Upload
others
View
4
Download
0
Embed Size (px)
Citation preview
IntroductionMethods
ResultsConclusions/Outlook
Assessing sub-pixel vegetation structure fromimaging spectroscopy data via simulation
Wei Yao, Martin van Leeuwen, Paul Romanczyk,Dave Kelbe, Mingming Wang, Scott Brown,
Adam Goodenough, and Jan van Aardt
Chester F. Carlson Center for Imaging ScienceRochester Institute of Technology
14 October 2015
2015-10-14 Wei Yao 2015 HyspIRI Science and Application Workshop 1
IntroductionMethods
ResultsConclusions/Outlook
Outline
1 IntroductionProject outline and objectivesDIRSIG
2 MethodsStudy areaAirborne and field dataBuilding virtual scenesDIRSIG simulation
3 ResultsEstimating LAI from VIsSimulating AVIRIS & HyspIRI dataAssessing sub-pixel vegetation structure
4 Conclusions/Outlook
2015-10-14 Wei Yao 2015 HyspIRI Science and Application Workshop 2
IntroductionMethods
ResultsConclusions/Outlook
Project outline and objectivesDIRSIG
IntroductionProject outline and objectives
EsimatingLAI
from VIs
Sub-pixelstructuralvariation
DIRSIGsimulation
Systemstudy
HyspIRIPSF
Airbornedata
Fielddata
2015-10-14 Wei Yao 2015 HyspIRI Science and Application Workshop 3
IntroductionMethods
ResultsConclusions/Outlook
Project outline and objectivesDIRSIG
IntroductionProject outline and objectives
Stage 1: Estimating LAI from VIs EstimatingLAI
from VIs
Sub-pixelstructuralvariation
DIRSIGsimulation
Systemstudy
HyspIRIPSF
Airbornedata
Fielddata
AVIRISData
FieldLAI
ExtractingVIs
Downsampling
VIsVIs
Downsampling
LAI
EstimatingLAI
from VIs
2015-10-14 Wei Yao 2015 HyspIRI Science and Application Workshop 4
IntroductionMethods
ResultsConclusions/Outlook
Project outline and objectivesDIRSIG
IntroductionProject outline and objectives
Stage 2: Sub-pixel structural variation EstimatingLAI
from VIs
Sub-pixelstructuralvariation
DIRSIGsimulation
Systemstudy
HyspIRIPSF
Airbornedata
Fielddata
Structuralvariation
Canopystruture
DBH
Stem position
Volume
Biomass
Ground-basedLiDAR
LAI(VIs)
Assess theimpact on
spectraAVIRIS data
Groundspectra
Grass biomass
NEON’shigh res data
Spa
tial
lyex
plic
it
2015-10-14 Wei Yao 2015 HyspIRI Science and Application Workshop 5
IntroductionMethods
ResultsConclusions/Outlook
Project outline and objectivesDIRSIG
IntroductionProject outline and objectives
Stage 3: DIRSIG simulation EstimatingLAI
from VIs
Sub-pixelstructuralvariation
DIRSIGsimulation
Systemstudy
HyspIRIPSF
Airbornedata
Fielddata
Structuralvariation
Virtual scenesDIRSIG
simulation
SimulatedHyspIRI pixel
SimulatedAVIRIS pixel
AVIRIS data Verification
DIRSIG: The Digital Imaging and Remote Sensing Image Generation
2015-10-14 Wei Yao 2015 HyspIRI Science and Application Workshop 6
IntroductionMethods
ResultsConclusions/Outlook
Project outline and objectivesDIRSIG
IntroductionProject outline and objectives
Stage 4: System study EstimatingLAI
from VIs
Sub-pixelstructuralvariation
DIRSIGsimulation
Systemstudy
HyspIRIPSF
Airbornedata
Fielddata
SimulatedHyspIRI pixel
Virtual scenes
SimulatedAVIRIS pixel
HyspIRI PSF
AVIRIS PSF
2015-10-14 Wei Yao 2015 HyspIRI Science and Application Workshop 7
IntroductionMethods
ResultsConclusions/Outlook
Project outline and objectivesDIRSIG
IntroductionDIRSIG simulation - overview
