uw cris sdr review december2013 20131213 · 200 220 240 260 280 300 bt (k) 700 800 900 1000 −0.2...
TRANSCRIPT
CrIS Radiometric Calibra1on: Uncertainty Es1mates and Evalua1ons
On behalf of the CrIS SDR Cal/Val team
Dave Tobin, Hank Revercomb, Joe Taylor, Bob Knuteson, Dan DeSlover, Lori Borg, Graeme Mar1n Space Science and Engineering Center, University of Wisconsin-‐Madison
SUOMI NPP SDR Science and Validated Product Maturity Review
NOAA Center for Weather and Climate Predic1on, College Park, MD
18-‐20 December 2013
Outline • Radiometric Uncertainty Es1mates
– Perturba)on of Calibra)on Equa)on and Parameter uncertain)es – On-‐orbit RU es)mates – Other terms
• Nonlinearity Refinements – Mo)va)on and Methodology, and changes to NLC equa)on and coefficients – Reprocessed dataset – Example changes in calibrated spectra
• Evalua1ons – AircraF underflights – CrIS/VIIRS comparisons – CrIS/IASI comparisons – CrIS/AIRS comparisons – Clear sky Obs-‐Calc
• Con1nuing Work – Spectral Ringing – Polariza)on – SW band biases
• Summary and Conclusions
2
Radiometric Uncertainty (RU) Es1mates • Perturba)on of Calibra)on Equa)on and Parameter uncertain)es • On-‐orbit RU es)mates • Other terms Ø Required in order to understand the size and dependencies of the
primary contributors to the CrIS SDR uncertain)es, for calibra)on improvements, weather, process, trend, and inter-‐calibra)on applica)ons.
3
Radiometric Uncertainty Es1mates
Simplified On-‐Orbit Radiometric Calibra1on Equa1on:
Rscene = Re {(C’scene – C’SP) /(C’ICT-‐C’SP)} RICT with:
Nonlinearity Correc)on: C’ = C � (1 + 2 a2 VDC) ICT Predicted Radiance: RICT = εICT B(TICT) + (1-‐εICT) [ 0.5 B(TICT, Refl, Measured) + 0.5 B(TICT, Refl, Modeled)]
Following Tobin et al. (2013), Suomi-‐NPP CrIS radiometric calibra>on uncertainty, J. Geophys. Res. Atmos., 118, 10,589–10,600, doi:10.1002/jgrd.50809.
Parameter Uncertain1es: Parameter Nominal Values 3-‐σ Uncertainty
TICT 280K 112.5 mK*
εICT 0.974-‐0.996 0.03
TICT, Refl, Measured 280K 1.5 K
TICT, Refl, Modeled 280K 3 K
a2 LW band 0.01 – 0.03 V-‐1 0.00403 V-‐1
a2 MW band 0.001 – 0.12 V-‐1 0.00128 – 0.00168 V-‐1
4
*Exelis at-‐launch es>mate
200
220
240
260
280
300
BT (K
)
700 800 900 1000
−0.2
−0.1
0
0.1
0.2
0.3
wavenumber
3−si
gma
RU
(K)
200 220 240 260 280 3000
0.1
0.2
0.3
BT (K)
3−si
gma
RU
(K)
wavenumber700 800 900 1000
1200 1300 1400 1500 1600 1700wavenumber
200 220 240 260 280 300BT (K)
wavenumber1300 1400 1500 1600 1700
2200 2300 2400 2500wavenumber
200 220 240 260 280 300BT (K)
wavenumber2200 2300 2400 2500
Example 3-sigma RU estimates
For a typical warm, ~clear sky spectrum
200
220
240
260
280
300
BT (K
)
700 800 900 1000
−0.2
−0.1
0
0.1
0.2
0.3
wavenumber
3−si
gma
RU
(K)
200 220 240 260 280 3000
0.1
0.2
0.3
BT (K)
3−si
gma
RU
(K)
wavenumber700 800 900 1000
1200 1300 1400 1500 1600 1700wavenumber
200 220 240 260 280 300BT (K)
wavenumber1300 1400 1500 1600 1700
2200 2300 2400 2500wavenumber
200 220 240 260 280 300BT (K)
wavenumber2200 2300 2400 2500
5
500 1000 1500 2000 25000
0.05
0.1
0.15
0.2
wavenumber (cm−1)
3m B
T R
U (K
)
TICT¡ICTTICT,Refl,MeasTICT,Refl,ModelTST¡STTST,Refla2Total RU
500 1000 1500 2000 25000
0.05
0.1
0.15
0.2
wavenumber (cm−1)
3m B
T R
U (K
)
TICT¡ICTTICT,Refl,MeasTICT,Refl,ModelTST¡STTST,Refla2Total RU
BT (K)
3−si
gma
RU
(K)
200 250 3000
0.1
