Resampling Effects on Classification 
Table 4 shows the relative contrast ratio (RCR) of the mean 
grey levels at uncorrected and resampled pixels in training 
sites in band 4,5,and 6. In other words, it represents the 
ratio of difference of mean grey levels between resampled 
pixels and uncorrected pixels to those at uncorrected pixel. 
Then, it may be summarized that in each band RCR of mean gr 
ey levels between resampled pixels via NN and uncorrected 
pixels seems to be negligibly small , whereas RCR of mean 
grey levels via BI and CC seems to show noticeable anomaly 
in a few classes,i•e., sea, lagoon and broad-leaved trees 
in band 4,5, and 6. In a manner similar to the mean grey le 
vels, we computed also RCR of the covariance for eleven clas 
ses via NN,BI, and CC in band 4,5, and 6. The situation was 
found almost to be identical with those of mean grey levels. 
It is of interest to mention that the resampling effects on 
classification of Landsat MSS data have recently been dis 
cussed by several authors (cf.Etheridge and Nelson 1979;Lo 
gan 1979; Tsuchiya and Arai 1980) 
GROUND ALBEDO MAPPING 
Diffuse Radiance Reflected by Terrestrial Atmosphere 
The apparent radiance of surface features which an arbitra 
ry spacecraft measures differs from the intrinsic surface 
radiance of the object, because of the presence of interve 
ning terrestrial atmosphere bounded by the horizontally non- 
uniform albedo of Lambertian surface. Then,removal of the at 
mospheric effects on terrestrial imagery can improve the ac 
curacy of pattern recognition and image interpretation in 
remote sensing. In what follows, in Cartessian coordinate 
system we deal with three-dimensional radiative transfer mo 
del of atmosphere bounded by horizontally non-uniform dif 
fuse reflector. 
Basic equations. Suppose that top z=z^ of a plane-parallel, 
vertically inhomogeneous, anisotropically scattering atmos 
phere of optical thickness t is monodirectionally illumina 
ted by parallel rays of net flux tF per unit area normal to 
the direction of propagation. On extending the principles 
of invariance in a slab geometry (cf.Chandrasekhar 1960) to 
the three-dimensional case, it is stated such that, in ad 
dition to the atmospheric path radiance along the sensor's 
line of sight, the diffuse radiance observd from target by 
a spaceborne sensor consists of the direct and diffuse tra 
nsmission of the total irradiance at ground level.One of 
the outstanding characteristics of the algorithm is to expr 
ess the required radiance at top (or bottom) in terms of the 
reflection (or transmission ) function of the atmosphere ap 
propriate to this case. They are the initial-value solution 
of the transfer equation,i.e., the Cauchy system of the Ri- 
ccati type of non-linear integro-differential equations.Phy 
sically speaking, when an assumed diffusely reflecting tar 
get is observed from space through hazy atmosphere, the at 
mosphere modifies the radiance from the target by attenuation 
i.e., the direct and diffuse transmission, in addition to 
the path radiance. In the three-dimensional radiation fie 
ld the reflection and transmission functions are also th 
ree-dimensional so that the exact solutions seem to be from 
computational aspects tremendously untractable. Then, in our 
present case, the approximation of three-dimensional reflec