DIRSIG = Digital Imaging and Remote Sensing Image Generation ModelUnder development for 20+ years at Rochester Institute of Technology
http://dirsig.org
Image ModalitiesVisible through thermal infrared(0.4 - 20.0 µm)Passive sensingActive Laser sensingActive RF sensing
InstrumentsSingle pixel, 1D arrays and 2D arrays.Filter, diffraction/refraction, orinterferogram-based photon collection
PlatformsGround, air or space on static or movingplatforms
2015-10-14 Wei Yao 2015 HyspIRI Science and Application Workshop 8
IntroductionMethods
ResultsConclusions/Outlook
Study areaAirborne and field dataBuilding virtual scenesDIRSIG simulation
MethodsStudy area
The National Ecological Observatory Network (NEON),Pacific Southwest Domain (D17)
1 San Joaquin Experiment Range (Core site)2 Soaproot Saddle (Relocatable site)
2015-10-14 Wei Yao 2015 HyspIRI Science and Application Workshop 9
IntroductionMethods
ResultsConclusions/Outlook
Study areaAirborne and field dataBuilding virtual scenesDIRSIG simulation
MethodsField collection
1 San Joaquin Experiment Range:June 9 - 14, 2013: 12 AOP sites (4, 8, 36, 112, 116, 361,824, 952)Oct 5 - 7, 2014: 3 AOP sites (36, 116, 824)
Site 36 Site 116
AOP: Airborne Observation Platform2015-10-14 Wei Yao 2015 HyspIRI Science and Application Workshop 10
IntroductionMethods
ResultsConclusions/Outlook
Study areaAirborne and field dataBuilding virtual scenesDIRSIG simulation
MethodsField collection
2 Soaproot Saddle:June 16 - 20, 2013: 8 AOP sites (43, 63, 95, 143, 299, 331,555, 1611)Oct 8 - 10, 2014: 3 AOP sites (43, 143, 299)
Site 143 Site 299
AOP: Airborne Observation Platform2015-10-14 Wei Yao 2015 HyspIRI Science and Application Workshop 11
IntroductionMethods
ResultsConclusions/Outlook
Study areaAirborne and field dataBuilding virtual scenesDIRSIG simulation
MethodsField collection
Measurements at each spotwithin 80× 80m site:
1 LAI (AccuPAR LP-80)
2 Ground-based lidar(SICK LMS-151, RITTL)
3 Spectra (SVC HR-1024i)
4 Hemispherical photos
5 GPS position
2015-10-14 Wei Yao 2015 HyspIRI Science and Application Workshop 12
IntroductionMethods
ResultsConclusions/Outlook
Study areaAirborne and field dataBuilding virtual scenesDIRSIG simulation
MethodsAirborne collection
AVIRIS data collected during HyspIRI preparatoryairborne campaign, summer 2013 & fall 2014.
NEON imaging spectrometer (NIS) & LiDAR datacollected in summer 2013.
2015-10-14 Wei Yao 2015 HyspIRI Science and Application Workshop 13
IntroductionMethods
ResultsConclusions/Outlook
Study areaAirborne and field dataBuilding virtual scenesDIRSIG simulation
MethodsRegister multiple Terrestrial Laser Scanner (TLS) scans and extract stem mapfrom TLS data
Algorithm:Extract coordinates of trunk-groundintersection as tie-points forregistrationRank potential correspondence setsusing geometric constraintsUse RANSAC to query candidatepoint set matches
Features:No markers are placed in the sceneNo initial pose estimation is required
By Dave Kelbe, PhD student
2015-10-14 Wei Yao 2015 HyspIRI Science and Application Workshop 14
IntroductionMethods
ResultsConclusions/Outlook
Study areaAirborne and field dataBuilding virtual scenesDIRSIG simulation
MethodsSimulate NEON’s high-resolution spectrometer data
NISdata
DIRSIGdata