0.2
0.3
0.4
0.5
BT (K)
200 250 300
log(count)0 5 10
BT (K)200 250 300
Example 3-sigma RU estimates
LW MW SW
Log scale RU distributions for one orbit of CrIS Earth view data, including all FOVs and spectral channels within the band:
6
Ø Uncertain)es are greatly reduced due to re-‐analysis of the TVAC data and on-‐orbit FOV-‐2-‐FOV analysis. In par)cular, MW band uncertain)es are greatly reduced due to the high degree of linearity of MW reference FOV9.
Ø Overall, RU is <0.3K (LW), <0.15K (MW), <0.15K (SW): Bemer than spec by approximately a factor of 4.
Other Terms
Ø Spectral Ringing Ø Polariza)on Ø Possible SW Nonlinearity • Other smaller/negligible terms:
– Detector temperature changes, Changes in DA Bias )lt over 4 minutes, Changes in op)cal flatness, OPD sample rate driF over 4 minutes, Electronic gain driF over 4 minutes, Electronic delay driF over 4 minutes, FOV to FOV crosstalk in same band, FOV to FOV crosstalk between bands, Stray light, Op)cs temperature change during cal, Changes in channel spectra
Smaller contributors not currently accounted for in the calibration algorithm or included in current RU estimates:
7
Addressed in later slides
Nonlinearity Refinements • Mo)va)on and Methodology, and changes to NLC equa)on
and coefficients • Reprocessed dataset • Example changes in calibrated spectra
8
Nonlinearity Refinements (Refinements developed since the Provisional Review)
Mo1va1on • To provide improved traceability of the Nonlinearity algorithm and coefficients to the CrIS TVAC
External Calibra)on Target residuals and analysis. • This is most important for the LW band where all FOVs display a significant level of nonlinearity, and
less so for the MW band where some FOVs display high levels of linearity and can be used as an in-‐orbit reference to assess the other MW FOVs.
Methodology • Involves determina1on of the a2 nonlinearity coefficients (other NL related terms are rela1vely well
known): 1. Ini)al values determined from analysis of TVAC ECT view data. 2. Change from pre-‐launch to in-‐orbit es)mates based on analysis of Diagnos)c Mode data (out-‐
of-‐band harmonics), leading to ini)al in-‐orbit values. 3. Followed by the selec)on of a reference FOV and adjustments of a2 values for the remaining
FOVs to create op)mal agreement with the reference FOV for Earth view data
Resul1ng changes to the NLC equa1on and coefficients – New a2 values (with respect to current opera)onal values, includes an overall increase for all LW
FOVs, along with smaller refinements to improve FOV-‐2-‐FOV agreement for LW and MW bands) – Equa)on change from C’ = C / (1 -‐ 2a2V) to C’ = C � (1 + 2a2V) – New modula)on efficiency (emod ) values
9
Reprocessed Dataset The refined CrIS SDRs for the full mission are available at:
Fp://peate.ssec.wisc.edu/allData/products/results/cris/cspp/SDR_1_4b_ILS_NLC_v33a-‐04/
Differences with respect to the opera1onal IDPS dataset:
1. Includes Nonlinearity algorithm and coefficient refinements* 2. Includes ILS algorithm and coefficient refinements* 3. Includes consistent SDR algorithm processing for the full mission 4. Processing takes place with ~24 hour latency to avoid missing packet issues
* The same Nonlinearity and ILS refinements are expected to be implemented in IDPS processing with MX8.1 and EPv36 in February 2014.
