In: Wagner W., Szekely, B. (eds.): ISPRS TC VII Symposium - 100 Years ISPRS, Vienna, Austria, July 5-7, 2010, IAPRS, Vol. XXXVIII, Part 7B 
Figure 1. Study area and location of field sampling plots 
2. METHODOLOGY 
every part, and finally the DW of each tree was calculated by 
summing the DW of all parts. 
The methodology (Fig. 2) of this study comprises two parts 
namely allometric model development for field biomass 
estimation, and processing of AVNIR-2 and SPOT-5 images. 
Due to the lack of an allometric model for converting the trees 
measured in the field to actual biomass, it was necessary to 
harvest, dry and measure a representative sample of trees. Since 
tree species in Hong Kong are very diverse, the harvesting of a 
large sample was required. This was done by selecting the 
dominant tree species comprising a total of 75 trees in 4 DBH 
classes (less than 10, 10-15, 15-20 and 20 & above cm) and 
standard procedures were followed for tree harvesting 
(Ketterings et al, 2001; Overman et al, 1994). 
Figure 2. Overall methodology of this research 
The harvested trees were separated into fractions including 
leaves, twigs, small branches, large branches and stem. After 
measuring the fresh weight, representative samples from every 
part of the tree were taken for dry weight measurement in an 
oven at 80°C temperature until a constant dry weight was 
obtained (Fig 3). The weight of every sample was estimated 
using the same electric weight balance at 0.002gm precision. 
The ratio of dry weight (DW) to fresh weight (FW) was 
calculated for every part of the samples using DW and FW of 
each part of the tree. Using the ratio, DW was calculated for 
Figure 3. Preparing the samples for dry weight measurement 
Regression models using DW as the dependent variable, and 
DBH and height as independent variables were tested, and the 
best fit model (Table 2) was found to be InDW = a+b*In DBF!, 
with the adjusted coefficient of determination (adjusted r 2 
0.932) and an RMSE of 13.50. This was deemed highly 
satisfactory in view of the great variety of tree species, and is 
similar to the accuracies of several other specialist forest 
inventories (Brown et al, 1989; 1997; Overman et al, 1994). 
To build a relationship between image parameters and field 
biomass, 50 circular plots with a 15m radius covering a variety 
of tree stand types were selected using purposive sampling. The 
DBH of trees was measured at 1.3 m above ground and the 
heights of small and large trees were measured by Telescopic-5 
and DIST pro4 respectively. Using the measured parameter 
DBH, the biomass of each tree and biomass of all trees in a plot 
were estimated 
3. IMAGE PROCESSING 
The DN of the AVNIR-2 and SPOT data were converted to 
Spectral Radiance, and the images were orthorectified using the 
Satellite Orbital Math Model to obtain RMS error within 0.5 
pixel. All individual spectral bands of AVNIR-2 and SPOT-5 as 
well as different combinations of band ratios and PCA were 
tested for biomass estimation. All individual spectral bands of 
AVNIR-2 and SPOT-5 as well as different combinations of 
band ratios and PCA were tested for biomass estimation. 
Additionally, nineteen different types of texture measurements 
(Table 1) from GLCM based (Haralick, 1973) and SADH based 
(Unser, 1986) were used to generate texture parameters from 4 
spectral bands each of AVNIR-2 and SPOT data using 4 
window sizes (3x3 to 9x9). All the generated parameters were 
tested by comparison with the field biomass using stepwise and 
multiple regression models of single and dual-sensor data.