Spectroscopic Analysis of Green, Desiccated
and Dead Tamarisk Canopies
Ran Meng and Philip E. Dennison
Abstract
Defoliation by the northern tamarisk beetle (Diorhabda cari-
nulata) causes changes in the reflectance of tamarisk (Tama-
rix spp.) canopies. Cross correlogram spectral matching was
used to examine spectral separability of green, yellow desic-
cated, brown desiccated, and dead tamarisk canopy types.
Using a feature selection technique (the instability index),
four spectral regions were identified as important for canopy
type discrimination, including one red (645-693 nm), one near
infrared (735-946 nm), and two shortwave infrared regions
(1,960-2,090 nm and 2,400-2,478 nm). The random forests
decision tree algorithm was used to compare classification
performances of full-range and feature-selected hyperspectral
spectra as well as simulated WorldView-2 spectra. Classifica-
tion results indicated that the process of feature selection
can reduce data redundancy and computation time while
improving accuracy of tamarisk canopy type classification.
Introduction
Tamarisk (
Tamarix
spp., a.k.a.
saltcedar) is one of the most
widely dispersed invasive plant species in the western United
States, occupying an estimated 526,000 hectares and causing
ecosystem service-related economic losses ranging between
133 and 285 million US dollars annually (Zavaleta, 2000), not
to mention millions of dollars spent on eradication and resto-
ration projects (Hultine
et al
., 2010a). Previous studies have
reported that tamarisk has cumulative negative effects on
riparian ecosystems, such as reduced biodiversity, increased
soil surface salinity, changes in riparian wildfire occurrence,
and water use (Dudley
et al
., 2000; Shafroth
et al
., 2005).
In order to control tamarisk, the northern tamarisk beetle
(
Diorhabda carinulata
) has been released in the western
United States (Tracy and Robbins, 2009). The beetle removes
the leaf cuticle of tamarisk and eats the leaf mesophyll cells
selectively in both the larval and adult stages, leading the leaf
to desiccate and drop (Plate 1) (Meng
et al
., 2012; Nagler
et
al
., 2014). Defoliation may not kill the tamarisk plant, and in
many cases, tamarisk can refoliate in six to eight weeks after
defoliation; however, studies show that the repeat defoliation
caused by the tamarisk beetle can increase tamarisk mortal-
ity (Carruthers
et al
., 2008; Dudley and Bean, 2012; Nagler
et
al
., 2014). Repeat herbivory caused up to 40 percent tamarisk
mortality near the release sites after five years (Hultine
et al
.,
2010a).
Many studies have been implemented to investigate and
monitor the impacts of beetle attack on tamarisk populations
(Hudgeons
et al
., 2007; Nagler
et al
., 2008; Dennison
et al
.,
2009; Hultine
et al
., 2010a; Hultine
et al
., 2010b; Pattison
et
al
., 2011; Meng
et al
., 2012; Nagler
et al
., 2012; Snyder
et al
.,
2012; Nagler
et al
., 2014). Due to the unexpected dispersal
speed of the tamarisk beetle, the corresponding defoliation
has spread to an extensive area that is unrealistic to track and
analyze from the ground (Nagler
et al
., 2014). Projected cli-
mate warming and drying trends in the southwestern United
States may increase the over-winter survival of beetle popu-
lations, and consequently lead to increased herbivory (Dale
et al
., 2001; Raffa
et al
., 2008). Remote sensing techniques
may be the most effective way to evaluate the effectiveness
of tamarisk bio-control at the landscape scale (Dennison
et
al
., 2009; Meng
et al
., 2012; Nagler
et al
., 2012; Snyder
et
al
., 2012; Nagler
et al
., 2014). Nevertheless, previous remote
monitoring studies of tamarisk defoliation have not differenti-
ated between desiccated (live) tamarisk canopies and dead
tamarisk canopies at the stand scale (Dennison
et al
., 2009;
Meng
et al
., 2012; Nagler
et al
., 2012; Nagler
et al
., 2014).
Desiccated and dead tamarisk canopies will have very differ-
ent ecosystem impacts, since desiccated canopies will regrow
leaves and resume photosynthesis and transpiration. Identi-
fying the spectral differences between green, desiccated and
dead tamarisk canopies may help establish more informative
and accurate assessment of tamarisk bio-control impacts and
assist development of more adaptive management plans.
We hypothesized that spectral analysis techniques devel-
oped for hyperspectral processing can be used to study spec-
tral features of tamarisk canopies and spectral separability
among green, desiccated and dead tamarisk canopy spectra. If
accurate classification of tamarisk canopy types based on field
spectroscopy is proven feasible, hyperspectral and/or high
spatial resolution imagery may be useful for mapping tama-
risk bio-control impacts. The objectives of this study are to:
(a) develop a methodology for selecting suitable wavelengths
for discrimination of green, desiccated and dead tamarisk
canopies, and (b) analyze the spectral signatures of these
canopy types at both fine and coarse spectral resolutions.
Background
Previous studies of tamarisk defoliation by the northern tama-
risk beetle have used multispectral remote sensing data from
the Advanced Spaceborne Thermal Emission and Reflection
Radiometer (
ASTER
), Landsat Thematic Mapper (
TM
), Landsat
Enhanced Thematic Mapper+ (
ETM
+) and Moderate Resolu-
tion Imaging Spectroradiometer (
MODIS
) instruments. Denni-
son
et al
. (2009) mapped defoliation of large, dense tamarisk
stands on the Colorado and Dolores Rivers using both
ASTER
and
MODIS
imagery.
ASTER
normalized difference vegeta-
tion index (
NDVI
) and
MODIS
enhanced vegetation index (
EVI
)
both declined during periods of defoliation. Using Landsat
TM
imagery, Meng
et al
. (2012) compared two algorithms for
Department of Geography, University of Utah, 260 South Central
Campus Dr., Room 270, Salt Lake City, UT 84112
(
).
Photogrammetric Engineering & Remote Sensing
Vol. 81, No. 3, March 2015, pp. 199–207.
0099-1112/15/813–199
© 2014 American Society for Photogrammetry
and Remote Sensing
doi: 10.14358/PERS.81.3.199
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
March 2015
199
