Examining Change Detection Approaches
for Tropical Mangrove Monitoring
Soe W. Myint, Janet Franklin, Michaela Buenemann, Won K. Kim, and Chandra P. Giri
Abstract
This study evaluated the effectiveness of different band
combinations and classifiers (unsupervised, supervised,
object-oriented nearest neighbor, and object-oriented de-
cision rule) for quantifying mangrove forest change using
multitemporal Landsat data. A discriminant analysis us-
ing spectra of different vegetation types determined that
bands 2 (0.52 to 0.6 µm), 5 (1.55 to 1.75 µm), and 7 (2.08
to 2.35 µm) were the most effective bands for differentiat-
ing mangrove forests from surrounding land cover types.
A ranking of thirty-six change maps, produced by compar-
ing the classification accuracy of twelve change detection
approaches, was used. The object-based Nearest Neighbor
classifier produced the highest mean overall accuracy (84
percent) regardless of band combinations. The automated
decision rule-based approach (mean overall accuracy of 88
percent) as well as a composite of bands 2, 5, and 7 used
with the unsupervised classifier and the same compos-
ite or all band difference with the object-oriented Nearest
Neighbor classifier were the most effective approaches.
Introduction
Mangrove forests, found in inter-tidal regions of the tropics
and subtropics, play an important role in stabilizing shore-
lines and reducing the impacts of extreme weather events and
strong wave action often associated with tropical storms (Teh
et al
., 2008; Yanagisawa
et al
., 2009). In addition, they are
among the most productive and biodiverse vegetated commu-
nities on Earth, providing important ecological and societal
goods and services including breeding and nursing grounds
for marine and pelagic species, and firewood and timber for
local communities (Aizpuru
et al
., 2000). Despite their impor-
tance, these forests are declining at an alarming rate: perhaps
even more rapidly than inland tropical forests (Aizpuru
et al
.,
2000). The worldwide extent of these forests is 137,760 square
kilometers which is 12.3 percent smaller than the most recent
estimate by the Food and Agriculture Organization (
FAO
) of
the United Nations (Giri
et al
., 2011a).
Furthermore, many remaining mangrove forests are degrad-
ed (UNEP, 2005) and under immense pressure from human
activities such as clear cutting, hydrological alterations, and
chemical spills (Blasco
et al
., 2001; McKee, 2005). This deg-
radation exacerbates the damage to coastal zones by extreme
weather events as the forests no longer serve as an effective
natural buffer against storm winds and waves (Weiner, 2005;
Williams, 2005; Rogan
et al
., 2011). Even in the absence of
direct human impacts, mangrove forests are at risk due to cli-
matic changes (Alongi, 2008; McKee, 2005). Giri
et al.
(2008)
reported that the tsunami-affected coastal areas of Indonesia,
Malaysia, Thailand, Burma (Myanmar), Bangladesh, India
and Sri Lanka in Asia lost 12 percent of its mangrove forests
between 1975 and 2005.
Past research has primarily focused on mapping mangrove
forests and their changes at a point in time as case studies
(Giri
et al
., 2011b; Bhattarai and Giri, 2011). In comparison,
little has been done to assess and monitor mangrove forest
change across larger and thus environmentally heterogeneous
areas (Long and Giri, 2011). To facilitate the appropriate
management, conservation, and restoration of these valuable
ecosystems given both current human and natural threats and
an uncertain future, timely and accurate maps of the broader
characteristics and spatial extent of mangrove forests as well
as their changes are needed.
Remote sensing using a variety of data and techniques has
played an important role in providing site- and time-specific
maps of mangrove ecosystems worldwide (Giri
et al
., 2011a).
For example, Chavaud
et al
. (1998) digitized benthic commu-
nity features using true-color aerial photography while Everitt
et al
. (2007) applied an unsupervised classification technique
to color-infrared aerial photography to discriminate black
mangrove populations in the south Texas Gulf Coast. There-
fore, it stands to reason that these data sets and techniques
will be equally useful for mapping mangrove forest changes.
In fact, the supervised approach for classifying remote
sensing images has proven effective in identifying mangrove
forests and change. For example, Gao (1999) employed the
widely used supervised maximum likelihood approach
applied to imagery of different spatial and spectral resolu-
tions in order to detect mangroves in the western Waitemata
Harbour, New Zealand and found that high spectral resolution
plays a more important role than fine spatial resolution in
identifying mangroves. Giri
et al
. (2011a) used a hybrid un-
supervised and supervised classification approach to perform
post-classification change analysis.
Soe W. Myint is with the School of Geographical Sciences and
Urban Planning, Arizona State University, P.O Box 875302,
Tempe, AZ85287 (
).
Janet Franklin is with the School of Geographical Scienc-
es and Urban Planning, Arizona State University, P.O. Box
875302, Tempe, AZ85287.
Michaela Buenemann is with the Department of Geography,
New Mexico State University, P.O. Box 30001, MSC MAP, Las
Cruces, NM 88003-8001.
Won K. Kim is with the Forest Economics and Management
Division, Korea Forest Research Institute, Seoul 132-712,
South Korea.
Chandra P. Giri is with the USGS Earth Resources Observation
and Science (EROS) Center, Sioux Falls, SD, 57198.
Photogrammetric Engineering & Remote Sensing
Vol. 80, No. 10, October 2014, pp. 983–993.
0099-1112/14/8010–983
© 2014 American Society for Photogrammetry
and Remote Sensing
doi: 10.14358/PERS.80.10.983
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
October 2014
983