Site 116 Site 299 Site 143
2015-10-14 Wei Yao 2015 HyspIRI Science and Application Workshop 15
IntroductionMethods
ResultsConclusions/Outlook
Study areaAirborne and field dataBuilding virtual scenesDIRSIG simulation
MethodsSimulate HyspIRI
DIRSIG key settings
Height = 600km
GSD = 60m
224 bands, 380 - 2500nm, 10nm FWHM
Use MODTRAN to simulate atmospheric radiativetransfer
MODTRAN key settings:
Enable multiple scattering (IMULT = +1)
Mid-latitude summer model (MODEL = 2)
RURAL extinction (IHAZE = 1)
2015-10-14 Wei Yao 2015 HyspIRI Science and Application Workshop 16
IntroductionMethods
ResultsConclusions/Outlook
Study areaAirborne and field dataBuilding virtual scenesDIRSIG simulation
MethodsSimulate HyspIRI
Point spread function (PSF)2-D Gaussian Function, FWHM = pixel size (60m GSD)
2-D Gaussian kernel Profile of the kernel
2015-10-14 Wei Yao 2015 HyspIRI Science and Application Workshop 17
IntroductionMethods
ResultsConclusions/Outlook
Estimating LAI from VIsSimulating AVIRIS & HyspIRI dataAssessing sub-pixel vegetation structure
ResultsEstimating LAI from VIs
Simulated forest LAI vs NDVI
5 m transect spacing 10 m transect spacing 15 m transect spacing
0 0.1 0.2 0.3 0.4 0.5 0.6 0.70
2
4
6
NDVI
LA
I
DataModel
0 0.1 0.2 0.3 0.4 0.5 0.6 0.70
2
4
6
NDVI
LA
I
DataModel
0 0.1 0.2 0.3 0.4 0.5 0.6 0.70
2
4
6
NDVI
LA
I
DataModel
LAI = 8.826 · NDVI - 1.506 LAI = 8.928 · NDVI - 1.566 LAI = 12.61 · NDVI - 2.457
R2 = 0.92 R2 = 0.77 R2 = 0.66
Forest LAI can be measured along multiple transects.
2015-10-14 Wei Yao 2015 HyspIRI Science and Application Workshop 18
IntroductionMethods
ResultsConclusions/Outlook
Estimating LAI from VIsSimulating AVIRIS & HyspIRI dataAssessing sub-pixel vegetation structure
ResultsEstimating LAI from VIs
Measured forest LAI vs NDVI
10 m transect spacing
0 0.1 0.2 0.3 0.4 0.50
0.5
1
1.5
2
2.5
NDVI
LA
I
DataModel
LAI = 8.858 · NDVI - 1.725
R2 = 0.61
Regression model from simulated data:
LAI = 8.928 · NDVI - 1.566
R2 = 0.77
Forest LAI can be measure along multiple transects.
2015-10-14 Wei Yao 2015 HyspIRI Science and Application Workshop 19
IntroductionMethods
ResultsConclusions/Outlook
Estimating LAI from VIsSimulating AVIRIS & HyspIRI dataAssessing sub-pixel vegetation structure
MethodsSimulating AVIRIS data
Verify the model using site 116 scene
AVIRISdata
Simulateddata
500 1000 1500 20000
500
1000
1500
2000
Wavelength (nm)
Ra
dia
nce
(µ
W c
m−
2 n
m−
1 s
r−1)
A1
B1
500 1000 1500 20000
500
1000
1500
2000
Wavelength (nm)
Ra
dia
nce
(µ
W c
m−
2 n
m−
1 s
r−1)
A2
B2
500 1000 1500 20000
500
1000
1500
2000
Wavelength (nm)
Ra
dia
nce
(µ
W c
m−
2 n
m−
1 s
r−1)
A3
B3
500 1000 1500 20000
500
1000
1500
2000
Wavelength (nm)
Ra
dia
nce
(µ
W c
m−
2 n
m−
1 s
r−1)
A4
B4
2015-10-14 Wei Yao 2015 HyspIRI Science and Application Workshop 20
IntroductionMethods
ResultsConclusions/Outlook
Estimating LAI from VIsSimulating AVIRIS & HyspIRI dataAssessing sub-pixel vegetation structure
ResultsAssessing sub-pixel vegetation structure
Density of the “forest”
60m
60m
500 1000 1500 20000
0.1
0.2
0.3
0.4
0.5
Wavelength (nm)
Re
fle
cta
nce
d = 0.20 Spectrum of the center pixel
Density of the “forest” is quantified by tree canopy cover (d), the proportion of land area covered by tree crowns.
