Title: Detection and Mapping of Invasive Leafy Spurge Using Orbital Hyperspectral Imagery from the EO-1 Mission

overratedbeltΤεχνίτη Νοημοσύνη και Ρομποτική

25 Νοε 2013 (πριν από 3 χρόνια και 6 μήνες)

109 εμφανίσεις

Final Report:

EO
-
1 EVALUATION AND VALIDATION: NRA
-
99
-
OES
-
01


Title
: Detection and Mapping of Invasive Leafy Spurge Using Orbital Hyperspectral
Imagery from the EO
-
1 Mission


Category:

B
-

Hyperspectral Applications


Principal Investigator Name
: Ralph Root


Department
: U.S. Department of the Interior


Institution
: U.S. Geological Survey, Rocky Mountain Mapping Center

DFC Bldg. 810, MS 516

Denver, CO 80225

USA

E
-
mail
:
ralph_root@usgs.gov

Phone
: (303) 202
-
4339
FAX
: (303) 202
-
4354


Co
-
Investigators:


Ste
ve Hager, National Park Service, Theodore Roosevelt National Park

(701)623
-
4466 x3433
steve_hager@nps.gov


Susan Ustin, University of California Davis, CSTARS Laboratory (530) 752
-
5092

susan@vache.ucdavis.edu


Gerry Anderson, Agricultural Research Service
, Sidney, MT (406) 482
-
9416

gerry@sidney.ars.usda.gov


Raymond Kokaly, U.S. Geological Survey, Spectroscopy Laboratory, Denver, CO

(303)236
-
1359
raymond@s
peclab.cr.usgs.gov


Pre
-
launch accomplishments:

Because the launch of EO
-
1 did not take place in early
2000 one of the first project activities was to reschedule the coordinated AVIRIS mission
from summer of 2000 to summer of 2001. We also arranged wit
h Theodore Roosevelt
National Park to postpone their intensive ground validation effort for delineating leafy
spurge polygons by multiple GPS teams from the 2000 growing season to 2001.


As a result of early (1999) success with AVIRIS in extracting leafy s
purge spectra from
neighboring vegetation types, (Root and Wickland, 2000, 2001), the USDA Agricultural
Research Service Northern Plains Agricultural Research Laboratory made the decision to
purchase a CASI II (Compact Airborne Spectrographic Imager) syste
m to test mapping at
higher spatial resolution within the visible and near infrared portions of the
electromagnetic spectrum. The first CASI data were flown over Theodore Roosevelt NP
and the Little Missouri Grasslands west of the park during the 2000 gro
wing season.
Success in calibrating these data and mapping leafy spurge have been reported by Kokaly
et al. (2001a, 2002), and detection/mapping has been demonstrated using both field
collected spectra (Kokaly et al., 2001b) and from training areas extrac
ted from the
imagery (Root et al., 2001a). CASI data were flown again in the 2001 growing season on
July 5 and 23, and are currently being calibrated and classified to enable comparisons
with the previous growing season. For both the 2000 and 2001 flight
s, the CASI II
spectrometer was configured to collect data from 400 to 1100 nm in 36 spectral bands
with 4 m spatial ground resolution. The CASI data completes the sequence of imaging
spectrometers from low altitude to high altitude to orbital. The CASI d
ata were a
significant addition to this research because it enabled us to make three
-
way comparisons
of classification results from each of these sensors, illustrating specific capabilities of
each platform for detecting and mapping leafy spurge. (Root
et
al.
, 2002).


Also during the 2000 growing season hundreds of ground spectra using three ASD
-
FR
(Analytical Spectral Devices) portable spectrometers. Spectra were collected at not only
the calibration site (a 2+ acre paved parking lot), but also for a wide

range of stands of
leafy spurge, a variety of grasses, sagebrush, rabbit brush, snowberry, yellow sweet
clover, and soil.


