PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
January 2017
27
Object-based Wetland Characterization Using
Radarsat-2 Quad-Polarimetric SAR Data,
Landsat-8 OLI Imagery, and Airborne Lidar-
Derived Geomorphometric Variables
Steven E. Franklin and Oumer S. Ahmed
Abstract
The goal of this research was to classify four wetland types in the Hudson Bay Lowlands in northern Canada using Radarsat-2
quad-polarization and Landsat-8 satellite sensor data and geomorphometric variables extracted from an airborne lidar digital
elevation model. Segmentation was followed by object-based image classification implemented with a Random Forest machine
learning algorithm. The classification accuracy was determined to be approximately 91 percent. This is a significant improve-
ment over the accuracy that was obtained using the Radarsat-2 (80 percent) or Landsat-8 sensor data alone (84 percent).
Variable importance (VI) was measured for geomorphometric measures related to the gravity-, wind- and solar-fields, which
were developed to explain eco-hydrological differences and increase the separability of wetland classes. Further research will
consider additional geomorphometric and spectral response variables that are useful in more detailed boreal wetland classifi-
cations and analysis of wetland characteristics over time.
Introduction
Wetland classifications are typically based on taxonomic, structural and functional differences among different vegetation com-
munities and hydrological regimes (Keddy, 2010; Zoltai and Vitt, 1995; National Wetlands Working Group, 1997; Hogg
et al.,
2009). In the field, identifying and classifying such wetlands can be difficult, especially as additional variability is introduced
by climate (e.g., drying) and environmental change (e.g., abiotic, biotic, or cultural modification of hydrological regimes). In
such cases, careful analysis of descriptive measures derived from field observations, remote sensing imagery, soils, topographic
and hydrological data may be required to classify adequately the different wetland types of interest at the level of detail and
scale desired (Tiner
et al
., 2015).
Early national and regional wetland mapping programs and protocols specifically outlined initial criteria for wetland classi-
fication (i.e., water regime, permanence, depth, chemistry, and vegetation cover) (e.g., Stewart and Kantrud, 1971; Cowardin
et
al.
, 1979; Carter, 1982; Lyon
et al.,
2001). Subsequent remote sensing research efforts and operational wetland mapping pro-
grams have documented increasingly effective and accurate methods and results for a wide range of satellite and aerial remote
sensing data (e.g., Dechka
et al.
, 2002; Murphy
et al.
, 2007; Peckham
et al.
, 2009; Bwangoy
et al.
, 2013; Gessner
et al.
, 2015;
Dingle Robertson
et al.
, 2015). Improved wetland classification has been obtained through image fusion (Haack and Bechdol,
2000; Grenier
et al.
, 2007; Huang and Jing, 2007), object-based segmentation (Frohn
et al
., 2009; Powers
et al
., 2012; Dronova
et al
., 2015), machine learning classification (Martin
et al
., 2012; Millard and Richardson, 2012; van Beijma
et al
., 2014), and
the use of digital elevation model (
DEM
) data (Goodale
et al
., 2007; Difebo
et al
., 2015). Precise and accurate
DEM
data can be
acquired from topographic maps, aerial- and satellite-based interferometry, airborne lidar, stereoscopic
DEM
generation, and
photogrammetry (Hengl and Reuter, 2009)
The quantitative description and measurement of the land surface represented in a
DEM
was introduced by Evans (1972) as a
system of geomorphometry. General geomorphometry refers to local and regional derivatives of the gravity field: (a) elevation,
(b) the first derivative of elevation (slope), (c) aspect (a directional measure of slope), (d) selected second and third derivatives
(or curvatures), and (e) certain measures (e.g., first and second moments) of their distributions (Florinsky, 2009; Minár
et al
.,
2013a). General geomorphometric variables are field-specific and field-invariant, and can be modeled with reference to the
behavior of fundamental geometrical properties of surfaces (Mitasova and Hofierka, 1993; Minár
et al
., 2013b). Specific geomor-
phometry refers to the large and still-expanding range of variables that are used to describe terrain features or objects (e.g., land-
forms, linear features, drainage networks, and terrain “objects”, including wetlands) (Hengl and Reuter, 2009). Many of these
specific geomorphometrics are defined relative to the gravity field. Others are area-based or positional (e.g., distance to stream),
or are used to describe phenomena in the solar radiation field (e.g., angle of incidence) and wind field (e.g., exposure or “topo-
graphic openness”). The enhancement and development of geomorphometry as a subfield of geomorphology (Pike, 1996 and
2000), hydrology (Moore
et al
., 1991), and mathematical geology (Shary, 1995) has been described by Zevenberger and Thorne
(1987), Papo and Gelbman (1984), Guth (1995), Shary
et al
. (2002), Deng (2007), and Hengl and Reuter (2009), among others.
No formal system of wetland geomorphometrics exists. However, Minár
et al
. (2013a) have outlined a scheme of more than
ten classes of geomorphometric variables, many of which are
known or are expected
a priori
to be helpful in wetland interpre-
tation and classification. Temporal and spatial scales are a criti-
cal consideration (Holling, 1992; Stallins, 2006). Derivatives may
School of the Environment, Trent University, Ontario, K9J
7B8, Canada (
).
Photogrammetric Engineering & Remote Sensing
Vol. 83, No. 1, January 2017, pp. 27–36.
0099-1112/17/27–36
© 2017 American Society for Photogrammetry
and Remote Sensing