2015-10-14 Wei Yao 2015 HyspIRI Science and Application Workshop 21
IntroductionMethods
ResultsConclusions/Outlook
Estimating LAI from VIsSimulating AVIRIS & HyspIRI dataAssessing sub-pixel vegetation structure
ResultsAssessing sub-pixel vegetation structure
Density of the “forest”
60m
60m
500 1000 1500 20000
0.1
0.2
0.3
0.4
0.5
Wavelength (nm)
Re
fle
cta
nce
d = 0.22 Spectrum of the center pixel
Density of the “forest” is quantified by tree canopy cover (d), the proportion of land area covered by tree crowns.
2015-10-14 Wei Yao 2015 HyspIRI Science and Application Workshop 21
IntroductionMethods
ResultsConclusions/Outlook
Estimating LAI from VIsSimulating AVIRIS & HyspIRI dataAssessing sub-pixel vegetation structure
ResultsAssessing sub-pixel vegetation structure
Density of the “forest”
60m
60m
500 1000 1500 20000
0.1
0.2
0.3
0.4
0.5
Wavelength (nm)
Re
fle
cta
nce
d = 0.24 Spectrum of the center pixel
Density of the “forest” is quantified by tree canopy cover (d), the proportion of land area covered by tree crowns.
2015-10-14 Wei Yao 2015 HyspIRI Science and Application Workshop 21
IntroductionMethods
ResultsConclusions/Outlook
Estimating LAI from VIsSimulating AVIRIS & HyspIRI dataAssessing sub-pixel vegetation structure
ResultsAssessing sub-pixel vegetation structure
Density of the “forest”
60m
60m
500 1000 1500 20000
0.1
0.2
0.3
0.4
0.5
Wavelength (nm)
Re
fle
cta
nce
d = 0.25 Spectrum of the center pixel
Density of the “forest” is quantified by tree canopy cover (d), the proportion of land area covered by tree crowns.
2015-10-14 Wei Yao 2015 HyspIRI Science and Application Workshop 21
IntroductionMethods
ResultsConclusions/Outlook
Estimating LAI from VIsSimulating AVIRIS & HyspIRI dataAssessing sub-pixel vegetation structure
ResultsAssessing sub-pixel vegetation structure
Density of the “forest”
60m
60m
500 1000 1500 20000
0.1
0.2
0.3
0.4
0.5
Wavelength (nm)
Re
fle
cta
nce
d = 0.30 Spectrum of the center pixel
Density of the “forest” is quantified by tree canopy cover (d), the proportion of land area covered by tree crowns.
2015-10-14 Wei Yao 2015 HyspIRI Science and Application Workshop 21
IntroductionMethods
ResultsConclusions/Outlook
Estimating LAI from VIsSimulating AVIRIS & HyspIRI dataAssessing sub-pixel vegetation structure
ResultsAssessing sub-pixel vegetation structure
Density of the “forest”
60m
60m
500 1000 1500 20000
0.1
0.2
0.3
0.4
0.5
Wavelength (nm)
Re
fle
cta
nce
d = 0.36 Spectrum of the center pixel
Density of the “forest” is quantified by tree canopy cover (d), the proportion of land area covered by tree crowns.
2015-10-14 Wei Yao 2015 HyspIRI Science and Application Workshop 21
IntroductionMethods
ResultsConclusions/Outlook
Estimating LAI from VIsSimulating AVIRIS & HyspIRI dataAssessing sub-pixel vegetation structure
ResultsAssessing sub-pixel vegetation structure
Density of the “forest”
60m
60m
500 1000 1500 20000
0.1
0.2
0.3
0.4
0.5
Wavelength (nm)
Re
fle
cta
nce
d = 0.40 Spectrum of the center pixel
Density of the “forest” is quantified by tree canopy cover (d), the proportion of land area covered by tree crowns.