2001 data acquisitions:

Three EO
-
1 data acquisitions occurred during 2001 for the
Theodore Roosevelt study area. Dates were May 23
, July 6, and September 24. Leafy
spurge was just emerging on May 3, was in full bract development on July 6, and was in
its fall coloration on September 24. A coordinated AVIRIS overflight occurred on June
21, 2001. Although the goal was to schedule th
e AVIRIS acquisition as close to the
hyperion DCE as possible, the two week interval still allow useful comparisons between
leafy spurge detection with each of the sensors.



Hyperion data analysis:

Level 1 hyperion data from the July 6 Data Collection
Event
(DCE) were manually edited to remove anomalous scan lines (by averaging values of
adjacent lines) and corrected for atmospheric effects. Ground spectra obtained from our
calibration site (large asphalt parking area) were used to calibrate the atmosp
herically
corrected data to surface reflectance. Surface reflectance data were then used to classify
leafy spurge via several methods (spectral angle mapper, red
-
edge spectral parameters,
and iso
-
class clustering), obtaining accuracies ranging from approx
imately 60% to 80%.
Details of methodologies and results are reported in Root
et al.
, 2002b.


2001 field data collection:

The leafy spurge EO
-
1 team and other collaborators working
on related projects conducted a week
-
long joint field data collection effo
rt from July 2
through July 6, 2001. A total of four ASD
-
FR full range ground spectrometers and 10
scientists participated in the 2001 field session. Several thousand spectra of leafy spurge
and related shrubs and grasses were collected, along with sever
al hundred calibration site
spectra which were obtained in conjunction with the EO
-
1 overpass on July 6. In a
separate field effort, several hundred calibration spectra were obtained simultaneously
with the AVIRIS overflight on June 21. Project collabora
tors eventually hope to compile
these spectra, plus those obtained in the three previous years, in a publishable form such
as a CDROM or website, for use by other researchers.


2001 leafy spurge canopy analysis
: (Pablo J. Zarco
-
Tejada (CSTARS, UCDavis)


A

field sampling campaign was carried out for biochemical analysis of leaf chlorophyll
concentrations through SPAD
-
502 chlorophyll meter readings (Minolta Camera Co.,
Ltd., Japan) along with leaf reflectance and transmittance. Leaf and bracts optical
measur
ements were acquired from the leafy spurge samples at different locations in order
to use reflectance and transmittance as inputs in the canopy reflectance models. Scaling
-
up from leaf level to canopy level requires leaf reflectance and transmittance data
as input
to canopy reflectance models.


Single leaf reflectance and transmittance measurements were acquired on all leaves and
bracts using a Li
-
Cor 1800
-
12 Integrating Sphere apparatus coupled by a 200

m
diameter single mode fiber to an GER
-
2600 spectrom
eter yielding a 0.5

nm sampling
interval and 2.5 nm spectral resolution in the 340
-
2500

nm range. Single leaf reflectance
and transmittance measurements were acquired with the integrated sphere following the
methodology described in the manual of the Li
-
Co
r 1800
-
12 system in which six signal
measurements are required: transmittance signal (TSP), reflectance signal (RSS),
reflectance internal standard (RTS), reflectance external reference (RST), and dark
measurements (TDP, RSD). Reflectance (Rfl) and transmi
ttance (Tns) are calculated
assuming a constant center wavelength and spectral bandpass (Equations [1] and [2]).


RSD
RTS
Rfl
RSD
RSS
Rfl
BaSO




4
)
(





[1]

RSD
RST
Rfl
TDP
TSP
Tns
BaSO




4
)
(





[2]


with
Rfl
BaSO4

the reflectance of Barium Sulfate.



The laboratory measurement
s of leafy spurge reflectance and transmittance along with
structural parameters permit the simulation of canopy reflectance through radiative
transfer modeling. The approach adopted (Zarco
-
Tejada
et al.,

2001) is to use leaf level
reflectance and transmit
tance data to calculate the corresponding above
-
canopy
reflectance through optically
-
thick vegetation (infinite reflectance R

) formulae and
through canopy reflectance models.


