ultraviolet reflectance for several illumination 
and canopy conditions. Therefore, the 
ultraviolet spectral region reflectance is 
assumed to be represented by the visible 
reflectance. 
The absence of a shortwave middle-IR detector 
on the AVHRR may introduce errors to over 20% in 
comparison to using a near-IR reflectance (0.72- 
1.30 ¡jinx) to represent a total shortwave IR region 
(0.72-1.30 /Jm) (Toll 1989). Brest and Goward 
(1987) used 0.5 times the near-IR reflectance to 
estimate a shortwave middle-IR reflectance. 
However, the near-IR optical properties for green 
vegetation are markedly different than for the 
middle-IR with a substantially higher absorptance 
of solar radiation (three to eight times higher 
green leaf absorptance) in the middle-IR. 
Landsat Thematic Mapper (TM) satellite 
derived reflectance in the visible and near-IR 
regions were compared against a middle-IR 
reflectance for the purpose of estimating a 
middle-IR reflectance from the AVHRR sensing in 
only the visible and near-IR regions of the solar 
band. The TM has three bands in the visible, one 
in the near-IR, and two in the middle-IR. The TM 
spectral data were radicmetrically calibrated to 
radiance, corrected for Sun zenith angle and 
Earth-Sun distance, and converted to an 
exoatmospheric reflectance using Neckel and Labs 
(1984) solar irradiance data. The available TM 
digital scenes in western Africa for analysis was 
low due to few scene acquisitions, a high cloud 
cover, and a low satellite overpass cycle (i.e., 
once every 18 days) . The Landsat Thematic Mapper 
data for three dates were analyzed. The three 
Landsat IM scenes selected for analysis were on 
August 21, 1984 (17057’N 8°0’E scene center 
point) September 24, 1984 (15055’N 14° 55’W), and 
October 23, 1986 (14o27’N 16o44’W). 
Regression analysis results between derived 
spectral reflectances (i.e, band combinations 
between the visible and near-IR versus the 
middle-IR spectral regions) are given in 
Tables 2(a-c). To reduce spatial autocorrelation 
effects and hence statistical interdependence, 
the IM pixel data were sampled by at least every 
tenth pixel (or greater) in both the across and 
along scan line directions (Labovitz et al. 
1982). An estimated exoatmopsheric NDVI is also 
included for an indication of green leaf 
vegetative density. 
i ! j S V° rtwaVe ^àdl^-JR spectral reflectance estimât 
related analyses using October 23, 1986 Landsat TM data. 
ANOVA STATISTICS 
TM-Band# 
Indep. Var. 
2 
3 
4 
mat 
(2*3) 
(2*3)*4 
(2*3) ,4*MAT 
Offset* Slope* 
-15(.00) 3.06(.03) 
-.02(.00) 2.18( .01) 
.23( .01) 0.10(.04) 
■ 48(.00) -0.56(.00) 
-. 08 (. 00) 2.92(. 02) 
F-Value 
R2 
c.v. 
10308.6 
0.82 
8.44 
23928.0 
0.91 
5.85 
6.3 
0.00 
19.71 
12596.8 
0.84 
7.77 
18506.8 
0.89 
6.57 
11277.5 
0.91 
6.01 
8594.6 
0.92 
5.65 
* - Standard error given in parentheses. 
Pearson Correlation Coefficients 
2 
2 
3 
4 
(2*3) 
15*7) 
3 
.96 
4 
.18 
.61 
(2*3) 
.98 
.99 
.09 
(5*7) 
.90 
.96 
-.05 
.94 
Mm 
-.83 
-.93 
.33 
-.90 
-.92 
Table 2b. Shortwave Middle-IR spectral reflectance estimate 
related analyses using September, 24, 1984 Landsat TM data. 
Indep. Var. 
Offset* 
Slope* 
F-Value 
R2 
C.V. 
2 
.14(.01) 
1.40(.03) 
1813.6 
0.36 
3.64 
3 
. 08 (. 00) 
1.31(.02) 
6432.4 
0.67 
2.63 
4 
,06(.01) 
1.31(.02) 
4694.1 
0.59 
2.90 
MAT 
.46(00) 
-. 42 ( . 03) 
250.3 
0.07 
4.37 
(2*3) 
08(.01) 
1.48(.02) 
3920.0 
0.55 
3.06 
(2*3) *4 
- 
- 
2607.1 
0.62 
2.81 
(2&3),4,JfcNDVI 
- 
- 
3589.2 
0.77 
2.18 
* - Standard error given in parentheses. 
Pearson Correlation Coefficients 
2 
2 
3 
4 
(2*3) (5*7) 
3 
.88 
4 
.79 
.86 
— 
(2*3) 
.96 
.97 
.85 
— 
(5*7) 
.59 
.81 
.77 
.74 
MAT 
-.34 
-.46 
.05 
-.42 -.26 
Table 2c. 
October 
23, 
1986 
, Pearson 
correlation coefficients. 
(2&3) 4 (5Ä7) 
(2&3) 
4 .83 
(5Ä7) .80 . 84 
Overall the relationship of the visible and 
near-IR reflectance to the middle-IR reflectance 
is strong, with a multiple regression coefficient 
(R) of 0.62 and higher (Tables 2a and 2b) . 
Except for the 1984 scene (r= - 0.05) the linear 
relationship of the near—IR to shortwave middle- 
IR reflectance relationship is strong (r=0.77 and 
r=0.84). In ccmparison, the linear relationship 
of a middle-IR to a visible reflectance is strong 
for all three dates (0.74<r<.94). A reason for 
the closer link between visible and middle-IR 
derived reflectance over the link between near-IR 
and middle-IR may be attributed to a closer 
similarity of leaf absorptance related effects 
between regions. Specifically, the plant 
pigments (primarily chlorophyll) absorb radiation 
in the visible and canopy water absorb radiation 
in the middle-IR. Of note, the strength of the 
visible to the middle-IR relationship is not 
significantly improved by the addition of NDVI in 
the analysis—of-variance for the dry October 23, 
1986 scene (Table 2a). However, for the densely 
vegetated Sept. 24, 1984 scene, the incorporation 
of an NDVI improved the estimation of a middle-IR 
reflectance (Table 2b). Based on the TM related 
findings and examination of the spectra published 
in Bowker et al. (1985), we estimated an area 
specific middle-IR reflectance frcm the AVHRR 
spectral data by multiplying the AVHRR visible 
reflectance by 1.5 (i.e., £>smir = Pvis * 1-5). 
Table 3 gives the percentage of solar 
irradiance incident at the surface in terms of 
diffuse, direct and global (direct + diffuse) 
radiation integrated over the four major spectral 
regions. The relative proportion of surface 
radiation by spectral region gives the weighting 
for the solar band estimation in Equation 5. The 
atmospheric radiative transfer data of Dave’ 
(1978) using three mid-latitude models (Model 2: 
atmospheric gaseous absorption with no aerosols; 
Model 3: gaseous absorption with a low aerosol 
loading; and Model 4: gaseous absorption with a