Stching)aMODIS/VIIRS)me)series)of)aerosol) …...N = 80591; R = 0.890 Bias = 0.004, RMSE = 0.103 Y =...
Transcript of Stching)aMODIS/VIIRS)me)series)of)aerosol) …...N = 80591; R = 0.890 Bias = 0.004, RMSE = 0.103 Y =...
S"tching a MODIS-‐VIIRS "me series of aerosol proper"es using the Dark Target algorithm:
Circa 2016 Robert C. Levy (NASA-‐GSFC)
Shana MaNoo, Virginia Sawyer* and Richard Kleidman (SSAI/GSFC) Falguni Patadia and Yaping Zhou* (Morgan State U / GSFC)
Pawan Gupta and Yingxi Shi* (USRA/GSFC) Lorraine Remer (UMBC/JCET), Robert Holz (SSEC/UWisconsin)
* New people in 2016.
And many, many, many others
CERES mee"ng; April 2016 @ NASA LaRC
http://earthobservatory.nasa.gov/
Aerosol from space
Smoke transported over Eastern Canada/USA (8 July 2002)
! Aerosol op"cal depth (AOD or τ) ! “Essen"al Climate Variable” (ECV)
! Requires accuracy <±0.02 ! Measured over mul"-‐decades
! Yet, mostly a “regional” problem. ! Required uncertainty (per pixel) = <15%. ! Desire: separa"on of aerosol types and effects
Outline
1. “Dark-‐target” (DT) remote sensing on MODIS 2. Terra vs Aqua (and calibra"on and trends) 3. DT applied to VIIRS (using Wisconsin IFF) 4. Challenges of MODIS"VIIRS con"nuity 5. Advancing the DT algorithm 6. Summary
MODIS on Terra and Aqua
Orbit: 705 km, sun-‐synchronous, over same point every 16 days Equator crossing: 10:30 (Terra, since 2000), 13:30 (Aqua, since 2002)
Swath: 2330 km (55° cross track) Spectral Range: 0.4-‐14.4µm (36 bands). 19 in solar spectrum (< 4.0 µm) Spa6al Resolu6on: 250m (2 bands) 500m (5 bands) 1000m (29 bands) Calibra6on: On-‐board and con"nuously updated
Moderate resolu"on Imaging Spectroradiometer
Twin MODIS instruments – Two views per day!
Terra (10:30, Descending) Aqua (13:30, Ascending)
Complicated TOA Signal
Mul"ple Reflec"on
Gas + Aerosol scaNering (path radiance)
Indirect Transmission (adjacency effect)
Direct Transmission
Contributions from: • Gas absorption (O3, CO2, etc) • H2O absorption • Rayleigh (molecular) scattering • Aerosol scattering and absorption • Surface reflection • Atmosphere / Surface interaction • Contamination from neighboring pixels (clouds, etc) • .. And cloud masks
clouds (%@(*%@!)
Aerosol retrieval from MODIS
OCEAN
GLINT
LAND
May 4, 2001; 13:25 UTC Level 2 “product”
AOD 1.0
0.0
May 4, 2001; 13:25 UTC Level 1 “reflectance”
What MODIS observes ANributed to aerosol (AOD)
There are many different “algorithms” to retrieve aerosol from MODIS Ours is Dark Target (DT); “Established 1997” by Kaufman, Tanré, Remer, etc)
6
Separate algorithms: Ocean and Land Both are mul6-‐channel inversions
Products = AOD at 0.55 µm, spectral AOD, diagnos6cs
MODIS Collec"on 6 (10 km product): “Validated since 2014”
All assump"ons related to assumed aerosol proper"es, surface reflectance, lookup tables, and cloud masks were updated for C6
7
Collec"on 6 “Webinars”: hNp://aerocenter.gsfc.nasa.gov/ext/registra"on/ “dark-‐target” website: hNp://darktarget.gsfc.nasa.gov MODIS product website: hNp://modis-‐atmos.gsfc.nasa.gov
C6 Land, Aqua, Mar 2003−Feb 2013
0.0 0.5 1.0 1.5 2.0AERONET 0.55 µm AOD
0.0
0.5
1.0
1.5
2.0
MO
DIS
0.5
5 µ
m A
OD
% within EE = 70.78% above EE = 15.85% below EE = 13.38N = 80591; R = 0.890Bias = 0.004, RMSE = 0.103Y = 1.014x−0.002
C6 Land, Aqua, Mar 2003−Feb 2013
0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4AERONET 0.55 µm AOD
−0.4
−0.2
0.0
0.2
0.4
MO
DIS−A
ER
ON
ET
0.55
µm
AO
D
1 3 10 30 100 300 1000
Frequency
Land: EE = ±(0.05 + 15%)
Two validated MODIS "me series: Do they represent the same world?