In more comprehensive approach, the single leaf reflectance and transmittance
data
collected from the ground
-
truth deployment are used to derive above
-
canopy level
reflectances through SAILH
(Verhoef, 1984)
,
Kuusk

(Kuusk, 1996)

and GeoSAIL
(Huemmrich, 2001) canopy reflectance models. The GeoSAIL mod
el allows for
designing multiple components in the simulated canopy reflectance, therefore considering
the optical properties of the sampled elements within the leafy spurge patches. Model
inversion methods as a function of the density of each component in

the simulated scene
will be carried out, estimating by iterative optimization and spectral matching the input
parameters that minimize the error between the measured Hyperion/AVIRIS
hyperspectral reflectance and the simulated reflectance from the model.


A second approach being investigated is based on the calculation of red edge spectral
parameters from hyperspectral data, which are a function of biophysical and biochemical
constituents of the vegetation canopy. A method for the classification of land cov
er has
recently been reported which exploits systematic differences by species of the reflectance
in the short wave infrared spectral regions sensitive to foliar chemistry (Martin
et al
.,
1998). Zarco
-
Tejada and Miller (1999) described classification of ve
getated land cover
based on spectral parameters that characterize the red edge reflectance region, which are
responsive to foliar chlorophyll pigment levels. Classification with three red edge spectral
parameters, red edge inflection point (

p
), the wavelength at the reflectance minimum
(

o
), and a shape parameter (

), as defined by the inverted
-
gaussian red
-
edge curve
-
fit
model (
Hare

et

al.
, 1984;

Miller
et al.,

1990; 1991) were carried out with the calibrated
Hyperion data. The separation
of land cover types and location of leafy spurge patches
using this classification method is based on cover
-
type systematic differences in the
variables known to affect red edge spectral parameters: vegetation chlorophyll content,
canopy structure, canopy
cover, and illumination. The inverted
-
gaussian red
-
edge curve
-
fit model for red
-
edge spectral parameter calculation, adapted for Hyperion reflectance
data, was used to classify leafy spurge with an overall accuracy of 75% (Root

et al.,

2002)


Ground verifi
cation data for leafy spurge distributions during the 2001 growing season:
In 2001 the National Park Service funded a seasonal team of scientists to document
locations and descriptions of leafy spurge populations over several square miles of the
Little Mi
ssouri River floodplain using sub
-
meter accuracy GPS field equipment
(Harrison, 2001). Ground data collected in this project, developed into an ArcInfo
coverage, was used to select spectra for image
-
derived training areas. In a separate study
supported b
y the NPS and USGS, field surveys were conducted for three consecutive
growing seasons from 1999 through 2001 to monitor responses of leafy spurge to
chemical and biological control measures. This project entailed the collection of data
from 550 3
-
m by 5
-
m plots to geographically document the presence/absence of leafy
spurge and to produce detailed estimates of crown cover (through stem counts) and
biomass for leafy spurge and associated native vegetation types. These data were used
for accuracy assessmen
t of the leafy spurge classifications developed in this study.


Final results and research outcomes:


Classification of leafy spurge from reflectance calibrated EO
-
1 Hyperion data obtained
July 6, 2001 (at the peak of bract development) was performed using

three methods: (1)
spectral angle mapper applied to selected principal components, (2) red
-
edge spectral
parameter calculation using the inverted Gaussian model to calculate

p
,

o,

and

and
Rs and Ro parameters pixel by pixel

isoclass clustering using all reflectance bands,
and (4) isoclass clustering with simulated landsat data. Overall classification accuracies
ranged from 63% using the spectral Angle Mapper algorit
hm to 72
-
78% for the red
-
edge
spectral parameters and isoclass clustering algorithms. Accuracy assessment details and
classification methods are discussed in Root

et al.,
2002e.