2015-10-14 Wei Yao 2015 HyspIRI Science and Application Workshop 21
IntroductionMethods
ResultsConclusions/Outlook
Estimating LAI from VIsSimulating AVIRIS & HyspIRI dataAssessing sub-pixel vegetation structure
ResultsAssessing sub-pixel vegetation structure
Density of the “forest”
60m
60m
500 1000 1500 20000
0.1
0.2
0.3
0.4
0.5
Wavelength (nm)
Re
fle
cta
nce
d = 0.43 Spectrum of the center pixel
Density of the “forest” is quantified by tree canopy cover (d), the proportion of land area covered by tree crowns.
2015-10-14 Wei Yao 2015 HyspIRI Science and Application Workshop 21
IntroductionMethods
ResultsConclusions/Outlook
Estimating LAI from VIsSimulating AVIRIS & HyspIRI dataAssessing sub-pixel vegetation structure
ResultsAssessing sub-pixel vegetation structure
Density of the “forest”
60m
60m
500 1000 1500 20000
0.1
0.2
0.3
0.4
0.5
Wavelength (nm)
Re
fle
cta
nce
d = 0.50 Spectrum of the center pixel
Density of the “forest” is quantified by tree canopy cover (d), the proportion of land area covered by tree crowns.
2015-10-14 Wei Yao 2015 HyspIRI Science and Application Workshop 21
IntroductionMethods
ResultsConclusions/Outlook
Estimating LAI from VIsSimulating AVIRIS & HyspIRI dataAssessing sub-pixel vegetation structure
ResultsAssessing sub-pixel vegetation structure
Density of the “forest”
60m
60m
500 1000 1500 20000
0.1
0.2
0.3
0.4
0.5
Wavelength (nm)
Re
fle
cta
nce
d = 0.61 Spectrum of the center pixel
Density of the “forest” is quantified by tree canopy cover (d), the proportion of land area covered by tree crowns.
2015-10-14 Wei Yao 2015 HyspIRI Science and Application Workshop 21
IntroductionMethods
ResultsConclusions/Outlook
Estimating LAI from VIsSimulating AVIRIS & HyspIRI dataAssessing sub-pixel vegetation structure
ResultsAssessing sub-pixel vegetation structure
Density of the “forest”
500 1000 1500 20000
0.1
0.2
0.3
0.4
Wavelength (nm)
Reflecta
nce
d = 0.20
d = 0.22
d = 0.24
d = 0.25
d = 0.30
d = 0.36
d = 0.40
d = 0.43
d = 0.50
d = 0.61
2015-10-14 Wei Yao 2015 HyspIRI Science and Application Workshop 22
IntroductionMethods
ResultsConclusions/Outlook
Estimating LAI from VIsSimulating AVIRIS & HyspIRI dataAssessing sub-pixel vegetation structure
ResultsNDVI vs density
Normalized DifferenceVegetation Index (NDVI)
NDVI =NIR − Red
NIR + Red
“Forest” densityTree canopy cover refers to theproportion of land area coveredby tree crowns (m2/m2). 0.1 0.2 0.3 0.4 0.5 0.6
0.2
0.3
0.4
0.5
0.6
0.7
0.8
ND
VI
Canopy cover (unitless)
R2 = 0.9735
2015-10-14 Wei Yao 2015 HyspIRI Science and Application Workshop 23
IntroductionMethods
ResultsConclusions/Outlook
Estimating LAI from VIsSimulating AVIRIS & HyspIRI dataAssessing sub-pixel vegetation structure
ResultsNarrow band vegetation indices (VIs) to characterize the forest density
VI =Band1− Band2
Band1 + Band2
R2 value of regression models
Wavelength of Band 1 (nm)
Wa
ve
len
gth
of
Ba
nd
2 (
nm
)
500 1000 1500 2000 2500
500
1000
1500
2000
2500
<0.95
0.955
0.96
0.965
0.97
0.975
0.98
Wavelength of Band 1 (nm)
Wa
ve
len
gth
of
Ba
nd
2 (
nm
)
500 1000 1500 2000 2500
500
1000
1500
2000
2500
<0.95
0.955
0.96
0.965
0.97