Evalua1ons of the CrIS RU presented here use the reprocessed dataset. Some example differences with respect to the opera1onal IDPS processing are shown on the following slides.
10
Example changes: New versus Old Nonlinearity Correc1on
600 650 700 750 800 850 900 950 1000 1050 1100200
220
240
260
280
300SCRIS_npp_d20120920_t0007219_e001239: FOR = 15 FOV = 9
CrIS
FM
1 B
.T. (
K)
600 650 700 750 800 850 900 950 1000 1050 1100
−0.2
−0.1
0
0.1UW NLC (mod4) minus CSPP default
Wavenumber (cm−1)
B.T
. Diff
(K)
672-‐682 cm-‐1
11
Ø LW stratospheric channels shiF to colder brightness temperatures (order ~0.1K) everywhere.
Ø For cold scenes, LW window channels also shiF to colder brightness temperatures.
Example changes: Long term FOV-‐2-‐FOV differences
Apr Jul Oct Jan Apr Jul
−0.2
−0.1
0
0.1
0.2
BT D
iff (K
)
Apr Jul Oct Jan Apr Jul
FOV1FOV2FOV3FOV4FOV5FOV6FOV7FOV8FOV9
LW (672-‐682 cm-‐1) BT Differences with respect to FOV5
IDPS processing
Feb Apr Jun Aug Oct Dec Feb Apr Jun Aug−0.25
−0.2
−0.15
−0.1
−0.05
0
0.05
0.10.15
0.2
BT D
iff (K
)
Feb Apr Jun Aug Oct Dec Feb Apr Jun Aug
FOV1FOV2FOV3FOV4FOV5FOV6FOV7FOV8FOV9
Reprocessed
Feb Apr Jun Aug Oct Dec Feb Apr Jun Aug−0.25
−0.2
−0.15
−0.1
−0.05
0
0.05
0.10.15
0.2
BT D
iff (K
)
Feb Apr Jun Aug Oct Dec Feb Apr Jun Aug
FOV1FOV2FOV3FOV4FOV5FOV6FOV7FOV8FOV9
12
Ø Reprocessing with the NLC refinements removes the inconsistencies in the IDPS )me series and also minimizes the magnitude of FOV-‐2-‐FOV differences.
Evalua1ons of RU es1mates (“Cal/Val”)
• AircraF underflights • CrIS/VIIRS comparisons • CrIS/IASI comparisons • CrIS/AIRS comparisons • Clear sky Obs-‐Calc
Ø A range of techniques, with various levels of uncertainty/sta)s)cs/traceability, to assess the CrIS SDRs and associated RU es)mates.
13
May 15 Underflight example: S-‐HIS and CrIS 895–900 cm-‐1 BTs overlaid on VIIRS true color image
May 2013 Suomi-‐NPP JPSS Aircrah Campaign Scanning-‐HIS evalua1ons of CrIS Calibra1on
14
BT [K]
BT [K]
Double Obs-‐Calc Comparison Methodology and Uncertainty (CrISobs-‐CrIScalc)-‐(SHISobs-‐SHIScalc)
15
650 700 750 800 850 900 950 1000 1050 11000
0.1
0.2
0.3
0.4
BT (K
)
wavenumber
Double Obs−Calc Uncertainty
Spatial VariabilitySHIS 3−sigma RUObs−Calc MethodologyTotal (RSS)
S-‐HIS Calibra1on, Calibra1on Verifica1on, and Traceability
16
NIST TXR Valida)on of S-‐HIS Radiances • Pre and post deployment end-‐to-‐end calibra)on verifica)on • Instrument calibra)on during flight using on-‐board calibra)on blackbodies • Periodic end-‐to-‐end radiance evalua)ons under flight-‐like condi)ons with NIST transfer sensors
Post SNPP End-‐to-‐End Calibra)on Verifica)on
333K
318K
298K
273K
Channel 1 @ 5µm
Channel 2 @ 10µm
CrIS/S-‐HIS Underflight Results Hamming apodiza)on
Ø AircraF underflights provide periodic end-‐to-‐end verifica)on of CrIS RU es)mates with 0.1-‐0.2K uncertainty over most of the spectrum.
