Page 1
Jet Propulsion LaboratoryCalifornia Institute of Technology
David R. Thompson, [email protected] !Joseph W. Boardman*!Michael Eastwood!Robert O. Green!Justin Haag!Pantazis Mouroulis!Byron Van Gorp!!
Jet Propulsion Laboratory, California Institute of Technology!*Analytical Imaging and Geophysics, Inc. Boulder, CO!!Copyright 2017 California Institute of Technology. All Rights Reserved. US Government Support Acknowledged.!!!
Imaging Spectrometer Stray Spectral Response: In-Flight Characterization and Correction!
11/1/17! [email protected] ! 1
Page 2
Motivation!• Non-Gaussian tails of spectral response functions
can be difficult to characterize in the laboratory!• Calibration can shift during deployment!• Small SSRF contributions can damage downstream
atmospheric correction!• In-flight techniques are useful for validating and
updating laboratory measurements. !
11/1/[email protected] ! 2
-20 -15 -10 -5 0 5 10 15 20Channel
10-4
10-3
10-2
10-1
Res
pons
e
Nominal SRFStray SRFActual SRF
Page 3
Method!
11/1/[email protected] ! 3
• Sequential estimation of Nominal and Stray SRF parameters. !
• Exploit predictable changes in the shape of the A band across varying surface elevation.!
• Diverse scene content provides numerical leverage to characterize spectral response tails!
Page 4
Estimation of nominal SRF[Thompson et al., Atmos. Meas. Tech 2015]!
Optimize a wavelength shift to match high-contrast atmospheric absorption features!
1/16/2017!AVIRIS-NG PSFs / [email protected] ! 4
Page 5
Estimation of nominal SRF[Thompson et al., Atmos. Meas. Tech 2015]!
1/16/2017!AVIRIS-NG PSFs / [email protected] ! 5
-0.0003!
-0.0002!
-0.0001!
0!
0.0001!
0.0002!
0.0003!
0! 100! 200! 300! 400! 500! 600! 700!
Wav
elen
gth
shift
(mic
ron)!
Sample!
Page 6
Stray SRF Measurement model Adapted from [Zhong et al., 2006]!
1/16/2017!AVIRIS-NG PSFs / [email protected] ! 6
StrayRadiance
NominalRadiance
MeasurementNoise
MeasuredRadiance
Page 7
Stray SRF Measurement model Adapted from [Zhong et al., 2006!
1/16/2017!AVIRIS-NG PSFs / [email protected] ! 7
StrayRadiance
NominalRadiance
MeasurementNoise
MeasuredRadiance
GHLA HLA 𝜖LMRadianceatsensorNominalSRFStraySRF
Page 8
1/16/2017!AVIRIS-NG PSFs / [email protected] ! 8
StrayRadiance
NominalRadiance
MeasurementNoise
MeasuredRadiance
RadianceatsensorNominalSRFStraySRF
Stray SRF Measurement model Adapted from [Zhong et al., 2006]!
Page 9
A Linear SRF Correction Matrix!
Calculate a Moore-Penrose Pseudoinverse:!!!!This estimates the nominal SRF:!!!!!!A similar correction fixes cross-track stray light !
1/16/2017!AVIRIS-NG PSFs / [email protected] ! 9
CorrectedRadiance
Correc=onmatrix
DistortedMeasurement
Page 10
Retrieve Stray SRF from a “Calibration Scene”!Death Valley Transect, 2014 (visible RGB)!
11/1/[email protected] ! 10
Predict A band radiances using a Digital Elevation Model!
Nonlinear least squares optimization finds SSRF parameters!
Page 11
Estimation accuracy for Gaussian SSRF (simulated)!
11/1/[email protected] ! 11
0.02 0.04 0.06 0.08 0.1�
4
5
6
7
Gau
ssia
n �
SNR 400SNR 200
(Straylightfrac=on)
Page 12
Estimation accuracy for Lorentz SSRF (simulated)!
11/1/[email protected] ! 12
0.02 0.04 0.06 0.08 0.1�
4.71
5.8875
7.065
8.2425
Lore
ntz
half
wid
th
SNR 400SNR 200
(Straylightfrac=on)
Page 13
Fit error for candidate SSRF shapes!