0.975
0.98
Reflectance VI Radiance VI2015-10-14 Wei Yao 2015 HyspIRI Science and Application Workshop 24
IntroductionMethods
ResultsConclusions/Outlook
Estimating LAI from VIsSimulating AVIRIS & HyspIRI dataAssessing sub-pixel vegetation structure
ResultsAssessing sub-pixel vegetation structure
Tree position
60m
60m
500 1000 1500 20000
0.1
0.2
0.3
0.4
0.5
Wavelength (nm)
Re
fle
cta
nce
Tree at (0, 0) Spectrum of the center pixel
2015-10-14 Wei Yao 2015 HyspIRI Science and Application Workshop 25
IntroductionMethods
ResultsConclusions/Outlook
Estimating LAI from VIsSimulating AVIRIS & HyspIRI dataAssessing sub-pixel vegetation structure
ResultsAssessing sub-pixel vegetation structure
Tree position
60m
60m
500 1000 1500 20000
0.1
0.2
0.3
0.4
0.5
Wavelength (nm)
Re
fle
cta
nce
Tree at (10, 0) Spectrum of the center pixel
2015-10-14 Wei Yao 2015 HyspIRI Science and Application Workshop 25
IntroductionMethods
ResultsConclusions/Outlook
Estimating LAI from VIsSimulating AVIRIS & HyspIRI dataAssessing sub-pixel vegetation structure
ResultsAssessing sub-pixel vegetation structure
Tree position
60m
60m
500 1000 1500 20000
0.1
0.2
0.3
0.4
0.5
Wavelength (nm)
Re
fle
cta
nce
Tree at (20, 0) Spectrum of the center pixel
2015-10-14 Wei Yao 2015 HyspIRI Science and Application Workshop 25
IntroductionMethods
ResultsConclusions/Outlook
Estimating LAI from VIsSimulating AVIRIS & HyspIRI dataAssessing sub-pixel vegetation structure
ResultsAssessing sub-pixel vegetation structure
Tree position
60m
60m
500 1000 1500 20000
0.1
0.2
0.3
0.4
0.5
Wavelength (nm)
Re
fle
cta
nce
Tree at (30, 0) Spectrum of the center pixel
2015-10-14 Wei Yao 2015 HyspIRI Science and Application Workshop 25
IntroductionMethods
ResultsConclusions/Outlook
Estimating LAI from VIsSimulating AVIRIS & HyspIRI dataAssessing sub-pixel vegetation structure
ResultsAssessing sub-pixel vegetation structure
Tree position
60m
60m
500 1000 1500 20000
0.1
0.2
0.3
0.4
0.5
Wavelength (nm)
Re
fle
cta
nce
Tree at (40, 0) Spectrum of the center pixel
2015-10-14 Wei Yao 2015 HyspIRI Science and Application Workshop 25
IntroductionMethods
ResultsConclusions/Outlook
Estimating LAI from VIsSimulating AVIRIS & HyspIRI dataAssessing sub-pixel vegetation structure
ResultsAssessing sub-pixel vegetation structure
Tree position
60m
60m
500 1000 1500 20000
0.1
0.2
0.3
0.4
0.5
Wavelength (nm)
Re
fle
cta
nce
Tree at (50, 0) Spectrum of the center pixel
2015-10-14 Wei Yao 2015 HyspIRI Science and Application Workshop 25
IntroductionMethods
ResultsConclusions/Outlook
Estimating LAI from VIsSimulating AVIRIS & HyspIRI dataAssessing sub-pixel vegetation structure
ResultsAssessing sub-pixel vegetation structure
Tree position
60m
60m
500 1000 1500 20000
0.1
0.2
0.3
0.4
0.5
Wavelength (nm)
Re
fle
cta
nce
Tree at (60, 0) Spectrum of the center pixel
2015-10-14 Wei Yao 2015 HyspIRI Science and Application Workshop 25
IntroductionMethods
ResultsConclusions/Outlook
Estimating LAI from VIsSimulating AVIRIS & HyspIRI dataAssessing sub-pixel vegetation structure
ResultsAssessing sub-pixel vegetation structure
Tree position
500 1000 1500 20000
0.1
0.2
