PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
September 2014
885
WorldView-2 High Spatial Resolution Improves
Desert Invasive Plant Detection
Temuulen Sankey, Brett Dickson, Steve Sesnie, Ophelia Wang, Aaron Olsson, and Luke Zachmann
Abstract
Sahara mustard (Brassica tournefortii) is an invasive spe-
cies common to the Mojave and Sonoran Deserts in the
southwestern US. Our objective was to assess WorldView-2
(
WV
2) satellite imagery potential to detect Sahara mustard
presence, cover, and biomass. We compared
WV
2 images
(2.4 m and 30 m resolution) to Landsat
ETM
+ image both
classified using a mixture tuned matched filtering (
MTMF
).
A total of 1,885 field plots (30 × 30 m) were established
across a 8,715 km
2
study area in spring of 2012, an excep-
tionally dry year. Average target canopy cover (7.5 percent)
and biomass (0.82 g/m
2
) were extremely low. The
WV
2
MTMF
classification had a much greater overall accuracy of 88
percent, while the resampled
WV
2 and the Landsat
ETM
+
MTMF
classification overall accuracies were 67 percent and
59 percent, respectively. Producer’s and user’s accuracies in
target detection were 86 percent and 94 percent, respectively,
although the exceptionally low canopy cover and biomass
were not well correlated with image-based estimates.
Introduction
Non-native plant invasions threaten to alter the structure and
function of ecosystems globally (D’Antonio and Vitousek,
1992). Highly invasive plant species can dramatically alter
hydrologic and nutrient cycles, fire regimes, and other ecolog-
ical processes. Invasive plants cost the western US economy
as much as 34 billion dollars per year (Barnett
et al
., 2007).
An efficient method to detect and determine the distribution
and abundance of non-native invasive plants over large areas
is critically needed, particularly for species undergoing rapid
expansion in fragile arid ecosystems.
Remote sensing provides a promising tool for targeted
monitoring or eradication by land management agencies
(Lass
et al
., 2005; Bradley and Mustard, 2006; Noujdina and
Ustin, 2008). Invasive species studies have successfully used
hyperspectral data such as
AVIRIS
imagery with 20 m resolu-
tion and 224 bands and HyMap imagery with 3.5 m resolution
and 126 bands (O’Neill
et al
., 2000; Root
et al
., 2002; Dudek
et
al
. 2004; Parker Williams and Hunt, 2002, 2004; Glenn
et al
.,
2005; Noujdina and Ustin, 2008). However, hyperspectral im-
agery can be expensive to acquire and tends to cover relative-
ly small spatial extents. Using freely available, moderate reso-
lution imagery such as Landsat can reduce costs and provide
data at a temporal resolution suitable for monitoring changes
in plant distributions, especially with the recent launch of the
Landsat-8 satellite. However, moderate to coarse resolution
data typically provide low rates of invasive plant detection
because of the mixed cover types within each pixel, especially
at early stages of invasion when invasive plant populations
are small and sparsely distributed (Lass
et al
., 2005; Mitchell
and Glenn, 2009).
Federal agencies in the US now have access to commer-
cial satellite data such as high resolution WorldView-2 (
WV
2)
imagery using the US Geological Survey Commercial Remote
Sensing Space Policy (CRSSP, 2003). Methods to effectively
utilize imagery for early detection of invasive species are
needed to add value to these data sources as other multispec-
tral and high resolution commercial satellite data become
more readily available (Kruse and Perry, 2013). The availabil-
ity of high resolution data, combined with efficient classifi-
cation methods for invasive species, can potentially improve
early detection rates thereby enhancing invasive species
management and mitigation efforts.
The
WV
2 satellite remote sensing system is a relatively new,
high spatial resolution (2.4 m pixels) sensor that is the first of
its kind to produce 8-band multispectral imagery (Figure 1)
(Kruse and Perry, 2013).
WV
2 might provide a unique oppor-
tunity to detect small populations of desert plants due to its
high spatial and spectral resolution and bands in the red (630
to 690 nm), red edge (705 to 745 nm), and near-infrared (770 to
895 nm and 860 to 1040 nm) spectral regions (Figure 1). The
fine spatial resolution of
WV
2 imagery has been demonstrated
to improve classification accuracy in forested environments,
where overall accuracies reached 98 percent (Ozdemir and Kar-
neli, 2011; Garrity
et al.,
2012). Latif
et al.
(2012) and Immitzer
et al.
(2012) further document that the high spectral resolution
of
WV
2 imagery result in successful tree species differentiation
(overall accuracy of 82 percent), although producer’s accuracies
at the species-level ranged widely between 33 percent and 92
percent (Immitzer
et al
., 2012).
WV
2 data have also been shown
to enhance classification accuracy for tree species differenti-
ation in a savanna ecosystem (Cho
et al.
, 2012), cover types
in urban areas (Zhang and Kerekes, 2012; Longbotham
et al
.,
2012; Pu and Landry, 2012), and coral reef detection in marine
environments (Botha
et al
., 2013). The utility of
WV
2 imagery
for mapping invasive plants has not been fully explored in hot
desert environments where invasive species can exhibit large
interannual variability in distribution and abundance.
Temuulen Sankey is with the School of Earth Sciences and
Environmental Sustainability, Northern Arizona University,
1298 S. Knoles Drive, Flagstaff AZ 86011. (Temuulen.San-
).
Brett Dickson, Ophelia Wang, Aaron Olsson, and Luke Zach-
mann are with the Lab of Landscape Ecology and Conserva-
tion Biology, Landscape Conservation Initiative, Northern
Arizona University, Box 5694, Flagstaff, AZ 86011, and Con-
servation Science Partners, Inc., 11050 Pioneer Trail, Suite
202, Truckee, CA 96161.
Steve Sesnie is with the US Fish and Wildlife Service, Albu-
querque, NM 87102.
Photogrammetric Engineering & Remote Sensing
Vol. 80, No. 9, September 2014, pp. 885–893.
0099-1112/14/8009–885
© 2014 American Society for Photogrammetry
and Remote Sensing
doi: 10.14358/PERS.80.9.885
