A Semiautomatic Extraction of Antarctic Lake
Features Using Worldview-2 Imagery
Shridhar D. Jawak and Alvarinho J. Luis
Abstract
We devised a semiautomatic approach for extracting lake fea-
tures based on a customized set of normalized difference wa-
ter index (
NDWI
) information which was obtained by incorpo-
rating high resolution, 8-band WorldView-2 data. An extensive
accuracy assessment was carried out for three semiautomatic
feature extraction approaches for extracting 36 lake features
on Larsemann Hills, Antarctica. The method was tested on
five existing
PAN
-sharpening algorithms, which suggest that
the customized
NDWI
approach renders intermediate perfor-
mance (root mean square error varies from ~227 to ~235 m
2
)
and highest stability when compared with existing feature ex-
traction techniques. In general, the customized
NDWI
rendered
a least misclassification (
≈
11 percent), followed by target
detection (
≈
16 percent) and spectral processing (
≈
17 percent)
methods for extraction of 36 lakes. We also found that cus-
tomized
NDWI
caused consistently least misclassification (
≈
21
percent) than the target detection (
≈
23 percent) and spectral
processing (
≈
30 percent) methods for extraction of partial-
ly snow or ice-covered 11 lakes. Our results indicate that
the use of the customized
NDWI
approach and appropriate
PAN
-sharpening algorithm can greatly improve the semiauto-
matic extraction of lake features in cryospheric environment.
Introduction
The semiautomatic (partly automatic or operating with
minimal human intervention) feature extraction method
interactively attempts to synergistically merge the intelligence
or knowledge of human visual system to robustly recognize
the targeted feature and the computer-aided system to carry
out rapid extraction of targeted feature and accurate shape
representation (Hu
et al
., 2004). In semiautomatic feature
extraction strategy, the target feature of interest is first recog-
nized by human vision and some approximations in terms of
seed points or training samples about the targeted feature are
often provided. The targeted feature is then delineated auto-
matically by the computer-aided algorithm.
Frazier and Page (2000) reviewed numerous methods em-
ployed by various authors to extract water bodies from Land-
sat Thematic Mapper (
TM
) and Multispectral Scanner (
MSS
).
It is apparent that the most common methods for extracting
water bodies use single band-based threshold methods, spec-
tral index ratio (
SIR
)-based multiband methods, image segmen-
tation methods, spectral-matching methods, and supervised
target detection methods.
Presently, the methods for extracting surface water bodies
are based on spectral index or multiband techniques, which
are spectrum property-based methods, such as the normal-
ized difference vegetation index (
NDVI
) and the normalized
difference water index (
NDWI
) (McFeeters, 1996; McFeeters,
2013). Because a single spectral index cannot delineate
water bodies effectively in different environments, many
improved indices have been proposed to obtain better results
in a particular environment (Lacaux
et al
., 2007; Xu, 2005).
Ouma and Tateishi (2006) proposed a novel water extraction
index for shoreline delineation by combining the tasseled cap
wetness index (
TCWI
) and
NDWI
. A comprehensive water body
information extraction technique was proposed by Wu
et al
.
(2008) through fusion of the spectral relationships between
various bands with supervised classification methods. Rogers
and Kearney (2004) proposed the
NDWI
for Moderate Resolu-
tion Imaging Spectroradiometer (
MODIS
) multispectral satellite
images (
MSI
). Furthermore, Xiao
et al
. (2005) proposed a land
surface water index (LSWI) by combining the shortwave infra-
red (
SWIR
) and near-infrared (
NIR
) data to identify water bodies
in
MODIS
images. Lu
et al
. (2011) recommended an integrated
water body extraction technique with HJ-1A/B satellite im-
agery by utilizing differences between
NDVI
and
NDWI
. These
modified indices have been commonly used to map surface
water bodies using Landsat and
MODIS
images (Zhong
et al
.,
2008; Li
et al
., 2009; Soti
et al
., 2009). However, because of
the complexity of cryospheric environment, various surface
targets may have the same spectrum characteristics. There-
fore, only one type of index method can extract water bodies
under certain conditions only (Jawak and Luis, 2013a and
2013b). Additionally, these spectral indices were developed
for well-known visible near-IR (
VNIR
) and shortwave infrared
(
SWIR
) systems. The WorldView-2 (WV-2) provides a duplet
of
VNIR
bands, which offers an opportunity to adapt a novel
set of
NDWI
using eight bands. The present study differs from
previous studies by reconstructing the
NDWI
for 8-band WV-2
data in order to semi-automatically extract lake features.
Spectral processing-based or matching-based feature iden-
tification methods, such as Matched Filtering (
MF
), Mixture
Tuned Matched Filtering (
MTMF
), Spectral Angle Mapper
(
SAM
),
MF/SAM
ratio, and Principal Component Analysis (
PCA
),
have been implemented for improved mapping and classifica-
tion of remote sensing images (Peterson
et al
., 2011; Williams
and Hunt, 2002; Boardman, 1998; Kruse
et al
., 1993; Jia and
Richards, 1999). Supervised target detection methods, such as
Constrained Energy Minimization (
CEM
) (Chang
et al
., 2000),
Adaptive Coherence Estimator (
ACE
) (Bourennane
et al
., 2011),
Target-Constrained Interference-Minimized Filter (
TCIMF
) (Ren
and Chang, 2010), Mixture Tuned
TCIMF
Polar Remote Sensing Department, National Centre for Ant-
arctic and Ocean Research (NCAOR), Earth System Sciences
Organization (ESSO), Ministry of Earth Sciences (MoES),
Government of India, Headland Sada, Goa 403804, India
(
).
Photogrammetric Engineering & Remote Sensing
Vol. 80, No. 10, October 2014, pp. 939–952.
0099-1112/14/8010–939
© 2014 American Society for Photogrammetry
and Remote Sensing
doi: 10.14358/PERS.80.10.939
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
September 2014
939