Terra (since spring 2000) Aqua (since summer 2002)
• Same instrument hardware (op"cal design) • Same spa"al and temporal sampling resolu"on • Same calibra"on/processing teams • Same aerosol retrieval algorithms • The two MODIS instruments are Iden"cal twins!
How do they behave????? 8
Aerosol Trends: If based on Collec"on 5 (C5)
• Consider that a trend of ±0.01/decade is significant • In C5, over land, Terra decreased (-‐0.05/decade) while Aqua was constant • Terra / Aqua divergence was similar everywhere on the globe! Not AM/PM • " Like iden"cal twins, the twin MODIS sensors have aged differently.
• MCST applied a new calibra"on for C6, based on observing psuedo-‐invariant desert targets
• Terra/Aqua divergence “mostly” removed for C6 • Terra AOD high by 0.027 land/0.017 ocean (13%), Global! • Residual trending (Terra-‐Aqua increasing by ~0.01/decade) • Bigger-‐amplitude seasonal cycle to Terra-‐Aqua azer 2011.
10
Global monthly median 0.47 µm AOD, Ocean
2000 2002 2004 2006 2008 2010 2012 20140.00
0.05
0.10
0.15
0.20
AO
D
−0.020.00
0.02
0.04
0.06
0.080.10
Te
rra
Min
us
Aq
ua
AO
D
Global monthly median 0.55 µm AOD, Ocean
2000 2002 2004 2006 2008 2010 2012 20140.00
0.05
0.10
0.15
0.20
AO
D
−0.020.00
0.02
0.04
0.06
0.080.10
Te
rra
Min
us
Aq
ua
AO
D
Global monthly median 0.67 µm AOD, Ocean
2000 2002 2004 2006 2008 2010 2012 20140.00
0.05
0.10
0.15
AO
D
−0.020.00
0.02
0.04
0.06
0.080.10
Te
rra
Min
us
Aq
ua
AO
D
Global monthly median 0.85 µm AOD, Ocean
2000 2002 2004 2006 2008 2010 2012 20140.000.02
0.04
0.06
0.08
0.100.12
AO
D
−0.020.00
0.02
0.04
0.06
0.080.10
Te
rra
Min
us
Aq
ua
AO
D
Global monthly median 1.24 µm AOD, Ocean
2000 2002 2004 2006 2008 2010 2012 20140.00
0.02
0.04
0.06
0.08
0.10
AO
D
−0.020.00
0.02
0.04
0.06
0.080.10
Te
rra
Min
us
Aq
ua
AO
D
Global monthly median 1.64 µm AOD, Ocean
2000 2002 2004 2006 2008 2010 2012 20140.00
0.02
0.04
0.06
0.08
0.10
AO
D
−0.020.00
0.02
0.04
0.06
0.080.10
Te
rra
Min
us
Aq
ua
AO
D
Global monthly median 2.1 µm AOD, Ocean
2000 2002 2004 2006 2008 2010 2012 20140.00
0.02
0.04
0.06
0.08
AO
D
−0.020.00
0.02
0.04
0.06
0.080.10
Te
rra
Min
us
Aq
ua
AO
D
Terra Aqua Terra−Aqua
Global monthly median 0.47 µm AOD, Land
2000 2002 2004 2006 2008 2010 2012 20140.0
0.1
0.2
0.3
0.4
AO
D
0.00
0.05
0.10
0.15
0.20
Terr
a M
inu
s A
qu
a A
OD
Global monthly median 0.55 µm AOD, Land
2000 2002 2004 2006 2008 2010 2012 20140.0
0.1
0.2
0.3
AO
D
0.00
0.05
0.10
0.15
0.20
Terr
a M
inu
s A
qu
a A
OD
Global monthly median 0.67 µm AOD, Land
2000 2002 2004 2006 2008 2010 2012 20140.000.050.100.150.200.250.30
AO
D
0.00
0.05
0.10
0.15
0.20
Terr
a M
inu
s A
qu
a A
OD
Terra Aqua Terra−Aqua
Terra Aqua
Terra-‐Aqua
C6 AOD: Terra versus Aqua LAND
OCEAN
Annual Mean AOD, Land
2004 2006 2008 2010 2012 20140.10
0.15
0.20
0.25
0.30
AO
D
AOD trend/decade = 0.011AOD trend/decade = −0.002
TerraAqua
Terra−Aqua, Land
2004 2006 2008 2010 2012 20140.00
0.01
0.02
0.03
0.04
AO
D D
iffer
ence
AOD difference trend/decade = 0.013
Annual Mean AOD, Ocean
2004 2006 2008 2010 2012 20140.10
0.12
0.14
0.16
AO
D
AOD trend/decade = 0.012AOD trend/decade = 0.004
Terra−Aqua, Ocean
2004 2006 2008 2010 2012 20140.000
0.005
0.010
0.015
0.020
0.025
AO
D D
iffer
ence
AOD difference trend/decade = 0.007
0.017
0.027
MODIS C6 (and C6+) • Trending issues reduced with C6 product, but:
– S"ll significant offsets (13%) and – S"ll residual co-‐trending (<0.01 / decade)
• Why? Sampling? diurnal cycles? Cloud masking? • Calibra"on?
