Challenges in Using Current Generation Solar
Reflectance Imager Observations for Climate
Change Detection and Future Directions
Steve Platnick 1
challenges: Rob Levy 1, Alexi Lyapustin 1, et al.
current imagers: Jim Butler 1, Jack Xiong 1
future imagers: Peter Pilewskie 2, Greg Kopp2, Kurt Thome 1
1 NASA
GSFC, 2 University of Colorado/LASP
SORCE STM
31 January 2014
Cocoa Beach, FL
Platnick et al., SORCE STM, Jan. 2014
Outline
1. Challenges
Examples of MODIS reflectometry trends aliasing into
geophysical product time series
Example stability requirements for trend detection
2. Current Generation Global Imagers
Overview of MODIS & VIIRS multispectral scanners
On-orbit reflectance calibration systems and approaches
3. Future Directions
Direct solar observations (SORCE/SRF/TRF legacy)
Hyperspectral pushbroom system examples
Platnick et al., SORCE STM, Jan. 2014
1. Challenges
MODIS/VIIRS
Current generation global imagers (MODIS/VIIRS) are
tasked to do a lot:
• Moderate resolution with wide swath global coverage
(MODIS: 2330 km, VIIRS: 3040 km)
scanners
• Multispectral coverage: VIS through IR (14.2 µm for MODIS)
• Meet a variety of science (aerosol, clouds, land, ocean
color) and operational needs
• On-orbit reflectometry accuracy on the order of 2-3% in the
VNIR-SWIR channels (MODIS: 20, VIIRS: 14). Stability not
specified for MODIS; 0.3% for VIIRS.
Platnick et al., SORCE STM, Jan. 2014
Example Trend Artifacts in MODIS Terra
Collection 5 Products
Platnick et al., SORCE STM, Jan. 2014
Trend Artifacts in MODIS Collection 5:
Aerosol Optical Depth Means over Land Surfaces
Aqua:JUL, 2002 to JUN, 2013 ; Terra: JUL, 2002 to JUN, 2013
AREA WEIGHTED = YES, PIXEL WEIGHTED = NO
Terra
Aqua
C5(Aqua & Terra) AOD zonal avg [60S, 60N]
= -0.001 per dec (abs)
AOD at 550nm (zonal avg ± 60°)
B
R
0.28
Terra
= -0.049 per dec
(abs)
MODIS Land scene
= -0.003 per dec (rel)
LAND
artifacts due to B3 (0.47 µm) degradation
B
R
= -0.267 per dec (rel)
Terra AOD
Aqua AOD
0.24
-0.3%/dec
0.20
0.16
0.12
-27%/dec
±60° latitude
Jan
2003
Jan
2004
Jan
2005
Jan
2006
Jan
2007
Jan
2008
Jan
2009
Year
Platnick et al., SORCE STM, Jan. 2014
Jan
2010
Jan
2011
Jan
2012
Jan
2013
Aqua:JUL, 2002 to JUN, 2013 ; Terra: JUL, 2002 to JUN, 2013
AREA WEIGHTED = YES, PIXEL WEIGHTED = YES
Terra
Aqua
C5(Aqua & Terra) (LIQUID) zonal avg [60S, 60N]
Cloud Optical Thicness Mean (± 60°)
22
Cloud Optical Thicness Mean (± 60°)
-0.052 per dec (rel)
per dec (rel)
Trend Artifacts in MODIS ==Collection
5:== -0.174
-0.043 per dec (rel)
-0.129 per dec (rel)
Liquid Water Cloud Optical Thickness Anomalies
13
B
R
B
R
Land COT ≈f (0.65µm)
Terra COT
Aqua COT
20
18
-5%/dec
-18%/dec
16
14
12
Ocean COT ≈f (0.86 µm)
12
11
-4%/dec
-13%/dec
10
9
Jan
2003
Jan
2004
Jan
2005
Jan
2006
Jan
2007
Jan
2008
Jan
2009
Jan
2010
Year
Platnick et al., SORCE STM, Jan. 2014
Monthly anomalies
Jan
2011
Jan
2012