Simulation of Landsat
-
sensor capabilities from Hyperion data shows promise fo
r
applications with less costly and more available types of data for the future analysis and
monitoring of leafy spurge infestations. However, this study has demonstrated that better
accuracy is achieved when hyperspectral satellite data are used for class
ification.



Suggested improvements to future imaging spectrometers that would further enhance the
detection and mapping of leafy spurge and perhaps other types of invasive plants with
similar growth characteristics would be (1) improved signal
-
to
-
noise ra
tio to enhance
spectral discrimination, (2) increased spatial resolution, and (3) greater swath width to
increase the coverage area and minimize data processing complexities associated with
merging multiple data swaths.


Other collaborative efforts relatin
g to the EO
-
1 study:


Predictive modeling of leafy spurge infestations and spreading characteristics:

As the
topic of his Ph.D. dissertation at Colorado State University, Karl Brown, USGS/BRD is
developing an expert system model incorporating various types

of GIS themes (i.e. soils,
geology, topographic variation, drainage, etc.) with image classification, to identify the
most likely areas of spread.

Spread prediction will incorporate best known factors of
plant physiology, soil preferences, moisture, and
other factors identified in current efforts
in leafy spurge control. By integrating biologic and physiographic factors, the expert
system will assist managers in planning control efforts. The expert system will be
designed to serve as a subroutine within

the Team Leafy Spurge DSS, with outputs
compatible to the DSS display and processing requirements. The factors affecting
spread of leafy spurge, which drive the expert system, would reflect environmental
conditions measured and modeled in the study area
. It is hoped that these results could
be applied in a larger regional context

Expected completion: Early summer, 2002.


Determination of changes in leafy spurge distributions in connection with biological
and chemical control measures undertaken by t
he park:

This is a collaborative Ph.D.
study by Kay Dudek of Colorado State University which involves analyzing 1999 and
2001 AVIRIS data missions to develop leafy spurge distribution maps for both dates,
and to quantify changes in leafy spurge distribut
ions with time/location documented pest
control measures undertaken by the National Park Service.


Exploration of regional mapping techniques:

Although leafy spurge can clearly be
detected and mapped with hyperspectral sensors (on low and high altitude a
ircraft as well
as from orbit), maps of leafy spurge over entire watersheds or states would be a
monumental undertaking with the small data swaths covered by any of today’s
hyperspectral sensors. There are also additional issues associated with storage an
d
analysis logistics. Another aspect of leafy spurge research being analyzed is the the
feasibility of using fused pan/MSS data from more regionally capable sensors such as
Landsat 7, and its likely successors such as is characterized by the ALI (Advance
Land
Imager) on EO
-
1. A team of researchers at the USGS Rocky Mapping Center is currently
examining the possibility of mapping leafy spurge over broader regions by developing
algorithms that take advantage of the combination of multispectral and higher re
solution
panchromatic images available on current medium resolution sensors such as Landsat.

A July 6, 2001 Landsat Scene, obtained 1 minute prior to the EO
-
1 Hyperion and ALI
data acquisitions, is being used to explore the hypothesis that leafy spurge can

be
detected and mapped with Landsat with appropriate types of specifically designed
algorithms. Expected analysis completion: June, 2002.



References


Anderson G. L., C. W. Prosser, R. Root, R. Kokaly, S. Hager, and B. Foster. 2001. A
five
-
year compa
rison of leafy spurge (Euphorbia esula) populations using remote sensing
and geographic information systems. Abstract: 221
st

Annual Meeting of the American
Chemical Society. April 1
-
5, 2001. San Diego, CA. (Invited presentation).


Hare, E. W., J. R. Miller
, and G. R. Edwards. 1984. Studies of the vegetation red
reflectance edge in geobotanical remote sensing, in
Proceedings of the 9
th

Canadian
Symposium on Remote Sensing
, pp. 433
-
440, Can. Remote Sens. Soc., Can. Aeronaut.
and Space Inst., Ottawa, 1984.


