheric layer due to a longer light path. This argument may explain why the difference between estimated and measured albedo decreases for the measurements taken close to noon for the two later tracks. The characteristics of surface bidirectional reflection may also play a role in producing the differences between the measured values and values estimated by various methods. However, the bidirectional-reflection effect cannot be separated using the existing data (only flux measurements were available) and could be further investigated if measurements of directional radiance become available.
In this study the main difficulty has been addressed for estimating surface shortwave broadband albedo from satellite observations of narrowband reflectances in the red and NIR wave bands such as AVHRR channels 1 and 2. Analyses based on simultaneous measurements of both multispectral narrowband reflectances and broadband albedo reveal that NTB conversion factors change with surface conditions, particularly the amount of vegetation. The relationships between two NTB conversion factors, f1 and f2, and NDVI were established from analysis of data taken over various surfaces. The f1–NDVI and f2–NDVI relationships developed through this study may be employed to infer the empirical coefficients often used to estimate surface albedo from AVHRR data.
The surface albedo calculated by using the new method has a better agreement with measurements than the previous empirical methods. The validity of the method was also examined by using the albedo measurements obtained by a UAV over a 180-km flight track covering large spatial variations of surface conditions and corresponding AVHRR 1-km-resolution data extracted for the UAV flight path. The new method that uses the NDVI-dependent NTB conversion factors improves the estimation of surface albedo on the basis of the AVHRR measurements. This investigation was focused on the spectral conversion from AVHRR narrow bands to albedo. Effects of bidirectional reflectance by surface and the atmosphere in association with changing solar angles were not addressed in detail. Improving the overall accuracy of estimating surface albedo from satellite data may require including these effects by developing a more complete empirical method that includes solar zenith angle as a controlling variable.
This work was supported by the U.S. Department of Energy, Office of Energy Research, Office of Health and Environmental Research, under Contract W-31-109-Eng-38 through the Atmospheric Radiation Measurement Program, and by the U.S. Department of Defense, Strategic Environmental Research and Development Program, through the ARM Unmanned Aerospace Vehicle Program. The lead author carried out this work during her summer research with the Argonne Faculty Research Participation Program and with funding support provided by Northern Illinois University summer research and artistry funds.
Corresponding author address: Dr. J. Song, Department of Geography, Northern Illinois University, DeKalb, IL 60115.