Examining the Effectiveness of Spectrally
Transformed SMA in Urban Environments
Yingbin Deng and Changshan Wu
Abstract
Spectral transformation has been applied to address the
spectral variability in spectral mixture analysis. However,
there is not a study addressing the necessity and applicabil-
ity of transformed models. This article, therefore, aims to
answer two questions: whether significantly different results
will be generated through applying a spectral transforma-
tion, and which spectral transformation performs better
in urban environments. In particular, 26 spectrally trans-
formed schemes were examined in three cities. Results of
paired-sample t tests demonstrated that normalized spectral
mixture analysis performed significantly better than the
untransformed scheme in all three study areas. Derivative
analysis, independent component analysis, and minimum
noise fraction outperformed the untransformed scheme
in one or two study areas but underperformed in others.
Other schemes are unnecessary, as they have significantly
lower accuracy compared to the untransformed scheme.
Introduction
Spectral mixture analysis (
SMA
) has been widely applied to
estimate fractional land covers within each pixel of coarse-res-
olution remote sensing imagery. Spectral transformation, which
converts the original spectra linearly or nonlinearly, is one of
the widely applied approaches in
SMA
to address spectral-vari-
ability problems. It focuses on enhancing the spectral character-
istics to reduce within-class variability and
classes variability. Spectral transformation h
in a large number of remote sensing applica
proposed a normalized
SMA
(
NSMA
) to extra
cover information in the city of Columbus, Ohio, United States.
Results indicated that brightness variations were reduced and
improved results could be achieved from the
NSMA
. Asner
and Lobell (2000) applied a tie spectral transformation before
estimating the fractions of vegetation and soil covers. They
concluded that significant variations of soil moisture, canopy
architecture, leaf and litter area index, and tissue optics could
be compressed using the tied spectra. Zhang, Rivard, and San-
chez-Azofeifa (2004) summarized that second-order derivative
spectral unmixing is promising for fraction estimation using
hyperspectral data. Similar studies include the research of Tsai
and Philpot (1998), Laba, Tsai, Ogurcak, Smith, and Richmond
(2005), and Huguenin and Jones (1986). Further, Debba, Car-
ranza, van der Meer, and Stein (2006) concluded that higher-or-
der derivatives contribute more to remote sensing imagery with
higher signal-to-noise ratios. Li (2004) compared the discrete
wavelet transform (
DWT
) with principal components analysis
(
PCA
; Richards and Richards 1999; Chuvieco and Huete 2009)
and found that
DWT
could bring more separability than
PCA
,
leading to improvement of fractional estimations. Similar con-
clusions were achieved in the studies of Bruce, Koger, and Li
(2002) and Zhang, Rivard, Sánchez-Azofeifa, and Castro-Esau
(2006). Youngentob
et al.
(2011) examined the performance of
continuum-removal (
CR
) analysis using hyperspectral data, and
the results showed improvement in overall accuracy. Further,
PCA
(Pearson 1901; Byrne, Crapper and Mayo 1980; Richards
and Richards 1999), the minimum noise fraction (
MNF
) trans-
form (Green, Berman, Switzer and Craig 1988; Geladi, Isaks-
son, Lindqvist, Wold and Esbensen 1989; Boardman and Kruse
1994), tasselled cap (
TC
; Kauth and Thomas 1976; Jensen and
Lulla 1987), and independent component analysis (
ICA
; Bayliss,
Gualtieri and Cromp 1998; Chen and Zhang 1999; Hyvärinen
and Oja 2000) are commonly employed for the land surface
feature enhancement before applying
SMA
. Spectral character-
istics of different land cover classes are highlighted in different
output layers of these transformations. Many researchers have
used these transformation techniques to assist in the selection
of end members, reduce spectral within-class variability, and
enhance between-classes variability. Further, different spatial
filters, such as low-pass (
LP
; Green
et al.
1988), high-pass (
HP
;
Yu
et al.
2006), Gaussian high-pass (
GHP
; Schowengerdt 2006),
and Gaussian low-pass (
GLP
; Schowengerdt 2006), are com-
nhance the spectral characteristics’ edges or to
sensing imagery (Xu
et al.
2011).
ny researchers have applied spectral-trans-
iques in remote sensing applications, there
is still a lack of comprehensive and systematic studies to
examine their effectiveness. In particular, researchers use
different transformed schemes based on their own knowl-
edge and expertise. The necessity of applying a transformed
scheme has not been adequately discussed in the literature.
In addition, because of the existence of spectral variability,
the reliability of each transformed scheme is still unclear, and
most researchers apply the transformed schemes in only one
study area. It is unknown whether consistent results can be
obtained other study areas. Therefore, this study aims to ex-
amine whether there is a significant difference in
SMA
results
after applying spectral transformation and to find out which
spectral-transformation approach generates consistently better
results. The structure of this article is as follows: The next
section introduces the background of
SMA
and spectral trans-
formation techniques, followed by the experiments and the
results in three study areas with Landsat Thematic Mapper
data. The next section discusses the results and is followed by
the conclusions of this study.
Yingbin Deng is with the Guangdong Open Laboratory of
Geospatial Information Technology and Application, Lab of
Guangdong for Utilization of Remote Sensing and Geographical
Information System, Guangzhou Institute of Geography,
Guangzhou, China; and the Department of Geography,
University of Wisconsin-Milwaukee, Milwaukee, WI 53211.
Changshan Wu is with the School of Geology and
Geomatics, Tianjin Chengjian University, Tianjin, China;
and the Department of Geography, University of Wisconsin-
Milwaukee, Milwaukee, WI 53211 (
.
Photogrammetric Engineering & Remote Sensing
Vol. 85, No. 7, July 2019, pp. 521–528.
0099-1112/19/521–528
© 2019 American Society for Photogrammetry
and Remote Sensing
doi: 10.14358/PERS.85.7.521
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
July 2019
521
