The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B7. Beijing 2008 
undulation could dramatically bias the reflectances of targets 
and further seriously attenuate the unmixing result for LSMM, 
and on the other hand, it also validate the effectiveness of C 
correction method on removing terrain undulation effect again. 
(at) 
FI = 1. 3138HDVI + 0. 048 
E? = 0.7774 
l 
* o.a 
0.6 
0.0 
o.z 
Jim 
0.2 0.4 0.6 0.3 
-0.4 
HDTI 
Cbl) 
Ct2) 
Figure 5 (al),(a2) Relationships between vegetation abundances and NDVI for pre-C corrected and post-C corrected ASTER 
imagery. (bl),(b2) Relationships between vegetation abundances and MVI for pre-C corrected and post-C corrected ASTER imagery. 
5. SUMMARIES AND CONCLUSIONS 
This paper has investigated the impact of terrain undulation on 
LSMM with reference to unmixing vegetation abundances using 
LSMM. C correction was used to remove or minimize terrain 
effects of the original ASTER data and the result showed that 
the C method was reasonable and effective. The endmember 
selection procedures such as minimum noise fraction (MNF), 
pixel purity index(PPI) and n-dimensional visualization were 
implemented respectively to pre-corrected and post-corrected 
ASTER data to determine the endmembers effectively. A full 
constrained least square LSMM was applied to the two data sets 
and the vegetation abundance images were sequentially 
derived .Multi-regression analysis between vegetation 
abundance and vegetation indexes which was employed to 
validate and estimate the terrain undulation impact on LSMM 
indicated that terrain undulation could constrain the application 
of LSMM, typically the unmixing precision was improved 
17.6% and 18.6% in R2 value for NDVI and MVI by 
minimizing or even removing terrain undulation effect using C 
correction method in our study. So specially effective removal 
or minimization of terrain impact was essential for LSMM 
applications in moderate or small-scale mountainous areas. The 
results not only proved the terrain undulation could dramatically 
bias the reflectances of targets and further seriously attenuate 
the unmixing result for LSMM but also validate the 
effectiveness of C correction method on removing terrain 
undulation impact again. 
Though we took vegetation abundance as a case study, it should 
be envisioned that the similar result for other endmember types 
(water, barren soil, impervious area and so on )could be achieve 
because of the same impact mechanism of terrain undulation 
and the identical unmixing procedure with LSMM. However 
further studies of different area different types of imagery and 
other endmembers are recommended with purpose to inspect the 
validation and applicability of our results and conclusions. In 
addition, to acquire specific unmixing precision the accurate 
and quantitative ground data should be collected. In fact, the 
atmospheric scattering and scales of the imagery can also 
behave negative impact on LSMM, to quantitatively evaluate 
the impacts is our further work in the future. 
ACKNOWLEDGEMENTS 
the authors like to extend our appreciations to College of 
Geographical Sciences ,Fujian Normal University, Fujian 
provincial department of science & technology for their 
foundation support. We also want to gratefully acknowledge all 
the anonymous reviewer for reviewing the manuscript. 
REFERENCES 
Adams, J. B., Sabol, D. E., Kapos, V., Almeida, R., Roberts, D. 
A., Smith, M.O., et al. (1995). Classification of multispectral 
images based on fractions of endmembers—Application to 
62