complicated transformations obtained by band group ratios did not improve the
correlation with above-ground forest carbon compared to the original TM bands and
their simple band ratios. The TM band 3 and its inversion were most highly correlated
with above-ground forest carbon with correlation coefficients of −0.3359 and 0.3521,
respectively. That is, in the areas with higher values of above-ground forest carbon,
there was more light absorbed and less light reflected in the red channel (Figure 6.2).
The above-ground forest carbon thus had negative correlation with the red channel
and positive correlation with its inversion. The inversion of TM band 3 was used in
the cosimulation to generate above-ground forest carbon maps.
Various spatial autocorrelation models were used and compared to fit the sample
variogram based on goodness of fit, a measure for quantifying how well the model fits
the sample data based on the sum of squares due to error. The closer to zero the
goodness of fit, the better the fit was. It was found that the spherical model fit the trend
of spatial autocorrelation best (Figures 6.3a, b). When the original data set was used,
the obtained model (Figure 6.3a) was
g h = 229:4 + 65:1 1:5
h
4
− 0:5
h
4
3
"
#
(6.4)
FIGURE 6.2 (a) Landsat TM band 3 and (b) its inversion, which had highest correlation with
above-ground forest carbon.
RESULTS
117
correlation with above-ground forest carbon compared to the original TM bands and
their simple band ratios. The TM band 3 and its inversion were most highly correlated
with above-ground forest carbon with correlation coefficients of −0.3359 and 0.3521,
respectively. That is, in the areas with higher values of above-ground forest carbon,
there was more light absorbed and less light reflected in the red channel (Figure 6.2).
The above-ground forest carbon thus had negative correlation with the red channel
and positive correlation with its inversion. The inversion of TM band 3 was used in
the cosimulation to generate above-ground forest carbon maps.
Various spatial autocorrelation models were used and compared to fit the sample
variogram based on goodness of fit, a measure for quantifying how well the model fits
the sample data based on the sum of squares due to error. The closer to zero the
goodness of fit, the better the fit was. It was found that the spherical model fit the trend
of spatial autocorrelation best (Figures 6.3a, b). When the original data set was used,
the obtained model (Figure 6.3a) was
g h = 229:4 + 65:1 1:5
h
4
− 0:5
h
4
3
"
#
(6.4)
FIGURE 6.2 (a) Landsat TM band 3 and (b) its inversion, which had highest correlation with
above-ground forest carbon.
RESULTS
117