extracted at coarser scales and paired with the corresponding
shadow cast by the “stand object.”
Conclusions
Saltcedar is best discriminated from other riparian species
during leaf senescence due to their unique spectral charac-
teristic. Nevertheless, methods are needed to classify these
species using images acquired during less favorable times due
to data availability. To date, this task has been challenging
because of the similar spectral responses between saltcedar
and cottonwood and the traditional pixel-based classification
methods using spectral information alone were unable to
separate these two species. Even when texture measures were
included in the pixel-based methods, the misclassification
rate remained high.
Cottonwood trees are either rare or in a state of decline
along many riparian corridors in both China and the US (Na-
gler
et al
., 2005; Silván-Cardenas and Wang, 2010; Cao
et al
.,
2012). Shape and shadow information associated with their
scattered distributions can be used to facilitate the discrimi-
nation between cottonwood and saltcedar. Our study found
that overall classification accuracy using a summer acquired
QuickBird image has increased by 15 percent compared to
Figure 7. Undetected shadows (in circles).
Figure 8. One of the two dense cottonwood stands near the seventh transect.
804
October 2015
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
