1090 
The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Voi. XXXVII. Part B7. Beijing 2008 
incorporates the watershed segmentation algorithm and 
subsequent classification using tree heights and tree crown 
signatures derived from both lidar data and multispectral 
imagery. 
2. STUDY AREA AND DATASETS 
The study area is part of the Moira State Forest, New South 
Wales, Australia where the dominant tree species are native 
eucalypts. River Red Gum (Eucalyptus camaldulensis ssp. 
obtusa Dehnh), Black Box (Eucalyptus largiflorens), and Grey 
Box (Eucalyptus microcarpa) are common tree species found in 
the Moira State Forest. 
2.1 Lidar data 
The lidar data used for this project was acquired by 
AAMGeoScan (now AAMHatch) in May,2001. The lidar 
system used was the ALTM 1225, which operates with a 
sampling intensity of 11000 Hz at a wavelength of 1.047 pm. 
The approximate flying height of this sensor was 1100m and 
the laser swath width was 800m. Vertical accuracy was 0.15m 
(la), the internal precision was 0.05m, and the original laser 
footprint was 22cm in diameter. The original lidar dataset was 
processed by AAMHatch and provided to the Victorian 
Department of Sustainability and Environment (DSE). The 
provided data were two separate files representing the first and 
last return point clouds. The original lidar data had point 
spacing in the order of 16 points per m 2 and was resampled to a 
lm grid. 
2.2 Multi-spectral imagery 
The multi-spectral imagery was captured over the study area 
using an Ultracam-D with a calibrated focal length of 
101.400mm. Three colour (red, green and blue) and one 
infrared (IR) band images were collected with a 28.125pm pixel 
size. The radiometric resolution of the images was 16-bit. This 
increased radiometric range captures more detailed information 
of the land cover features. As a result, in extreme bright and 
dark areas we still mange to get redundant information, which is 
beyond what is visible in images with lower radiometric 
resolution (Leberl and Gruber, 2005). 3 
3. METHODOLOGY 
The proposed scheme includes five parts: (1) data pre 
processing, (2) watershed segmentation, (3) data processing, (4) 
supervised classification and (5) accuracy assessment. The 
flowchart in Figure 1, illustrates the major steps, which are 
performed through this data fusion project. 
3.1 Data pre-processing 
The data pre-processing stage consists of two major steps: (1) 
normalised digital surface model (nDSM) generation from lidar 
data and (2) geometric correction of multispectral imagery. The 
lidar first and last return height data were used to generate the 
nDSM for the tree crowns. The last return of the lidar normally 
represents the digital terrain model (DTM) and the first return 
as the digital surface model (DSM). A height difference 
between the DSM and DTM represents the absolute height of 
the trees. A height threshold was applied to remove low-lying 
vegetation (<1.5m) close to the terrain surface. This nDSM 
along with lidar 1 st return intensity and multispectral images 
was used in the data fusion process. 
Using the exterior orientation parameters (X, Y, Z, co, cp, tc) 
derived from onboard GPS and INS sensors and ground control 
points the geometric correction of the multispectral imagery 
was accomplished. The exterior orientation parameters for each 
aerial photograph were supplied with the camera calibration 
certificate, which were used for the orthorectification of the 
aerial photographs. An optimal number of ground control points 
were derived using differential GPS to increase the geometric 
corrections of the multispectral imagery. 
Figure 1. Flowchart of tree type classification using lidar and 
multispectral imagery. 
Some temporal effects were expected in the data fusion process, 
due to the differences in acquisition time of the lidar (in 2001)