1/16/2017!AVIRIS-NG PSFs / [email protected] ! 13
bestfit
-30 -20 -10 0 10 20 30Channel
10-6
10-4
10-2
100
Response
gaussianlorentzpareto
Page 14
Improvement in O2 A band fit!
11/1/[email protected] ! 14
745 750 755 760 765 770 775Wavelength (nm)
3
4
5
6R
adia
nce
(uW
nm
-1 s
r-1 c
m-2
)
NominalMeasuredCorrected
745 750 755 760 765 770 775Wavelength (nm, with offset for clarity)
0
0.01
0.02
0.03
0.04
0.05
Rad
ianc
e sq
uare
d er
ror
MeasuredCorrected
Page 15
Correction fixes a bias in pressure altitude estimates!
11/1/[email protected] ! 15
0 0.5 1 1.5 2 2.5 3Elevation, km ASL
0
0.5
1
1.5
2
2.5
3
Estim
ated
Pre
ssur
e Al
titud
e, k
m A
SL
1.372
0 0.5 1 1.5 2 2.5 3Elevation, km ASL
0
0.5
1
1.5
2
2.5
3
Estim
ated
Pre
ssur
e Al
titud
e, k
m A
SL0.721
Beforecorrec6on A8ercorrec6on
Ivanpahvalida6onsite
Page 16
Reflectance validation!
400 600 800 1000 1200 1400 1600 1800 2000 2200 2400Wavelength (nm)
0
0.1
0.2
0.3
0.4
Ref
lect
ance
400 600 800 1000 1200 1400 1600 1800 2000 2200 2400Wavelength (nm)
0
0.1
0.2
0.3
0.4
Ref
lect
ance
950 1000 1050 1100 1150 1200Wavelength (nm)
0.28
0.3
0.32
0.34
Ref
lect
ance
950 1000 1050 1100 1150 1200Wavelength (nm)
0.25
0.3
0.35
0.4
Ref
lect
ance
test interval
Reference interval
q(x) = 0.0045
q(x) = 0.0032
11/1/[email protected] ! 16
Reflectancequalitymetric:
Page 17
India Validation Results!
11/1/[email protected] ! 17
• 26of37flightdaysshowsignificantimprovements(p<0.001)• Typicalimprovementis20-35%• Noflightdayshowsasta=s=callysignificantaccuracyreduc=on
Frac=onalimprovementfor277scenes
Page 18
Agreement with laboratory data!
[email protected] ! 18
-10 -8 -6 -4 -2 0 2 4 6 8 10Channel
10-4
10-2
100
Relat
ive re
spon
se
Laboratory measurementAtmospheric fitNominal SRF
11/1/17!
Page 19
Spatial dimension!
1/16/2017!AVIRIS-NG PSFs / [email protected] ! 19
• ExploitNear-Infrared(NIR)oceanreflectance• Useahaze-freedaytoconstrainpathradianceandadjacencyeffects• Useawind-freedaywithnadirobserva=onstolimitglint• DarkwatershouldbehighlyabsorbantinNIR• Dataset:2015Greenlandiceflow
500 1000 1500 2000 2500Wavelength (nm)
0
0.1
0.2
0.3
0.4
0.5
0.6
Ref
lect
ance
Sea!
Ice!
Page 20
“Halo” reduction!
1/16/2017!AVIRIS-NG PSFs / [email protected] ! 20
Original RGB! 612 nm, equalization stretch!(0-3 uW nm-1 sr-1 cm-2)!
612 nm, after CRF correction!
Page 21
Discussion!• Can leverage scene invariant properties to fit PSFs!• Some advantages to using separable functions!
– Numerical stability, fairly easy to prevent ringing & overcorrection!– Can model CRF or SRF or both, and fit them independently!
• Positive results on held-out validations!– Appears to fix our pressure altitude bias!– Improves H2O residuals!– Improves spatial halos!
• Implemented in latest India release, and all AVIRIS-NG datasets starting from 2016!
1/16/2017!AVIRIS-NG PSFs / [email protected] ! 21
Page 22
Thanks!!NASA Earth Science!The AVIRIS-NG Team, including Sarah Lundeen, Brian Bue, Winston Olson-Duvall, Ian McCubbin, Mark Helmlinger, and others!
11/1/[email protected] ! 22