Transcript SST from GHRSST/MODIS

MISST Science Team Meeting
November 28, 2006
Sea-surface Temperature from
GHRSST/MODIS –
recent progress in improving
accuracy
Peter J. Minnett & Robert H. Evans
with
Kay Kilpatrick, Ajoy Kumar, Warner Baringer, Erica Key,
Goshka Szczodrak, Sue Walsh and Vicki Halliwell
Rosenstiel School of Marine and Atmospheric Science
University of Miami
Outline
•Leverage of Efforts Across Multiple Grants
•Generation of Single Sensor Error Statistics for AVHRR
and MODIS
• GHRSST MODIS product generation division of effort
• Status of MODIS SST
• MODIS approach to SSES
• Initial observations
– Space and Time resolution of sst analysis fields has
important implications for sst retrieval coverage and
quality
– Regions of IR (MODIS) and microwave (AMSR)
difference not correlated with water vapor
• Conclusions
Leverage of Effort for MISST
•NOPP - MISST support for generation of SSES
- Minnett, Evans
•NOPP - ISAR and voluntary ship program to
acquire radiometric in situ SST - Minnett
•NASA - MODIS algorithm support - Evans
• NASA - MAERI in situ observations for
MODIS calibration and validation - Minnett
• NASA - Transfer of GHRSST/MODIS SST to
JPL with product production at OCPG, GSFC Evans
GDAC
Real Time MODIS processing for GHRSST
July, 2005 formation of MODIS SST processing team
(JPL, OBPG - GSFC, Miami)
Division of effort:
Miami - algorithm development, cal/val, base code development
OBPG (Bryan Franz) integrate code into OBPG processing, process MODIS
Terra, Aqua; day, night; global 1km; SST, SST4; transfer files to JPL
JPL PO.DAAC (J. Vazquez, E. Armstrong) convert OBPG files into L2P, add remaining fields, ice mask, distance to
clouds…, transfer files to Monterrey, interested users
Program ‘near real-time’ processing MODIS Terra and Aqua
L2P (OCPG) and transferring to JPL GDAC
Background – Algorithm Maintenance
and Validation
The foundation of Algorithm Maintenance is the comparison of MODIS SST
retrievals with surface-based measurements of equal or superior accuracy
(reference field). This is usually referred to as “validation.” It is needed to
give:
–
–
–
confidence in the values of the geophysical fields.
knowledge under what circumstances an algorithm performs well, and
when it performs badly (i.e. not enough to know that the retrievals
represent the mean conditions well) - error statistics.
guidance to improve algorithm performance.
But in reality…..
•There are no perfect reference fields….
•For validation we must rely on imperfect reference fields, with known or
unknown uncertainties, inadequate spatial coverage, and incomplete
sampling of the governing parameters.
•The uncertainties in the reference field must be well known so they are not
attributed to the satellite retrieval.
Marine-Atmospheric Emitted Radiance
Interferometer (M-AERI)
Laboratory tests of M-AERI accuracy
Target Temp.
20oC
30oC
60oC
LW
(980-985 cm-1)
+0.013 K
-0.024 K
-0.122 K
SW
(2510-2515 cm-1)
+0.010 K
-0.030 K
-0.086 K
The mean discrepancies in the M-AERI 02 measurements of the NIST –
characterized water bath blackbody calibration target in two spectral
intervals where the atmosphere absorption and emission are low.
Discrepancies are M-AERI minus NIST temperatures.
Specifications
Spectral interval ~3 to ~18µm
Spectral resolution 0.5 cm-1
Interferogram rate 1Hz
Aperture 2.5 cm
Detectors InSb, HgCdTe
Detector temperature 78oK
Calibration Two black-body cavities
SST retrieval uncertainty <<0.1K (absolute)
Constructed by SSEC, U. Wisconsin - Madison
Traceable to National Standards: NIST EOS TXR
Surface radiometry
• Use ship-based radiometers,
e.g. M-AERI, ISAR,
CIRIMS and others.
• M-AERI is the reference
standard for satellite SST
retrievals (AVHRR, AATSR,
as well as MODIS), and for
other ship-board
radiometers.
• M-AERI also being used for
AMSR-E & AIRS SST
validation.
