0 tree
ectral
IS can
stand
to the
na for
0409-
Using
sure".
yectral
rborne
87. A
rborne
e used
> XS,
It was
e; Tie
- on
17/44.
study.
spatial
3.Data Processing
The image processing techniques of these remotely
sensed data were implemented through the use of the
Earth Resources Data Analysis System (ERDAS). In
the geometric correction aspect, 22 * 21 and 26 widely
scattered ground control points were selected
respectively from the airborne MSS, SPOT HRV and
Landsat-TM images to compute the equations that
transformed the images into the Transverse Mercator
Coordinate System of the base map. The pixels of these
remotely sensed data were resampled to a 10m X 10m »
20m x 20m and 30m X 30m respectively by the nearest
neighbor algorithm, The root-mean-squared error
(RMSE) of the linear transformation model of these
three images were 0.831 ~ 0.711, and 0.489.
4.Independent Variables
On the sample plots selecting, forest maps of the study
area were overlaid on the three remotely sensed images
respectively by the assistance of GIS data. The total 78
sample plots were drawn systematically from each
image. The digital data utilized in this study were
collected from these sample plots. The independent
variables of the equation in this study were the digital
mumber of individual band, vegetation index, tree ages
and crown closures. The digital number of individual
band were collected from each sample plots. Six
vegetation indices were computed from the equation as
follows IND] =NIR-R, IND2-(NIR-R)/(NIR*R),
IND3=(IND, +053, IND4=NIR-G, IND5=(NIR-
G)/(NIR+G), and IND6= (IND5+0.5) ^, where, NIR is
near-infrared band, R is red band , and G is grcen band.
Tree ages were adopted from tree plantation's records.
Forest crown closures were measured with a crown
density scale under a stereoscope from the
corresponding plots on the aerial photographs.
5.Regression equtation
Regression equations were derived from multiple
regression analysis by using the digital number of
spectral individual bands and vegctation indices. The
individual band and vegetation index were the
independent variables, and the forest crown closures and
trce volumes were the dependent variables. The stepwise
regression approach in the Statistical Analysis System
(SAS) package was used to select the equations. Among
the selected equations, one optimum equation was
selected by signifjcant F-value and higher coefficient of
determination (R)value.
RESULTS AND DISCUSSIONS
1.Individual Band
As mentioned precviously, the Statistical Analysis
System (SAS) was used to perform the multiple
regression analysis. In these regression equations, the
individual band of remotely sensed data were used as the
independent variable, and the forest crown closure
which was measured in the corresponding plot on the
acrial photographs was the dependent variable. Stepwise
regression approach was uscd to selection the equation.
Only the one with a significant F-value and higher R2
value was adopted. Table 2 was the result of regression
analysis of crown closure for the three indiviaual bands
of the airborne multispectral scanning data. Selected
individual bands were MSSs(green) ^ MSSg(near-
infrared) and MSSo(near-infrared). Follow the same
125
procedure, the results of regression analysis of stand
volume for three individual bands of airborne
multispectral data was obtained. The three selected
individual bands of stand volume equation were
MSS5(green) » MSSgo(red) and MSSg (near-infrored).
The SPOT data and Landsat-TM data were treated in
the same way. All of the three SPOT XS bands were
selected in the crown closure and stand volume
estimation equation. In the Landsat data aspect, the
TM, * TM, and, TMs were selected in the crown closure
equation; the TM; * TM; and TM, were selected in the
stand volume estimation equation. The results were
shown in Table 3.
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B7. Vienna 1996