13 
X [ (Lj-Lj)/(Lj»Lj)] 
/ Ivania 
:e (0.64pm) 
ze reflectance in the red 
Renter, 1993) 
t very satisfactory, since 
id the vegetation index 
)d to be applied only to 
re present in the image. 
>r 3.7 pm) that are less 
erosol particles) and are 
d to find pixels that are 
7HRR images over the 
tive to the presence of 
itive to the presence of 
emer, 1993) (see Fig. 5). 
1 for thermal emission. 
This introduces uncertainties in the procedure. As a result an effort is made to test if a 
shorter wavelength that is not affected by emission can be used to identify forests and other 
dark pixels. Fig. 6 shows examples of the relationships between the surface reflectance at 2.13 
pm and that at 0.47 pm and 0.64 pm, derived from Landsat TM images over Washington 
DC and the Chesapeake Bay area. The uncertainty in the estimate of surface reflectance in 
the visible channels from the 2.1 pm channel reflectance is ±0.005 to ±0.01 in the red and 
blue channels respectively for dark targets (reflectance at 2.1 pm <0.1). 
reflectance ch 1 (0.47|±m) reflectance ch 3 (0.66p.m) 
Fig. 6: Example of the relationships between the surface reflectance at 2.13 pm and that at 0.47 pm 
and 0.64 pm, derived from Landsat TM images (channels 1 and 3) over Washington DC and the 
Chesapeake Bay area. The data were first reduced to resolution of 240 m. The + are for specific sites 
of different reflectance properties chosen for the analysis. The dots are subsampled pixels from the 
image. All data representing water pixel were eliminated by requiring that the apparent 
reflectance at 0.86 pm larger than 0.15. Only data for reflectance at 2.1 pm less than 0.15 are 
plotted. The uncertainty in the estimate of surface reflectance in the visible channels from the 2.1 
pm channel is ±0.005 to ±0.01 in the red and blue respectively for dark targets (for reflectance at 2.1 
pm <0.1. The dashed lines show the relationship after atmospheric correction. The correction is 
based on the aerosol parameters for the image given by Kim (1986) ???? 
An algorithm that uses these principles is being developed for global operational 
monitoring of aerosol and atmospheric correction from EOS-MODIS. The algorithm selects 
pixels that are expected to have a low reflectance in one or more of the MODIS bands (0.41, 
0.47 and 0.66 pm) and attempts to estimate their reflectance using information from other 
bands. The procedure is based on the following physical principles: 
Except for dust, the aerosol effect decreases with wavelength as A,' 1 to X' 2 (Kaufman, 
1993). Therefore the effect is much smaller in the near and mid IR than in the visible. 
The aerosol effect includes backscattering of sun light and absorption of the direct 
sunlight and light reflected from the surface. For dark surfaces the scattering effect 
dominates while for brighter surfaces the effect is mixed. As a result we can expect a 
smaller effect of the aerosol on the apparent surface reflectance for surface reflectance in 
the range 0.2 < p < 0.4. 
The surface reflectance in the MODIS bands is correlated to some extent. Soils have 
usually a smoothly increasing reflectance as a function of the wavelength with 
correlation between the reflectances slowly decreasing with an increase of the 
wavelength span. Parallel processes affect the reflectance in the 0.41, 0.47 and 0.66