M-AERI cruises
Number of deployments
40
Number of ships
23
Number of days
3352
ISAR – an autonomous IR radiometer
• ISAR – Infrared
SST Autonomous
Radiometer
• Filter radiometer,
internal calibration
• Deployed on Jingu
Maru, Atlantic
crossings
• Currently on Mirai
in Indian Ocean
Buoy measurements
MODIS SST4 - Buoy Residuals
Feb 2000 - Aug 2006
Spatial Distribution
N = 12536
@ qf = 0
MODIS v5 global error statistics
M-AERI
Buoys
SST 11-12 μm
Year
2000
2001
2002
2003
2004
2005
2006
all years
TERRA
day
mean
RMS
-0.139
-0.262
-0.135
-0.086
-0.068
-0.110
-0.105
-0.108
SST 11-12 μm AQUA
da y
Ye a r
m ean
RMS
2002
-0.153
2003
-0.133
2004
-0.137
2005
-0.152
2006
-0.135
all years
-0.142
SST 4μm
Ye a r
2000
2001
2002
2003
2004
2005
2006
all years
TERRA
night
m ean
RMS
-0.161
-0.220
-0.191
-0.176
-0.178
-0.178
-0.174
-0.179
Count
0.797
1.430
0.621
0.607
0.579
0.549
0.574
0.650
3091
6321
9244
15685
24964
39826
32495
131626
Count
0.538
0.577
0.562
0.539
0.550
0.553
10293
22988
26415
40941
34687
135324
Count
0.829
0.663
0.528
0.500
0.493
0.471
0.473
0.505
1993
5397
7580
12006
18452
31130
25294
101852
1800
4935
6935
11058
16943
28460
23149
93280
SST 11-12μm TERRA
day
Year
mean
RMS
2000
-0.015
2001
0.115
2002
0.174
2003
0.016
2004
0.155
2005
0.121
2006
-0.032
all years
0.090
night
m ean
RMS
Count
-0.235
0.499
5906
-0.224
0.508
12977
-0.219
-0.484
15471
-0.235
0.461
25083
-0.205
0.452
22187
-0.222
0.475
81624
SST 11-12μm AQUA
day
Year
mean
RMS
2002
0.079
2003
-0.087
2004
0.087
2005
0.171
2006
-0.176
all years
0.037
night
mean
RMS
-0.186
-0.228
-0.204
-0.190
-0.167
-0.213
-0.208
-0.200
AQUA
night
m ean
-0.224
-0.217
-0.214
-0.223
-0.208
-0.216
Count
0.794
0.707
0.580
0.558
0.559
0.519
0.524
0.555
SST 4μm
Year
RMS
Count
0.449
0.455
0.426
0.414
0.404
0.423
6429
14095
16765
27280
24140
88709
2000
2001
2002
2003
2004
2005
2006
all years
TERRA
night
mean
RMS
-0.055
-0.046
0.004
-0.123
0.010
0.008
-0.017
-0.030
116
510
236
382
364
176
164
1948
night
mean
RMS
-0.035
-0.036
0.020
-0.060
0.086
0.092
-0.041
0.006
0.493
0.475
0.502
0.453
0.510
0.466
0.430
0.408
134
323
249
113
75
803
night
mean
RMS
-0.061
-0.262
0.034
0.061
0.008
-0.039
0.440
0.473
0.534
0.469
0.510
0.513
Count
0.613
0.557
0.448
0.513
0.687
0.723
0.515
0.584
Count
0.544
0.621
0.615
0.578
0.459
0.593
AQUA
night
mean
Count
0.462
0.387
0.390
0.358
0.407
0.458
0.373
0.400
115
714
397
453
597
350
316
2942
-0.158
-0.213
-0.019
-0.038
0.007
0.063
Count
1
6
3
4
5
2
3
23
Count
RMS
2
4
2
1
11
Count
0.384
0.389
0.486
0.382
0.432
0.442
3
5
2
1
13
But bias & rms alone do not tell the
whole story…
Systematic patterns in residual uncertainties
indicate shortcomings in the atmospheric
correction algorithms, and indicate how they
can be improved……
SSES for AVHRR
•Pathfinder Match-up database analyzed for
behavior of residuals (AVHRR to buoy and MAERI)
•Residuals structured as function of:
–Satellite zenith angle
–Temperature
–Time
Measure of satellite retrieval
uncertainty for MODIS
Standard uncertainty approach is to provide a global
estimate of bias and standard deviation.
