Mapping Wetlands and
Phragmites
Using
Publically Available Remotely Sensed Images
Yichun Xie, Anbing Zhang, and William Welsh
Abstract
Using publically available remotely sensed images to map
wetlands and invasive plants is attractive to ecologists,
environmental scientists, and managers. However, wetland
and invasive plant mapping on the basis of no- or low-cost
images has been challenged by the variability of mapping
accuracy. In this paper, we are developing an innovative
wetland and invasive plant mapping technique character-
ized with three integrations: the integration of image inter-
pretation with feature extraction, the integration of high
spatial-resolution images with high spectral-solution images,
and the integration of field reference data with interpreted
and classified images. This technique advocates standard
procedures for integrating
NAIP
(National Agriculture Imag-
ery Program) and Landsat images with multiple processes
of ground truthing, image classification, and validation.
The case study conducted in the Detroit River International
Wildlife Refuge concludes that the integration of
NAIP
and
Landsat images provides sufficient spatial and spectral
information for mapping coastal wetlands and Phragmites.
Introduction
Wetlands provide critical ecosystem services, such as support
of biodiversity, improvement of water quality, flood abate-
ment, and carbon sequestration (Mitsch and Gosselink, 2007).
Despite their value, wetlands have often been filled, drained,
or otherwise destroyed. Wetland losses in Michigan over the
past two centuries are estimated at over 50 percent (Dahl,
1990; Reyer
et al
., 2009). The wetlands that remain often face
severe stressors, especially in urban landscapes. These stress-
ors include altered hydrology, increased loads of nutrients and
contaminants from within the watershed, fragmentation, and
invasion of non-native species. The Great Lakes region has a
long history of biological invasions, with over 40 percent of
established exotic species being wetland plants (Mills
et al
.,
1993). At least 10 percent of invasive species in this region
have caused well-documented environmental problems and
substantial economic losses (Mills
et al
., 1993). One such
problematic invasive species currently spreading across the
Great Lakes region is the common reed,
Phragmites australis
.
Although native strains of this large (2 to 4 m) clonal reed are
endemic to North America, an aggressive non-native genotype
(Galatoswitsh
et al
., 1999) has resulted in its recent expansion
throughout the Great Lakes (Wilcox
et al
., 2003) and other
regions. Mapping of wetlands and
Phragmites
will provide
baseline data for monitoring
Phragmites
invasion and distri-
bution and for assessing the effects of
Phragmites
invasion and
Phragmites
removal efforts on wetland ecosystem function.
Remote sensing technology has proven to be a practical and
economical means to study land cover changes, for resource
monitoring, and resource assessment, especially over large
areas (Langley
et al
., 2001; Nordberg and Evertson, 2003; Niel-
sen
et al
., 2008). Recent advances in hyperspectral, microwave
(radar) and multispectral remote sensing provide powerful and
efficient techniques for monitoring plant activities at multiple
spatial and temporal scales (Xie
et al
., 2008). Hyperspectral
data, also known as imaging spectroscopy, is generally com-
posed of hundreds of spectral bands with narrow bandwidths
(5 to 10 nm), and can accurately detect the absorption features
of individual plant components (Varshney and Arora, 2004).
Hence, hyperspectral remote sensing is widely applied to
identify different species in plant communities. (Andrew and
Ustin, 2008; Asner and Martin, 2008; Pengra
et al
., 2007; Pig-
nattia
et al.
, 2009; Lopez
et al
., 2006; Ustin
et al
., 2002; Zhang
and Xie, 2014). Radar data have also been used to identify wet-
land plant communities (Corcoran
et al
., 2011; Henderson and
Lewis, 2008; Kasischke and Bourgeau-Chavez, 1997), includ-
ing
Phragmites
in the Great Lakes Basin (Bourgeau-Chavez
et
al
., 2004). Radar’s active sensors emit energy at a very low an-
gle and create backscattered energy. The backscattered energy
is sensitive to the dielectric constant and is primarily affected
by the volume, physical structure, and amount of moisture in
a material (Kozlov
et al
., 2001; Kwoun and Lu 2009; Skolnick,
2008). However, in general, the accessibility to hyperspectral
and radar data is limited and the associated costs of acquiring
and processing are very high.
Multispectral images are the largest family among remotely
sensed images. Many among them are free for public access.
For instance,
NAIP
imagery is publically available multispec-
tral data, acquired at a one-meter ground sample distance
with four bands of data: RGB and Near-infrared.
NAIP
imagery
has a high spatial resolution, but a coarse spectral resolution.
Another publically free accessible image dataset is the Landsat
imagery archive, which dates back to 1972. The Landsat imag-
es have been a major component of
NASA
᾿s Earth observation
program, with five primary sensors evolving over forty years:
MSS
(Multi-spectral Scanner),
TM
(Thematic Mapper), E
TM
+
(Enhanced Thematic Mapper Plus),
OLI
(Operational Land Im-
ager), and
TIRS
(Thermal Infrared Sensor)
.
Landsat-8 contains
11 bands and their spatial resolution is 30 m ground area, ex-
cept the panchromatic band 8 (15 m) and the thermal infrared
Yichun Xie is with the Institute for Geospatial Research & Ed-
ucation, Eastern Michigan University, 125 King Hall, Ypsilan-
ti, MI 48197 (
).
Anbing Zhang is with The School of Hydropower, Hebei Uni-
versity of Engineering, Handan 056021, China.
William Welsh is with the Department of Geography, Eastern
Michigan University, 203 Strong Hall, Ypsilanti, MI 48197.
Photogrammetric Engineering & Remote Sensing
Vol. 81, No. 1, January 2015, pp. 69–78.
0099-1112/15/811–69
© 2014 American Society for Photogrammetry
and Remote Sensing
doi: 10.14358/PERS.81.1.69
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
January 2015
69