Ha
rrison, L. 2001. Ground Truthing Remotely Sensed Leafy Spurge Infestations at
Theodore Roosevelt National Park: Methodology for Mapping Project using Global
Positioning Systems and Geographic Information Systems.
Theodore Roosevelt National
Park. US Depa
rtment of the Interior, National Park Service.
Project Date


Summer
2001. Unpublished manuscript, Theodore Roosevelt National Park.


Huemmrich, K.F. 2001. The GeoSail model: a simple addition to the SAIL model to
describe discontinuous canopy reflectanc
e, Remote Sensing of Environment, 75(3), 2001.


Kokaly, R., R. Root, K. Brown, G. L. Anderson, and S. Hager. 2001a. Calibration of
Compact Airborne Sepctrographic Imager (CASI) data to surface reflectance at Theodore
Roosevelt National Park. Abstract: 2
21
st

Annual Meeting of the American Chemical
Society. April 1
-
5, 2001. San Diego, CA. (Invited paper)


Kokaly, R. R. Root, K. Brown, S. Hager, and G. L. Anderson. 2001b. Discriminating
leafy spurge spectral signature from native vegetation using field ref
lectance
measurements from Theodore Roosevelt National Park, North Dakota. Abstract: 221
st

Annual Meeting of the American Chemical Society. April 1
-
5, 2001. San Diego, CA.
(Invited paper)


Kokaly, R. F., Root, R., and K. Brown. 2001c. Mapping the Distr
ibution of the Invasive
Species Leafy Spurge (
Euphorbia esula
) in Theodore Roosevelt National Park Using
Field Measurements of Vegetation Spectra and CASI Imaging Spectroscopy Data. Third
International Conference on Geospatial Information in Agriculture a
nd Forestry. Denver,
Colorado. 5
-
7 November, 2001. Poster presentation: (won best in session award, in
“Geospatial Information and Rangeland Management” poster session)


Kokaly, R.F., R. Root, K. Brown, S. Hager, and G. Anderson. 2002. “Calibration of

CASI Data to Surface Reflectance at Theodore Roosevelt National Park for Mapping
Invasive Species.” Submitted June 7, 2002, to
International Journal of Remote Sensing
.


Kuusk, A. 1996. A computer
-
efficient plant canopy reflectance model,
Computers &
Geos
ciences.

22
:
149
-
163.


Martin, M. E., S. D. Newman, J. D. Aber, and R. G. Congalton. 1998. Determining forest
species composition using high spectral resolution remote sensing data,
Remote Sens.
Environ
, 65, 249
-
254, 1998.


Miller, J. R., E. W. Hare, and J.

Wu. 1990. Quantitative characterization of the vegetation
red edge reflectance I. An inverted
-
Gaussian reflectance model,
Int. J. Remote Sens.,

11,
121
-
127, 1990.


Miller, J. R., J. Wu, M. G. Boyer, M. J. Belanger, and E.W. Hare. 1991. Seasonal patterns
i
n leaf reflectance red edge characteristics,
Int. J. Remote Sens.
, 12, 1509
-
1524, 1991.


Root, R., R. Kokaly, K. Brown, and G. L. Anderson. 2001. Detection of leafy spurge
infestations via imaging spectroscopy using the Compact Airborne Spectrographic
Im
ager (CASI). Abstract: 221
st

Annual Meeting of the American Chemical Society. April
1
-
5, 2001. San Diego, CA. (Invited paper)


Root, R. and D. Wickland. 2000. Hyperspectral Technology Transfer to the U.S.
Department of the Interior: Status and Results of

the NASA/Department of the Interior
Hyperspectral Technology Transfer Project. Proceedings of the Ninth JPL Airborne
Earth Science Workshop. NASA Jet Propulsion Laboratory, Pasadena, CA. JPL
Publication 00
-
18.