0.3
0.4
0.5
Wavelength (nm)
Reflecta
nce
(0, 0)
(10, 0)
(20, 0)
(30, 0)
(40, 0)
(50, 0)
(60, 0)
2015-10-14 Wei Yao 2015 HyspIRI Science and Application Workshop 26
IntroductionMethods
ResultsConclusions/Outlook
Estimating LAI from VIsSimulating AVIRIS & HyspIRI dataAssessing sub-pixel vegetation structure
ResultsAssessing sub-pixel vegetation structure
Tree position
500 1000 1500 20000
0.1
0.2
0.3
0.4
0.5
Wavelength (nm)
Reflecta
nce
(0, 0)
(10, 0)
(20, 0)
(30, 0)
(40, 0)
(50, 0)
(60, 0)
500 1000 1500 20000
0.1
0.2
0.3
0.4
0.5
Wavelength (nm)
Re
fle
cta
nce
(0, 0)
(10, 0)
(20, 0)
(30, 0)
(40, 0)
(50, 0)
(60, 0)
2015-10-14 Wei Yao 2015 HyspIRI Science and Application Workshop 26
IntroductionMethods
ResultsConclusions/Outlook
Estimating LAI from VIsSimulating AVIRIS & HyspIRI dataAssessing sub-pixel vegetation structure
ResultsAssessing sub-pixel vegetation structure
Tree position
500 1000 1500 20000
0.1
0.2
0.3
0.4
0.5
Wavelength (nm)
Reflecta
nce
(0, 0)
(10, 0)
(20, 0)
(30, 0)
(40, 0)
(50, 0)
(60, 0)
500 1000 1500 20000
0.1
0.2
0.3
0.4
0.5
Wavelength (nm)
Re
fle
cta
nce
(0, 0)
(10, 0)
(20, 0)
(30, 0)
(40, 0)
(50, 0)
(60, 0)
500 1000 1500 20000
0.1
0.2
0.3
0.4
0.5
Wavelength (nm)
Re
fle
cta
nce
(0, 0)
(10, 0)
(20, 0)
(30, 0)
(40, 0)
(50, 0)
(60, 0)
2015-10-14 Wei Yao 2015 HyspIRI Science and Application Workshop 26
IntroductionMethods
ResultsConclusions/Outlook
Estimating LAI from VIsSimulating AVIRIS & HyspIRI dataAssessing sub-pixel vegetation structure
ResultsAssessing sub-pixel vegetation structure
Tree position
500 1000 1500 20000
0.1
0.2
0.3
0.4
0.5
Wavelength (nm)
Reflecta
nce
(0, 0)
(10, 0)
(20, 0)
(30, 0)
(40, 0)
(50, 0)
(60, 0)
500 1000 1500 20000
0.1
0.2
0.3
0.4
0.5
Wavelength (nm)
Re
fle
cta
nce
(0, 0)
(10, 0)
(20, 0)
(30, 0)
(40, 0)
(50, 0)
(60, 0)
500 1000 1500 20000
0.1
0.2
0.3
0.4
0.5
Wavelength (nm)
Re
fle
cta
nce
(0, 0)
(10, 0)
(20, 0)
(30, 0)
(40, 0)
(50, 0)
(60, 0)
500 1000 1500 20000
0.1
0.2
0.3
0.4
0.5
Wavelength (nm)
Re
fle
cta
nce
(0, 0)
(10, 0)
(20, 0)
(30, 0)
(40, 0)
(50, 0)
(60, 0)
2015-10-14 Wei Yao 2015 HyspIRI Science and Application Workshop 26
IntroductionMethods
ResultsConclusions/Outlook
Estimating LAI from VIsSimulating AVIRIS & HyspIRI dataAssessing sub-pixel vegetation structure
ResultsThe spectral angle distribution
θ = cos−1
[x1 · x2‖x1‖‖x2‖
]
−80 −60 −40 −20 0 20 40 60 80−80
−60
−40
−20
0
20
40
60
80
Distance from the center of the terrain (m)
Dis
tance fro
m the c
ente
r of th
e terr
ain
(m
)
0.2
0.4
0.6
0.8
1
1.2
−80−60
−40−20
020
4060
80
−50
0
50
0
0.2
0.4
0.6
0.8
1
1.2
1.4
Distance (m)Distance (m)
Spectr
a a
ngle
(degre
e)
2015-10-14 Wei Yao 2015 HyspIRI Science and Application Workshop 27
IntroductionMethods
ResultsConclusions/Outlook
Conclusions/OutlookConclusions
Results indicate:
1 VIs could be used to estimate forest density from HyspIRIdata.
2 The effect of vegetation’s position is mainly determinedby the PSF of spectrometer.
3 The system’s suitability for consistent global vegetationstructural assessments could be improved by adaptingcalibration strategies to account for this variation insub-pixel structure.
