Airborne Hyperspectral Research

1
Airborne Hyperspectral Research Development
for Invasive Species Detection and Mapping
Kenneth McGwire1, Timothy Minor1, Bradley
Schultz2, and Christopher Kratt1 1Desert Research
Institute, Reno, Nevada, 89512 USA 2University of
Nevada Cooperative Extension, Winnemucca, NV
89445 USA
Field Data Collection
Introduction
Accuracy assessments will be performed for both
discrete and continuously varying map products.
This will provide information on the smallest
detectable amount of vegetative cover at a given
confidence level. In order to better understand
the limitations of space-borne hyperspectral
remote sensing for these invasive species,
simulations will be performed in which the high
resolution SpecTIR imagery will be sequentially
degraded, both spatially and radiometrically, to
match the acquired Hyperion imagery. A
regression-based accuracy analysis will be
applied across the simulations of degraded
SpecTIR products to provide information on the
tradeoffs for differing satellite sensor design
parameters for spectral and spatial response.
This will provide insight into the required
spatial and spectral resolutions and
signal-to-noise performance that will be required
to map these species at a given level of
confidence using space-borne instruments.
Field data were collected over a period of three
days following the overpass of the aircraft.
Percent cover data for the selected invasive
species were collected for quadrats corresponding
to the 1.5 meter pixels that were nested within a
3x3 pixel block. These plots covered a variety
of plant size and cover classes, as well as
different mixtures of plant species. In
addition, polygons representing complete cover
and complete absence by the targeted species were
surveyed with precision GPS. Field spectra were
acquired for vicarious calibration and to support
the generation of vegetative endmembers. A
combination of precision GPS measurements and
recent orthophotography will be used to ensure
proper registration of field measures with
airborne imagery.
High spatial resolution hyperspectral imagery
collected from airborne platforms present a
promising way to address basic and applied
research questions regarding invasive species, as
well as providing a mechanism for better
understanding how available methods scale up to
satellite-based systems. This project is testing
the ability of airborne and space-borne
hyperspectral remote sensing to map and monitor
the distribution of specific noxious, non-native
plants into rangeland, agricultural, and riparian
landscapes of northern Nevada. This analysis
will test multiple endmember spectral mixture
analysis (MESMA) and stepwise discriminant
functions for identifying the extent and percent
cover of Lepidium latifolium (tall whitetop),
Tamarix spp. (tamarisk), and Acroptilon repens
(Russian knapweed) in the Humboldt River basin
and elsewhere in Nevada. Research efforts will be
coordinated with collaborators from the
University of Nevada Cooperative Extension (UNCE)
who work with agricultural and land management
interests in the area. The developed monitoring
techniques will complement ongoing control
activities such as herbicide spraying and the use
of biological control agents.
Hyperspectral Data Acquisition
Measuring cover of tall white-top (bottom-left)
Measuring cover of Russian knapweed
The first acquisition of high resolution imagery
for this effort was on July 17, 2006 under
clear-sky conditions. The imagery was collected
by SpectIR, a Nevada-based provider of airborne
hyperspectral data. The imagery was collected at
1.5 meter resolution with SpectIRs new Dual
hyperspectral imaging system that uses
bore-sighted VNIR and SWIR sensors
(specifications below). Absolute positional error
for the SpectIR imagery is reported as /-3
meters. Space-borne hyperspectral imagery for
the study areas was also collected by the
Hyperion sensor on the Earth Observing-1
satellite. The Hyperion sensor measures radiance
in 220 spectral bands ranging from 0.4 to 2.5
microns covering a ground swath of approximately
7.5 kilometers with 30 meter spatial resolution.
Data Analysis
The first mapping approach to be tested will be
multiple endmember spectral mixture analysis
(MESMA), a form of linear mixture modeling that
allows different, limited combinations of
candidate spectral endmembers for vegetation to
be tested in estimating the composition of each
pixel. (Roberts et al., 1998). This analysis will
be performed using the new Viper Tools ENVI
plugin that has been developed by Dar Roberts
research group at the University of California,
Santa Barbara. While field spectra were acquired,
it is expected that the high spatial resolution
of the SpecTIR imagery will allow the use of
image-based vegetation endmembers. The second
approach to be tested is a stepwise discriminant
function analysis (DFA). The DFA approach
presents an efficient way to identify unique
spectral features of the target plant species and
creates a transformation function that maximizes
the difference of the target species from other
image components. A stepwise approach will
reduce problems with collinearity arising from
correlation between adjacent spectral channels.
As part of this effort we will develop software
to integrate the DFA mapping method into the ENVI
software environment.
Status
Our first field effort was successful, though we
did encounter a number of challenges, including
flooding and idiosyncratic scheduling of cutting
and herbicide application in the agricultural
study area. At this time, data entry has been
completed from the field data sheets for plant
cover that were recorded at the time of the
overflight. SpecTIR is processing the
hyperspectral data to a finished georeferenced
radiance product, and initial quick-look products
suggest good image data quality.
Acknowledgments
This work was funded through the National
Aeronautics and Space Administration
(NNS06AA91G).