– Test different op"ons – “C6+” of Alexei Lypus"n et al., – Ocean vicarious correc"ons – Dave Doelling’s one – Me, playing on my own. – Etc.
• Yet, overall convergence
11
June 2013, land grid cells
−0.15 −0.10 −0.05 −0.00 0.05 0.10 0.15Terra−Aqua AOD
0
200
400
600
800
1000
1200
Freq
uenc
y
C6C6+
June 2013, ocean grid cells
−0.15 −0.10 −0.05 −0.00 0.05 0.10 0.15Terra−Aqua AOD
0
1000
2000
3000
4000
Freq
uenc
y
Munchak et al., (in prep)
Beyond MODIS
12
• Terra is driving in Virginia (16!) • Aqua already celebrated its “Sat-‐mitzvah” (13). • Both have well-‐exceeded their planned mission life"mes • Calibra"on con"nues to get trickier, and there are end-‐of-‐life"me plans How do we make AOD climate data record? (20+ years of global AOD)?
VIIRS? Visible-‐Infrared Imager Radiometer Suite
aboard Suomi-‐NPP (and future JPSS)
VIIRS versus MODIS Orbit: 825 km (vs 705 km), sun-‐synchronous, over same point every 16 days
Equator crossing: 13:30 on Suomi-‐NPP, since 2012 (vs on Aqua since 2002) Swath: 3050 km (vs 2030 km) Spectral Range: 0.412-‐12.2µm (22 bands versus 36 bands) Spa6al Resolu6on: 375m (5 bands) 750m (17 bands): versus 250m/500m/1km Aerosol retrieval algorithms: “Physics” similar, but different strategies Wavelength bands (nm) / DT aerosol retrieval: 482 (466), 551 (553) 671 (645), 861 (855),
2257 (2113) " differences in Rayleigh op"cal depth, surface op"cs, gas absorp"on.
MODIS-‐Aqua – 29 May 2013 VIIRS-‐SNPP – 29 May 2013
Already a “validated” NOAA-‐based aerosol product
14 Qualita"vely similar, yet quan"ta"vely different. Especially over land Cannot be used for con"nua"on of MODIS.
Solu"on? Port the DT algorithm!
15
• We use Intermediate File Formats (IFF) and tools developed at the “Atmosphere-‐SIPS”, at the University of Wisconsin
• Results of MODIS-‐like on VIIRS include: • Reduced global AOD differences and more similar global sampling • Now a systema"c bias over ocean (VIIRS high by 20%). • Déjà vu? Terra versus Aqua? (Terra high by 13%) • " VIIRS also needs calibra"on study?