Jan
2013
45
10.0
Trend Artifacts in MODIS Collection 5:
Cloud Optical Thickness Regional Trends
0
Time Series, Monthly :JUL, 2002 to JUN, 2010
-90
0
JUL, 2000 to JUN, 2001
COT Annual Mean (10° bins)
July 2000 – June 2001
90E
30.0
45
0
15.0
180
90W
-20.0
0
5% significance level
90
90
20.0
90
20.0
45
10.0
45
10.0
0
0.0
0
0.0
-45
-45
[ Annual mean ]
-45
-10.0
AQUA ! (C51) anomaly (LIQUID) TREND (TOTAL)
COT Trends
masked by p-value ≤ 0.05
(w/autocorrelation correction in t-test)
TERRA ! (C51) LIQUID
MODIS
Terra
[ %b0 (per decade) ]
-45
[ Annual mean ]
0.0
-10.0
-10.0
AQUA ! (C51) LIQUID
JUL, 2002 to JUN, 2001
-90
0
90E
180
0.0
90W
0
-90
-900
0
90E
90E
July 2002 – June 2003
30.0
30.0
45
45
0
0
15.0
15.0
-45
-45
90E
90E
0
180
180
90W
90W
15
0
0
0.0
0.0
30.0
30
0
15.0
-45
0.0
90E
180
90W
-20.0
-20.0
90
20.0
45
10.0
0
0.0
-10.0
-90
0
-20.0
90E
≤20
Platnick et al., SORCE STM, Jan. 2014
45
-90
0
0
0
-45
-90
-900
0
90
90W
90W
5% significance level
90
90
MODIS
Aqua
180
180
0
180
%/decade
0
90W
0
≥20
Trend Artifacts in MODIS Collection 5:
DATA: JUL,Optical
2002 to JUN,Thickness
2013
Liquid Water Cloud
Means
AREA WEIGHTED = YES, PIXEL WEIGHTED = YES
C5/C6(Terra) (LIQUID) zonal avg [60S, 60N]
Cloud Optical Thicness Mean (± 60°)
OCEAN
12
Ocean ±60° latitude
11
10
Ohring et al. 2005
(2%/dec stability)
L1B Collection 5
L1B Collection 6 test
9
Jan
2003
Jan
2004
Jan
2005
Jan
2006
Jan
2007
Jan
2008
Jan
2009
Year
Platnick et al., SORCE STM, Jan. 2014
Jan
2010
Jan
2011
Jan
2012
Jan
2013
Weatherhead Equation (constant = 3.3)
The Challenge
of Trend Detection
example : n = 16.3 yrs, with ! = 10%, and "=5%/dec
*
Number of Years Required to Detect a Trend
*
n (yrs)
required
to detect
the trend
(90% prob. of detecting
a trend
to a 0.05
statistical
level, no autocorrelation)
10
10
8
% trend/dec (%)
Trend/decade
~17 yrs
6
20
~30 yrs
4
30
2
40
5
10
50
0
0
20340050
5
10
!
15
20
y
Natural Variability in monthly residual
time series σy/⟨y⟩ (%)
Platnick et al., SORCE STM, Jan. 2014
90% prob of detec.ng means your t-­‐test must exceed 3.3 instead of 2.0
Note: image is for 3.3 approxima2on which should be fine for monthly anomalies
COT(rel.)/dec
The Challenge of Trend Detection
40 yrs
30
Wednesday, February 6, 13
20
10
from Platnick, S., S. A. Ackerman, B. A.
Baum, A. K. Heidinger, R. E. Holz, M. D.
King, W. Paul Menzel, S. Nasiri, El Weisz, P.
Yang: Assessment of IDPS VIIRS Cloud
Products and Recommendations for EOS-era
Cloud Climate Data Record Continuity, Suomi
NPP Science Team report, 2013 (http://
npp.gsfc.nasa.gov/teaminfo.html).
0
Fig. 1.2. Time required to have a 90% probability of detecting a trend to a statistical
significance of 0.05 if a 5%/decade change in cloud optical thickness was occurring in
each 10° grid box. Variability and correlation statistics needed for deriving time-todetection are from 12 years of MOD06 C5 cloud data.
