tanbul 2004
1odel.
overlaid
be adjusted
on each tree
In order to
nsity at the
the plot are
d to the top
. defined as
nsity within
anopy base
the canopy
11 kg/m).
st height at
011 kg/m’ ;
1el weight,
iopy height
An example
the control
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences
Figure 4. Estimation of fuel parameters at inventory plot within
dense canopy unit, Capitol Forest study area.
32 IFSAR canopy fuel estimation
Multiple regression analysis was used to develop a
mathematical model relating the radar backscatter and
interferometric information to the observed fuel parameters at
the plot level. At each inventory plot, the observed IFSAR data
from all passes were extracted and aggregated.
Summary statistics for each set of aggregated plot-level IFSAR
observables were then calculated and used as predictor
variables. These data included mean, median, maximum,
minimum, and coefficient of variation of: 1) Canopy heights
(X-band minus P-band elevations) 2) X-band backscatter , 3)
X-band coherence, 4) P-band VV backscatter, 5) P-band VH
backscatter, 6) P-band HH backscatter, 7) 1* optimized
coherence (P-band), 8) ond optimized coherence (P-band) , 9) a
optimized coherence (P-band), 10) Wrapped phase (X-band),
and 11) 3" optimized unwrapped phase (P-band) .
In addition, an IFSAR-based canopy density estimate was
caleulated as the percentage of observed canopy elevations
greater than 2 meters in height. Stepwise and "best" subsets
variable selection algorithms were then used to develop the
"best fit" model for each fuel parameter at the plot level.
4. RESULTS
41 Canopy height
The correlation between IFSAR observables and canopy height
was quite high, with a coefficient of multiple determination (RY)
of 0.89 and an adjusted R? of 0.88. It should be noted that the
adjusted R^ measure will place a penalty on models with
numerous extraneous predictor variables. ^ A scatterplot of
field-based canopy height measurements versus predicted
values is shown in Figure 5.
1033
Vol XXXV, Part B3. Istanbul 2004
Field Canopy Height (m)
T T T T T
10 20 30 40 50
Predicted Canopy Height (m)
Figure 5. Field-based (y) versus predicted canopy height
measurements (x) (with 1:1 line shown).
4.2 Canopy base height
The regression model developed to predict canopy base height
had an R? of 0.85 and adjusted R? of 0.83. A scatterplot of field-
based canopy base height measurements versus predicted values
is shown in Figure 6.
Field Canopy Base Height (m)
T T T T T T
0 5 10 15 20 25 30
Predicted Canopy Base Height (m)
Figure 6. Field-based (y) versus predicted canopy base height
measurements (x) (with 1:1 line shown).
4.3 Canopy bulk density
The regression model developed to predict canopy bulk density
had an R? of 0.74 and an adjusted R^ of 0.71. In this case a
logarithmic transformation of the independent variable was
used to stabilize the error variance. It should be noted that the
