The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B8. Beijing 2008 
1056 
data sets. The Voroni-plot results were generally similar to the 
plot level results, save for marginally higher RMSE values, 
which indicated that modelling based on Voroni-plots were less 
precise than those developed at the plot level. 
Data 
All 
R 2 
Adj- R 2 
RMSE (m 2 /ha') 
Optical 
.0810 
.0493 
6.38 
SAR 
.0834 
.0403 
5.44 
Fused 
.1840 
.1371 
6.05 
4-6 yrs 
Optical 
.4191 
.3483 
3.05 
SAR 
.2497 
.1799 
3.73 
Fused 
.3324 
.2673 
5.71 
7-11 yrs 
Optical 
.5544 
.4974 
3.67 
SAR 
.4174 
.2939 
3.57 
Fused 
.5418 
.4552 
6.68 
Table 2. Voroni level Basal Area results 
4.3 Merchantable Volume: Plot level 
Plot level results for mvl multiple regression models are shown 
in table 3. Similar to ba when all plots are used to model mvl 
results are poor with the SAR data returning the highest R 2 and 
adjusted-R 2 . Results improved when the data set was subdivided 
into young and mature plots. Once again there appears to be a 
disparity between optical and SAR models with respect to age. 
In the young age group optical data return far superior models 
while in the mature age group the difference between the two 
was smaller. SAR data consistently return higher R 2 and 
adjusted R 2 values when modelling mvl in mature stands. 
Following from this the fused data sets also return improved 
models in mature stands, where nearly 50% enumerated mvl 
variance was explained compared to less than 10% and less than 
30% in the all age group and young group respectively. 
RMSE results shown in table 3 reflect the goodness of fit 
statistics mentioned above - an interesting result was that while 
the models developed with the mature data set return higher R 2 
and adjusted R 2 values, the RMSE results in the young data set 
were in some cases lower than those reported for the mature 
data sets. This indicated that while modelling plot level volume 
in the older stands produced superior models, they may not be 
that precise when compared to the younger stands. 
Data 
All 
R 2 
> 
¿Eh 
l 
5« 
RMSE (nf/ha' 1 ) 
Optical 
.1026 
.0730 
109.33 
SAR 
.2014 
.1298 
94.507 
Fused 
.0925 
.0503 
244.811 
4-6 yrs 
Optical 
.5679 
.4770 
33.84 
SAR 
.2729 
.1898 
44.075 
Fused 
.4376 
.3690 
38.246 
7-11 yrs 
Optical 
.4832 
.4454 
75.54 
SAR 
.4958 
.4093 
73.55 
Fused 
.4906 
* .4369 
76.42 
Table 3. Plot level Volume results 
4.4 Merchantable Volume: Voroni level 
Table 4 presents the results from the Voroni level mvl modelling. 
Analogous to results already presented in tables 1-3, when all 
plots are used to model inventory attributes results are poor. 
Goodness of fit statistics improved when the data were 
subdivided into young and mature plots. Once again optical 
models returned superior goodness of fit statistics in the 
younger age group when compared to the SAR results. SAR 
results did, however, improve in the mature age group but still 
remain inferior when compared to the optical data. Combining 
the optical and SAR data using the DWT-IHS transformation 
produced superior models in both the mature and all age groups. 
While results in the all age group explained less than 15% 
variance in enumerated mvl data, this value increased to just 
below 60% in the mature stands. Once again the precision of the 
mature stand models was called into question when observing 
the RMSE results. Younger plots seem to return more precise 
models regardless of the independent variables used. 
Data 
All 
R 2 
Adj- R 2 
RMSE (m 3 /ha J ) 
Optical 
.0702 
.0495 
101.07 
SAR 
.0867 
.0437 
99.10 
Fused 
.1498 
.1208 
103.26 
4-6 
yrs 
Optical 
.5970 
.5317 
29.83 
SAR 
.3250 
.2426 
44.09 
Fused 
.3025 
.2131 
40.20 
7-11 yrs 
Optical 
.4987 
.4309 
71.83 
SAR 
.4871 
.4040 
80.21 
Fused 
.5716 
.5167 
69.65 
Table 4. Voroni level Volume results 
5. DISCUSSION 
Both optical and SAR data returned poor results when 
compared to those in the published literature. Foody et al. (2001) 
used artificial neural networks and multiple independent 
variables to model above ground biomass, explaining 80% 
variance in field enumerated data. Zheng et al. (2004) used 
multiple regressions and achieved an R 2 of 0.67 for both pine 
and hardwood species. Lu (2005) found significant differences 
between mature and successional forests reporting R 2 values of 
0.50 and 0.76, respectively. The major difference between the 
present study and those cited above is that the present study 
occured in plantation forests, while research in the case of 
Foody et al. (2001), Zheng et al. (2004), and Lu (2005) were 
conducted in natural forests where forest canopies display 
significantly more spectral variability, associated with structural 
variability. In contrast, plantation forests do not display as much 
canopy spectral variability, thereby making it more difficult to 
use reflectance from these canopies to explain structural 
variability. The very same observation was evident when 
investigating the SAR results. 
Past studies have shown that saturation of the relationship 
between SAR backscatter is common with asymptotes usually 
determined by wavelength (Dobson et al., 1992; Rauste et al., 
1994; Imhoff, 1995; Ramsey, 1999; Fransson and Israelsson, 
1999) and to some extent the polarisation (Van de Griend and 
Seyhan, 1999; Santos et al., 2003). It proved impossible to