MODIS MODIS-‐like on VIIRS Difference M -‐ V
0
20
40
60
80
100
% R
efle
ctan
ce
DeciduousConiferSeawater
MODISVIIRSOverlap
0.5 1.0 1.5 2.0Wavelength [µm]
Wavelength bands & surface spectra
Levy et al., 2015
Comparing to AERONET and calibra"on
16
MODIS Collection 6, Ocean
0.0 0.2 0.4 0.6 0.8 1.0AERONET 0.55 µm AOD
0.0
0.2
0.4
0.6
0.8
1.0
Sate
llite
0.5
5 µ
m A
OD
% within EE = 75.95% above EE = 18.69% below EE = 5.36N = 1418; R2 = 0.884Y = 0.983x + 0.020Bias = 0.024RMSE = 0.073
MODIS Collection 6, Ocean
0.0 0.2 0.4 0.6 0.8AERONET 0.55 µm AOD
−0.4
−0.2
0.0
0.2
0.4
Sate
llite−A
ERO
NET
0.5
5 µ
m A
OD
Y = −0.009x + 0.019
MODIS−like MODIS, Ocean
0.0 0.2 0.4 0.6 0.8 1.0AERONET 0.55 µm AOD
0.0
0.2
0.4
0.6
0.8
1.0
Sate
llite
0.5
5 µ
m A
OD
% within EE = 77.27% above EE = 17.01% below EE = 5.72N = 1399; R2 = 0.896Y = 0.986x + 0.018Bias = 0.021RMSE = 0.070
MODIS−like MODIS, Ocean
0.0 0.2 0.4 0.6 0.8AERONET 0.55 µm AOD
−0.4
−0.2
0.0
0.2
0.4
Sate
llite−A
ERO
NET
0.5
5 µ
m A
OD
Y = −0.003x + 0.017
MODIS−like VIIRS, Ocean
0.0 0.2 0.4 0.6 0.8 1.0AERONET 0.55 µm AOD
0.0
0.2
0.4
0.6
0.8
1.0
Sate
llite
0.5
5 µ
m A
OD
% within EE = 65.74% above EE = 32.30% below EE = 1.96N = 2297; R2 = 0.915Y = 1.170x + 0.015Bias = 0.044RMSE = 0.082
MODIS−like VIIRS, Ocean
0.0 0.2 0.4 0.6 0.8AERONET 0.55 µm AOD
−0.4
−0.2
0.0
0.2
0.4
Sate
llite−A
ERO
NET
0.5
5 µ
m A
OD
Y = 0.170x + 0.015
VIIRS EDR, Ocean
0.0 0.2 0.4 0.6 0.8 1.0AERONET 0.55 µm AOD
0.0
0.2
0.4
0.6
0.8
1.0
Sate
llite
0.5
5 µ
m A
OD
% within EE = 69.47% above EE = 26.32% below EE = 4.21N = 2139; R2 = 0.887Y = 1.040x + 0.024Bias = 0.033RMSE = 0.064
VIIRS EDR, Ocean
0.0 0.2 0.4 0.6 0.8AERONET 0.55 µm AOD
−0.4
−0.2
0.0
0.2
0.4
Sate
llite−A
ERO
NET
0.5
5 µ
m A
OD
Y = 0.040x + 0.024
1 2 3 4 5 6 7 8 9 10
Frequency
MODIS-‐like on VIIRS has great correla"on but 1.17 slope! Studies such as Uprety et al., (2013) do radiometric comparisons between VIIRS and MODIS and find that VIIRS may be 2% high in some bands. 2% high bias is sufficient to give a 1.17 slope over ocean without the adding same bias to land.
0.856 or 0.861 Reflectance % Difference Reflectance
MODIS: 0.856 um Reflect VIIRS: 0.861 um Reflect VIIRS – MODIS Reflect
Calibra"on: Match files
Cloud Optical Properties: Granule Example
20 April 2016IRS16, Platnick et al.
6 July 2014
common view zenith & scattering angle
“common” geometry/angles
• Can we “prove” calibra"on differences? It’s hard! • Slight differences in orbit " no true matches inside ±70° la"tude • Common geometry is very limited • University of Wisconsin is crea"ng “match” files for us to look at
From Steve Platnick
Cloud Optical Properties: 0.86 µm Channel Radiometry
20 April 2016IRS16, Platnick et al.
Spectral Response FunctionsMODIS B2
VIIRS ~3-4% more reflective than expected
VII
RS
M7
VII
RS
CO
T
MODIS COT MODIS COT
MO
DIS
CO
T w
/b2
+3
% b
ias
RetrievedVIIRS vs MODIS COT scatterplot
MODIS COT w/3% increase in reflectance vs. baseline MODIS COT
Calibra"on: Wavelength issues • Can we “prove” calibra"on differences? It’s hard!
• Slight differences in wavelength " no true matches • Slight differences in Rayleigh op"cal depths, • Some"mes major differences in gas absorp"ons • With of lack of true spa"al overlap, hard to find common points .
From Steve Platnick
Calibra"on: Timing issues • Can we “prove” calibra"on differences? It’s hard!
• Drizing orbit "mes " • Tolerance for “matches” vary • With of variety of "me overlap, hard to find common points .
Equatorial local solar crossing times, ascending node
2012 2013 2014 2015 2016 2017Year
13:20
13:25
13:30
13:35
13:40
Cro
ssin
g t
ime,
UT
C
S-NPP
Aqua
Plot drawn by Andy Sayer (GSFC), source data from Greg Quinn at SSEC Wisconsin.