Platnick et al., SORCE STM, Jan. 2014
Detection Time (yrs)
Fig. 1.1. Zonal mean decadal trend in cloud optical thickness (relative change/decade)
derived from simulator output for 5 CMIP5 models [Zelinka et al., 2013; private
communication]. Individual models are given by the gray lines; the inter-model mean is
given by the red line.
2. Current Imagers
Reflectometry Characterization Issues with
Current Generation Global Imagers (MODIS, VIIRS)
• Reflectometry derived from solar diffuser (SD) and solar diffuser
stability Monitor (SDSM) – ratio radiometer w/small integrating
sphere and Si detectors/filters
• Scan mirror angle of incidence varies across scan. Response vs.
Scan (RVS) angle must be characterized, including polarization
sensitivity (e.g., Terra mirror side differences in VIS)
• Two-sided mirror (MODIS scan mirror, VIIRS rotating telescope w/
Half Angle Mirror)
• Along-track linear detector arrays (~10km)
• Use of lunar observations for stability monitoring (most important
for climate)
Platnick et al., SORCE STM, Jan. 2014
MODIS%On8orbit%Calibra<on%Schema<c%
Solar%diffuser%(SD)%and%solar%
diffuser%stability%monitor%
(SDSM)%for%reflec<ve%solar%
bands%(RSB)%calibra<on%
Spectroradiometric%Calibra<on%
Assembly%(SRCA)%for%instrument%
spectral%and%spa<al%characteriza<on%
Solar
Diffuser
SRCA
SDSM
Blackbody
Scan%%
Mirror%
AOI%:%
MODIS
Scan Mirror
EV%:%10.5865.5⁰%
Angle
of Incidence
Nadir:%38⁰%
Earth View:
10.5-65.5°
SD%:%50.2⁰%
Nadir: Moon:%11.2⁰%
38°
Diffuser: 50°
Moon: 11°
Blackbody%(BB)%for%
thermal%emissive%bands%
(TEB)%calibra<on%
Space
View
VIIRS Scan Mirror
Angle of Incidence
Earth View: 29-56.5°
Nadir: 36°
Diffuser: 60°
Moon: 60°
Lunar%views%
through%SV%port%
Platnick et al., SORCE STM, Jan. 2014
Solar Diffuser Characterization Challenges
• Degradation: Spectralon (scintered/pressed Teflon) degrades on-orbit w/
•
•
BRDF, sr-1
•
Spectralon BRDF at 60 deg Incidence
exposure time (short wavelengths in particular)
0.390
300 nm
Exposure: Diffuser door reduces exposure
on MODIS (though Terra
400 nm
0.370
500 nm
door is stuck open). No door on VIIRS.
600 nm
0.350
700 nm
Diffuser screens: reduce incident sunlight,
800 nmbut can cause interference
0.330
patterns and scattered light (less of issue for VIIRS w/elliptical holes) 6%
0.310
Uncertainty in spectral BRDF of diffuser and BTDF of a solar diffuser
0.290
screen
VIIRS
MODIS
VIIRS
0.270
Current NIST BRDF measurement uncertainties
at
MODIS/VIIRS
SDSM
MODIS
SDSM
0.250
viewing angles gives an uncertainty
-70 -60 of
-50 0.57%
-40 -30 -20(k=2)
-10 0 for
10 a
20 Spectralon
30 40 50 60 70
Viewing angle, deg
diffuser.
Inconsistent viewing angle: SDSM on MODIS doesn’t view diffuser at
same angle as MODIS instrument.