Root, R. and D. Wickland. 2001. Hyperspe
ctral Technology Transfer to the U.S.
Department of the Interior: Hyperspectral Technology Transfer to the U.S. Dept. of
Interior: Summary of Results of the NASA/DOI Hyperspectral Technology Transfer
Project Proceedings of the Tenth JPL Airborne Earth S
cience Workshop. NASA Jet
Propulsion Laboratory, Pasadena, CA.


Root, R.,
Ray Kokaly, Karl Brown, Carol Mladinich, Susan Stitt, and Alex Konduris,
Susan Ustin, Pablo Zarco
-
Tejada, Steve Hager, Gerry Anderson, Kay Dudek, and Ed
Holroyd. 2002a.
Comparison
s of orbital and aircraft hyperspectral systems for
classification and mapping of invasive leafy spurge in southwestern North Dakota. U.S.
Geological Survey Colloquium, January 24, 2002. Denver Federal Center, Denver, CO.
Monthly scientific forum, presen
ted to the general public by the USGS.


Root, R.,
Susan Ustin, Pablo Zarco
-
Tejada, Carlos Pinilla, Ray Kokaly, Karl Brown,
Gerry Anderson, Steve Hager, Kay Dudek, and Ed Holroyd.

2002b. Comparison of
High, Medium, and Low Spatial Resolution Hyperspectra
l Sensors for Mapping of
Invasive Leafy Spurge in Theodore Roosevelt National Park, North Dakota

2002
ASPRS
-
ACSM Annual Conference and FIG Congress, Washington, DC, April 22
-
26,
2002 (abstract).


Root, R., Kokaly, R., Anderson, G., and Hager, S. 2002
c. Comparison of EO
-
1
hyperion, AVIRIS, and CASI hyperspectral imaging systems for detecting and mapping
leafy spurge infestations in southwestern North Dakota. Preceedings of the 11
th

JPL
Airborne Earth Science Workshop. NASA
-
JPL, Pasadena, CA. March,

2002


Root, R.,
Susan Ustin, Pablo Zarco
-
Tejada, Carlos Pinilla, Ray Kokaly, Karl Brown,
Gerry Anderson, Steve Hager, Kay Dudek, and Ed Holroyd. 2002d.
Identification,
Canopy Characterization, and Mapping of Invasive Leafy Spurge with the EO
-
1
Hyperion
Orbital Imaging Spectrometer. 2002 International Geoscience and Remote
Sensing Symposium, Toronto, June 25, 2002.


Root, R., Zarco
-
Tejada, P., Ustin, S., Anderson, G., Kokaly, R., Hagar, S., Brown, K.
2002e. Comparison of orbital EO
-
1 hyperion and high

altitude aircraft AVIRIS imaging
spectrometers for detecting and mapping invasive leafy spurge at Theodore Roosevelt
National Park, ND. IEEE transactions for the International Geoscience and Remote
Sensing Society, Special Issue on initial results of the

investigations conducted by the
NASA EO
-
1 Science Validation Team. Submitted May, 2002, currently under peer
review.


Verhoef, W. 1984. Light scattering by leaf layers with application to canopy reflectance
modelling: the SAIL model, Remote Sensing of E
nvironment, 16:125
-
141, 1984.


Zarco
-
Tejada, P. J. and Miller, J. R. 1999. Land cover mapping at BOREAS using red
edge spectral parameters from CASI imagery,
Journal of Geophysical Research.

104
:
27921
-
27933.


Zarco
-
Tejada, P.J., J.R. Miller, G.H. Mohammed,

T.L. Noland and P.H. Sampson. 2001.
Scaling
-
up and Model Inversion methods with narrow
-
band Optical Indices for
Chlorophyll Content Estimation in closed Forest Canopies with Hyperspectral Data, IEEE
Transactions on Geoscience and Remote Sensing, 39(7), 2
001.