Based on high quality radiometry and buoy SST
measurements
MODIS - GHRSST (GODAE High Resolution Sea Surface
Temperature Pilot Project) approach:
– To provide a statistical estimate of expected bias and standard
deviation for each satellite-retrieved SST
– Partition satellite - in situ match-up database along 7 dimensions
(environmental conditions and observing geometry)
– The “uncertainty hypercube” has been implemented for MODIS
SST and SST4 products and applied to the AQUA and TERRA
instruments
MODIS Single Sensor Error Statistics Approach
Bias and Standard Deviation Hypercube
Hypercube dimensions (partitioning of Match-up database):
- Time- quarter of year (4)
- Latitude band (5):
"60S to 40S" "40S to 20S" "20S to 20N" "20N to 40N" "40N to 60N"
- Sat Zenith angle intervals (4):
"0 to 30 deg" "30+ to 40 deg" "40+ to 50 deg" "50+ deg"
- Surface temperature intervals (8): 5 degree intervals
- Channel difference intervals:SST(3), SST4(4)
ch31-32 (SST): 0.7<, 0.7->2.0, >2.0
ch22-23 (SST4) 0.5 degree intervals: -0.5<, -0.5->0, >0 ->0.5, >0.5
-Quality level (2)
cube created only for ql=0 and 1
Note for ql2 and 3 the bias and standard deviation are each fixed to a single value
-Day/Night
No interpolation between adjacent cells in Hypercube
SSES Characteristics
SSES Bias wrt In Situ
MODIS SST – Reynolds OI SST
Predicted
SSES Bias
SST Difference wrt
Weekly OI Field
St. Dev. of
SSES Error
Terra MODIS SST
for GHRSST
SSES St. Dev wrt In Situ
MODIS SST
February 1
SSES Bias wrt In Situ
SSES St. Dev wrt In Situ
MODIS SST – Reynolds OI SST
MODIS SST
Water-vapor dependence…
• Water vapor is one of the main
atmospheric constituents that
contribute to the atmospheric
effect in the infrared.
• Water vapor is not an
independent variable in the
atmospheric correction
algorithm, but is represented by
a proxy (brightness temperature
difference).
• Non-linearity in the current
algorithm water vapor
dependence treated with 2 part
linear fit.
Microwave SST accuracies
AMSR-E M-AERI
comparisons
during AMMA,
May-July 2006.
Parts of the cruise
tracks under
clouds of ITCZ
Mean
K
0.033
0.143
0.088
St.
Dev.
K
0.478
0.350
0.421
18
18
36
NOAA S Ronald H Brown
Ascending arc (daytime) 0.105
Descending arc (night)
0.081
Both
0.092
0.439
0.281
0.358
15
17
32
Both Ships
Ascending arc (daytime)
Descending arc (night)
0.065
0.113
0.455
0.321
33
35
0.090
0.390
68
N/O L'Atalante
Ascending arc (daytime)
Descending arc (night)
Both
All
N
MODIS to AMSR-E SST comparisons
MODIS – AMSR-E SST
Differences in MODIS
and AMSR-E SSTs
have spatial patterns,
that do not correlate
with the water vapor
proxy.
Water vapor proxy
Other geophysical
parameters also
involved.
11-12μm brightness temperature differences
24 August 2004
Summary
- V5 monthly coefficients removed seasonal bias trends, Terra
mirror side trends
- SST4 rms order 0.4K, SST order 0.5K
- SST4 less affected by dust aerosols, water vapor
- Improved quality filtering removed most cold clouds and
significant dust aerosol concentrations
- Hypercube developed and tested for Terra and Aqua, provided
to OCPG and included in Aqua and Terra L2P processing
- Introduction of SSES hypercube provides insight into bias and
standard deviation trends as a function of time, latitude,
temperature, satellite zenith angle, brightness temperature
difference as a proxy for water vapor and retrieval quality level
Conclusions
- MODIS SSTs of “climate record” quality, having extensive error
characterization, and traceability to NIST standards
- No evidence that Terra SSTs are of poorer quality than Aqua SSTs
- MODIS SSTs are an important component of GHRSST-PP
- An important focus of GHRSST-PP is quantifying effects of diurnal
heating… benefits from Terra AND Aqua
- Hypercube provides insight leading to improved retrieval equation
coefficient generation
- DT analysis and hypercube bias comparable for most retrievals
Challenges:
- Many areas of climate interest are very cloudy – approach to follow is to
use AMSR-E SSTs as a “transfer standard”
- M-AERIs are still the best source of validation data, but are “showing their
age….”
END