Assump2on made that SD reflectance degrada2on will be isotropic Platnick et al., SORCE STM, Jan. 2014
VIIRS and MODIS Solar Diffuser (SD) Degradation
VIIRS SD Degrada.on vs. Time
Similar to MODIS (as a func4on of solar exposure 4me)
MODIS & VIIRS SD Degrada.on
vs. Spectral Channel
14 yrs
11.5 yrs
2 yrs
Platnick et al., SORCE STM, Jan. 2014
Collection 6 Corrections Implemented in MODIS L1B
for Problem Channels
Terra & Aqua MODIS Band 2 (0.86 µm)
11-yr
Time
of time
CEOS
Desert
Sites
MODIS band 2 (0.86
µm) Record
reflectance
record
at CEOS
desert sites
Terra
Aqua
5%
Collection 5
Collection 5
days since Jan 2000
days since Jan 2002
different sensor view angles
Platnick et al., SORCE STM, Jan. 2014
Collection 6 Corrections Implemented in MODIS L1B
for Problem Channels
Terra MODIS band 2 (0.86 µm) 11-yr Time Record of Desert Sites
Terra
Terra
Collection 6
tentative correction using
desert sites
Collection 5
days since Jan 2000
days since Jan 2000
different sensor view angles
Platnick et al., SORCE STM, Jan. 2014
Collection 6 Corrections Implemented in MODIS L1B
for Problem Channels
13.5 yr degradation (%)
Terra MODIS Response vs. View Angle
C5 method
C6 method
Frame # (view angle)
Lunar
View
Platnick et al., SORCE STM, Jan. 2014
Solar
Diffuser
MODIS Band 1 (0.65 µm) and 2 (0.86 µm) vs. vs. VIIRS SNOs
VIIRS IDPS SDR (operational LUTs)
MODIS Collection 6 L1B
Platnick et al., SORCE STM, Jan. 2014
Lunar Observations
VIIRS Lunar Roll Maneuver
(view from the Sun)
ASTER&560nm&Band&
MODIS&645.5nm&Band&
MISR&672nm&Band&
17&
Platnick et al., SORCE STM, Jan. 2014
Lunar Observations: State of the Art
• Stability Monitoring. ROLO model (USGS, T. Stone): spectral
irradiance vs. lunar phase/libration from 8+ years of data at Flagstaff
facility
Relative precision ~1% over full valid phase angle range (eclipse to
90 degrees before/after Full Moon), from 350–2300 nm.
Stability demonstrated to ~ 0.13% (SeaWiFS, Eplee et al., Appl. Opt.,
2012) with small range of phase angles using pitch maneuvers.
Part of standard data processing stream for SeaWiFS, MODIS T&A,
VIIRS. Used in cal/val of Landsat-8 OLI, EO-1 ALI, EO-1 Hyperion,
geostationary meteorological imagers (GOES-7 thru GOES-15, etc.)
• Absolute Irradiance
ROLO irradiance accuracy ~ 5-10%
NIST lunar Irradiance Program (Claire Cramer et al., CLARREO STM
2014): 2 datasets from CAS spectrometer at Mt. Hopkins, 1 % (k=2)
from 500–920 nm (ROLO for variation in brightness during obs).
Setting up facility at Mauna Loa (goal of k=2 0.5 % from 380–980
nm). Working on a high-altitude flight campaign, and extending
Platnick et al., SORCE STM, Jan. 2014
3. Future
Future Directions
• Improved scanner system characterization (some combination of
cost, risk, and/or technical issues)
Improvements in diffuser material and screens (?)
Include diffuser doors w/design opening to limit scattered light
Optimizing instrument optical layout to view the diffuser and improved
diffuser monitoring
Better ground cal facilities, e.g., improved standards for diffuser
BRDF ground cal of and traceability into orbit
Separate Solar Reflectance and IR scanners to optimize optics
(Landsat), polarization scramblers (SeaWiFS), etc.
• Reflectometry derived from direct solar views
CLARREO pushbroom concepts (not wide swath): HySICS (LASP),
SOLARIS (GSFC)
• Hyperspectral coverage through the VNIR/SWIR (~5-10 nm) for
improved information content
Platnick et al., SORCE STM, Jan. 2014
SOLARIS Instrument
Offner system covering 320–2300 nm,
500 m GIFOV, 100 km swath width
Technology demonstration:
• Thermal control of
attenuators/detector
• Design/production of optics
• Depolarizer technology
Lunar data provide calibration verification
Benchmark reflectance from ratio
of earth view to measurements of
irradiance while viewing the sun
Cross-calibration of other imagers
K. Thome, et al.
Platnick et al., SORCE STM, Jan. 2014
SOlar, Lunar for Absolute Reflectance Imaging Spectroradiometer (SOLARIS)
HySICS on ECHO Platform (Earth Venture proposal)
Aperture&wheel&
Alignment&Cube&
Enclosure&door/&
drive&assembly&
Filter&wheel&
Fine&Sun&Sensor&
Offner&mirrors&
Four&mirror&
anas-gmat&
telescope&
Fold&mirror&
Gra-ng&
Instrument&&
enclosure&
Detector&
Cryo8cooler&
Detector&
vacuum&
enclosure&
Platnick et al., SORCE STM, Jan. 2014
HySICS: HyperSpectral Imager for Climate Science
P. Pilewskie et al.
A"enua'on)Methods)U'lized)by)HySICS)
•  Aperture'a)enua+on!–!Reduc(on!of!
input!light!collec(ng!area!
Earth!Viewing!
Solar!Viewing!
–  Can!achieve!a6enua(ons!~10;3!
–  Limited!by!diffrac(on!
•  Integra+on'+me'a)enua+on!–!
Reduc(on!of!light!collec(ng!(me!
–  Can!achieve!a6enua(ons!~10;2!
–  Limited!by!electronics!
Input!Aperture!
•  Filter'a)enua+on!–!spectral!filters!
calibrated!with!on;orbit!lunar!views!
Can!achieve!a6enua(ons!10;1!
–
–  Limited!by!S/N!
All)a"enua'on)methods)are)rela've)
measurements;)direct)measurements)of)
solar)or)Earth)irradiances)not)required.)
Filter!
Hyper;
spectral!
Imager!
Exposure!(me!
FPA!
(me!
Integra(on!
Time!
Hyper;
spectral!
Imager!
FPA!
Exposure!(me!
(me!
G. Kopp et al., IIP
Platnick et al., SORCE STM, Jan. 2014
A:enua4on Uncertain4es (Very Preliminary – Not End-­‐to-­‐End)
Actual'system'apertures'
Platnick et al., SORCE STM, Jan. 2014
G. Kopp, et al., IIP
Hyperspectral VNIR/SWIR Information Content
Cloud detection and thermodynamic phase discrimination over Arctic snow
REFF [µm]
Probability
95%
66%
liquid
over snow
Wavelength [µm]
2.3
What is retrieval skill for (1) cloud detection,
(2) cloud phase discrimination, (3) COD, REF,
LWP/IWP retrieval for:
(a) MODIS (b) spectral imager?
Use generalized retrieval analysis (GENRA,
Coddington et al., 2012 to ingest spectral
information sequentially, given measurement
and model uncertainty (including snow
surface albedo variability).
cloud free
COD
(b) spectral imager
REFF [µm]
100 0
50%
Cloud
COD=5,
REF=25 µm
over ocean
Probability [%]
0.6
ice
ice
66%
95%
50%
liquid
cloud free
COD
0
0
Reflectance
measured snow
albedo + variability
(a) MODIS
Probability [%]
1
SW imagery: Also low cloud/surface contrast for λ<900 nm, but can use several
λ>900 nm where snow is ‘dark’
100
Motivation: Clouds < 5 km are important for Arctic surface energy budget, water-ice-cloud feedback processes
Problem: IR imagery has low skill detecting clouds due to low cloud/surface temperature contrast
0.6
Wavelength [µm]
2.3
S. Schmidt et al.
Platnick et al., SORCE STM, Jan. 2014
Summary
1. Current generation imagers (MODIS/VIIRS) have a
number of challenging technical requirements: moderate
spatial resolution and wide swath, VIS to IR spectral
coverage, variety of science and operational users.
2. While MODIS/VIIRS instrument characterization teams
have/are doing an excellent job of accounting for
degradation issues, it has not been demonstrated that
current global imagers can be characterized to the
stability/accuracy requirements needed for multi-decadal
trend detection of important geophysical datasets.
3. Requirements for climate studies by imagers requires new
technical capabilities/approaches.
Platnick et al., SORCE STM, Jan. 2014