2015-10-14 Wei Yao 2015 HyspIRI Science and Application Workshop 28
IntroductionMethods
ResultsConclusions/Outlook
Conclusions/OutlookFuture work
1 Re-run current simulations according to updated VSWIRspecification:
30 m GSDnew PSF
2 Increase the number of simulations to assess othersub-pixel vegetation structural variables:
height of treescrown sizeforest species
3 Quantify the simulation results.4 Investigate Lidar-based approaches for calibration of
HyspIRI structural estimates.
2015-10-14 Wei Yao 2015 HyspIRI Science and Application Workshop 29
IntroductionMethods
ResultsConclusions/Outlook
Conclusions/OutlookAssociated Publications
Refereed journal articlesYao, W., van Aardt, J., Romanczyk, P., Kelbe, D., van Leeuwen, M., Brown,
S., and Goodenough, G. “Reviewing LAI collection protocol – a simulationstudy focused on PAR sensor”. Sensors. In preparation.
Yao, W., van Aardt, J., Romanczyk, P., Kelbe, D., van Leeuwen, M., Brown,S., and Goodenough, G. “A simulation approach to assessing sub-pixelvegetation structure impacts on imaging spectrometer response”. IEEETransactions on Geoscience and Remote Sensing. In preparation.
2015-10-14 Wei Yao 2015 HyspIRI Science and Application Workshop 30
IntroductionMethods
ResultsConclusions/Outlook
Conclusions/OutlookAssociated Publications
Conference Proceedings & PostersYao, W., van Aardt, J., Romanczyk, P., Brown, S., Kelbe, D., van Leeuwen,
M., and Kerekes, J. (2015b). “Towards robust forest leaf area index assessmentusing an imaging spectroscopy simulation approach”. In: Proceedings ofInternational Geoscience and Remote Sensing Symposium (IGARSS). Milan,Italy. July 26-31, 2015.
Yao, W., van Aardt, J., Romanczyk, P., Kelbe, D., and van Leeuwen, M.(2015a). “Assessing the Impact of Sub-pixel Vegetation Structure on ImagingSpectroscopy Via Simulation”. In: Proceedings of SPIE. Vol. 9472. Baltimore,MD, USA, 94721K94721K-7. doi: 10.1117/12.2176992.
Yao, W., van Aardt, J., Romanczyk, P., Kelbe, D., van Leeuwen, M., andKampe, T. (2015c). “Constructing Virtual Forest Scenes for Assessment ofSub-pixel Vegetation Structure From Imaging Spectroscopy ”. In: AmericanGeophysical Union (AGU) Fall Meeting. San Francisco, CA, USA. December14-18, 2015.
2015-10-14 Wei Yao 2015 HyspIRI Science and Application Workshop 31
IntroductionMethods
ResultsConclusions/Outlook
Acknowledgements
This material is based upon work supported by the NASAHyspIRI Mission under Grant No. NNX12AQ24G.
Thanks to field team: Ashley Miller, Terence Nicholson,Claudia Paris, and Alexander Fafard
Thanks to Chris DeAngelis for building virtual scenes
Collaborators: Dr. Crystal B. Schaaf (UMB) andDr. Alan H. Strahler (BU)
Fine-scale airborne data were provided by the NEONAOP team; NEON is a project sponsored by the NationalScience Foundation and managed under cooperativeagreement by NEON, Inc.
2015-10-14 Wei Yao 2015 HyspIRI Science and Application Workshop 32
IntroductionMethods
ResultsConclusions/Outlook
Thanks!
Wei Yao ([email protected]), PhD student
Jan van Aardt ([email protected]), Adviser
2015-10-14 Wei Yao 2015 HyspIRI Science and Application Workshop 33