The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B7. Beijing 2008 
1093 
The classification accuracies achieved are comparable to those 
produced by visual interpretation. Map scatter plots of lidar 
return combined with multispectral imagery and field data 
enabled, in some cases, visual discrimination at the individual 
tree level between Black Box and Grey Box. While a clear 
distinction between these two species was not always visually 
obvious at the individual tree level, due to other extraneous 
sources of variation in the dataset, the observation was 
supported in general at the site level. Sites dominated by Black 
Box generally exhibited a lower proportion of singular lidar 
returns compared to sites dominated by Grey Box. River Red 
Gums can easily be distinguished from others by their unique 
spatial distribution. This species is largely populated throughout 
the forest surrounding the Riverine wetlands that are subject to 
periodic inundation (see Appendix A). 
The fusion procedure proposed in this study demonstrates the 
usefulness of the five main processing steps to cope with the 
classification of very hight spatial resolution lidar and 
multispectral imagery. This approach can be used in principle 
for species classification of high spatial resolution data. 
However, sensor-specific modifications to these different 
processing steps have to be made in order to maximise the 
fusion results. 
5. CONCLUSIONS 
The investigation presented in this paper has been conducted to 
establish an automated procedure for forest species 
identification at the tree level from high spatial resolution 
imagery and lidar data. For this purpose, four-band 
multispectral imagery and lidar data were used to develop a 
feature-level fusion approach. This technique consists of five 
steps: the preprocessing of the lidar and multispectral data; tree 
crown polygon extraction using marker-control watershed 
segmentation; masking of spectral, height and textural 
information using the crown polygons; classification of the 
polygon data; and, the accuracy assessment. 
In contrast to the original four-band multispectral data sets, the 
average classification accuracy was considerably improved 
through the generation of additional features using the principal 
component transformation, filtering techniques, and texture 
analysis. Principal component transformation of the 
multispectral imagery added more layers to separate different 
tree species. The addition of the height and texture features 
derived from lidar data resulted in an improved discrimination 
of the tree classes. 
The superposition of the watershed derived crown polygon on 
to the images was essential for achieving good classification 
results and for a more standardised classification. The proposed 
procedure can be used as a model for fusing high spatial 
resolution multispectral imagery and lidar data for assessing 
forest attributes at the tree level. In addition, this fusion 
procedure has the potential to minimise human interaction in 
the interpretation of forest attributes. 
ACKNOWLEDGEMENT 
This work was supported by the Australian Research Council 
(ARC) under Discovery Project DP0450889. The Ultracam-D 
data set was provided by IFMS Germany 
(http: //www. arc forest. com/). 
REFERENCES 
Baltsavias, E.P.(1999) A comparison between photogrammetry 
and laser scanning, ISPRS Journal of Photogrammetry and 
Remote Sensing, Vol. 54, p. 83-94. 
Beucher, S. and Lantuejoul, C.(1979) Use of watersheds in 
contour detection, Proceedings of International Workshop on 
Image Processing, Real-time Edge and Motion 
Detection/Estimation, Rennes, France. 
Chen, Q., Balddocchi, D., Gong, P. and Maggi, K.(2006) 
Isolating individual trees in a Savnna Woodland using small 
footprint lidar data, Photogrammetric Engineering & Remote 
Sensing, Vol. 72, No. 8, pp. 923-932. 
Clark, D.B., Castro, C.S., Alvarado, L.D.A. and Read, 
J.M.(2004) Quantifying mortality of tropical rain forest trees 
using high-spatial-resolution satellite data, Ecological Letters, 
Vol. 7, pp. 52-59. 
Gong, P., Biging, G.S., Lee, S.M., Mei, X., Sheng, Y., Pu, R., 
Xu, B., Schwarz, K. and Mostafa, M.(1999) Photo ecometrics 
for forest inventory, Geographic Information Science, Vol. 5, 
pp. 9-14. 
Kalayeh, H.W. and Landgrebe, D.A.(1983) Predicting the 
required number of training samples, IEEE Transactions on 
Pattern Analysis and Machine Intelligence, Vol. 5, pp. 493-504. 
Kelly, M., Shaari, D., Guo, Q.H. and Liu, D.S(2004) A 
comparison of standard and hybrid classifier methods for 
mapping hardwood mortality in areas affected by "sudden oak 
death", Photogrammetric Engineering & Remote Sensing, Vol. 
70, pp. 122-1239. 
Leberl, F. and Gruber, M.(2005) Ultracam-d: understanding 
some noteworthy capabilities, Proceedings of Photogrammetric 
Week 2005, Stuttgart, Germany. 
Leckie, D., Gougeon, F., Hill, D., Quinn, R., Armstrong, L. and 
Shreenan, R.(2003) Combinedhigh density LiDAR and 
multispectral imagery for individual tree crown analysis, 
Canadian Journal of Remote Sensing, Vol. 29, No. 5, p. 633- 
649. 
Meyer, F. and Beucher, S.(1990) Morphological segmentation, 
Journal of Visual Communication and Image Representation, 
Vol. 1, No. l,pp. 21-46. 
Soille, P.(2003) Morphological image analysis: principles and 
applications (2nd Edition), Springer, Berlin ; New York. 316 pp. 
Sua'reza, J.C., Ontiveros, C., Smith, S. and Snape, S.(2005) 
Use of airborne LiDAR andaerial photography in the estimation 
of individual tree heights in forestry, Computers & Geosciences 
(2005), Vol. 31, No. 2, p. 253-262.