MODIS – VIIRS overlap with the IFF
• 2012-‐2015. • Ocean: Consistent offset = 0.03 (20%) with spikes in summer
• Land: Average offset is near zero, but seasonal dependence
20
MODIS (Aqua): MAM 2013
What is good enough? • Convergence: of gridded (Level 3 –like) data
– For a day? A month? A season? – What % of grid boxes must be different by less than X?
• in AOD? In Angstrom Exponent? Size parameters?
• Valida"on: Comparison with AERONET, etc?
• “Retrievability”: Do algorithms make same choices under same condi"ons? • Other metrics?
21
What’s s"ll missing from IFF • I-‐Bands:
– High resolu"on data (375 m) could help with cloud-‐masking/pixel selec"on
• Decision on NxN pixel size: – MODIS scans are units of 10 detectors (e.g.
10, 20, 40) – VIIRS scans are units of 8 detectors (e.g. 8
or 16) – Current MODIS-‐like is 10x10, but that
mixes can lines for VIIRS – Doesn’t make too much of a difference "
• Land surface reflectance ra"os (that exactly follow MODIS logic).
• Cloud mask (thermal-‐infrared tests)
• Formats, etc: – We are repor"ng products in MODIS-‐like
formats. – S"ll awai"ng science-‐team decision on
archival formats, meta-‐data, etc. • Note from C. Hsu (GSFC): Deep blue
(DB) products (V1) will be delivered, independent of rest of atmospheres
AOD using 10x10
ΔAOD if using 8x8
DT retrieval: Improvements
• Improving coverage • Removing bias over urban areas • A beNer dust retrieval over ocean • Etc
• Figuring out which updates will be in “forward” stream, and which can go into reprocessing.
MODIS (C6) misses many AOD events during winter months (AERONET confirms not cloud)
Leiku Yang and Yingxi Shi
Improving coverage
Clear Cloudy
2013282.0500
Case study over Beijing area shows that our cloud mask is working
C6 AOD
Instead it is the “In-‐land water mask” that is preven6ng retrieval over Beijing.
Modified AOD
Can we relax masks, but not degrade global retrieval?
Revised urban algorithm works very well in the US Global implementa6on is challenging, but forthcoming
Surface scheme is revised over urban areas by integra6ng land cover type informa6on in the retrieval algorithm.
DISCOVER AQ – BAL/DC
Urban %
(MDT AODs over urban surface are biased high w.r.t. AERONET) Urban % in the U.S. (Ci"es)
Characterizing / correc"ng urban surface bias
Urban Percentage
AOD (M
ODIS-‐AE
RONET)
Global implementa"on
Gupta et al., (in revision)
Finding and retrieving dust • C6 DT retrieval uses VIS-‐NIR-‐
SWIR, plus TIR for cloud and snow masking.
• Over ocean, C6 DT assumes spherical coarse models
• No angular informa"on to use non-‐spherical assump"ons.
• But there is dust informa"on in TIR.
• Can we detect dust, then retrieve with beNer models?
• " BeNer fine/coarse mode separa"on? BeNer spectral AOD? BeNer use for CERES?
Dust Occurance (04/10 - 05/11, 2011)
180W 150W 120W 90W 60W 30W 0 30E 60E 90E 120E 150E 180E90S
60S
30S
0
30N
60N
90N
1
2
3
4
5
6
7
8
9
(days) 10
Dust detec"on from cloud-‐mask product (Apr-‐May 2011)
Summary (MODIS "VIIRS) • MODIS-‐DT Collec"on 6
– Aqua/Terra level 2, 3; en"re record processed – “Trending” issues reduced – S"ll a 15% or 0.02 Terra vs Aqua offset. – Terra/Aqua convergence improved with C6+, but
bias remains.
• VIIRS-‐DT in development – VIIRS is similar, yet different then MODIS – With 50% wider swath, VIIRS has daily coverage – Ensures algorithm consistency with MODIS DT. – Currently: 20% NPP vs Aqua offset over ocean. – Only small bias (%) over land (2012-‐2016) – Can VIIRS/MODIS create aerosol CDR? – S"ll need to define “how good is good enough?”
• DT improvements / expansion – Will be applied to both MODIS and VIIRS – DT can be applied to addi"onal sensors (GOES-‐R,
Himawari, PACE, etc) 27
hNp://darktarget.gsfc.nasa.gov
• Reference for all things “dark target” – The algorithms and assump"ons – Examples – Valida"on – Primary publica"ons – Educa"onal material – FAQ – Links to data access – Considering a “forum”
28 THANK YOU!