17
Mean DOMC ± uncertainty CrIS RU
CrIS/VIIRS comparisons Example Daily Comparisons, M15 band @ 10.8µm, Descending
CrIS convolved with VIIRS SRF VIIRS mean within CrIS FOVs
VIIRS standard devia1on within CrIS FOVs Differences for uniform scenes
BT (K) 210 220 230 240 250 260 270 280 290 300
0 1 2 3 4 5 6 7 8 9 10 -‐5 -‐4 -‐3 -‐2 -‐1 0 1 2 3 4 5
BT (K) 210 220 230 240 250 260 270 280 290 300
BT (K) VIIRS -‐ CrIS (K)
Ø Each day includes ~500,000 coloca)ons which pass a spa)al uniformity test 18
−0.1
−0.05
0
0.05
0.1VI
IRS −
CrIS
BT
(K)
CrIS/VIIRS Daily Mean Differences
−0.02
−0.01
0
0.01
0.02
VIIR
S −
CrIS
BT
(K)
Mar
Apr
May
Jun Jul
Aug
Sep
Oct
Nov
Dec
Jan
Feb
Mar
Apr
May
Jun Jul
Aug
Sep
Oct
Nov
Dec
Jan
−0.1
−0.05
0
0.05
0.1
VIIR
S −
CrIS
BT
(K)
CrIS/VIIRS Daily Mean Differences
M13, 4umM15, 10.8umM16, 12um
−0.02
−0.01
0
0.01
0.02
VIIR
S −
CrIS
BT
(K)
Mar
Apr
May
Jun Jul
Aug
Sep
Oct
Nov
Dec
Jan
Feb
Mar
Apr
May
Jun Jul
Aug
Sep
Oct
Nov
+0.074 K
-‐0.064 K
-‐0.022 K
Mean Bias over past 21 months:
with mean biases subtracted off
19
-‐6.4 ± 1.0 -‐1.6 ± 1.2 -‐4.4 ± 1.0
Trends, mK/yr:
Ø CrIS/VIIRS daily mean differences are < 0.1K and trends are < 10 mK/yr
CrIS/VIIRS comparisons with uncertainties
Legend
Mean differences
CrIS 3σ RU
VIIRS 3σ RU
Uncertainty due to Out-of-Band SRF effects
Statistical uncertainty
RSS
20
Ø VIIRS RU es)mates provided by Jeff McIn)re et al.
Ø CrIS/VIIRS differences are bounded by combined RU for all scene temperatures
Ø Larger VIIRS RU for cold scenes at M15 and M16 are due to c0 offset term and under inves)ga)on by the VIIRS SDR team.
VIIR
S −
CrIS
BT
(K)
BT (K)
M13 @ 4µm
200 220 240 260 280 300 320−0.5
0
0.5
VIIR
S −
CrIS
BT
(K)
BT (K)
M15 @ 10.8µm
200 220 240 260 280 300 320−0.5
0
0.5
VIIR
S −
CrIS
BT
(K)
BT (K)
M16 @ 12µm
200 220 240 260 280 300 320−0.5
0
0.5
SNO Comparison Methodology
21
−2 0 2101
102
103
104
AIRS − CrIS (K)
The SNO comparison technique is aimed at minimizing differences in the spa)al/temporal colloca)on process and providing well understood uncertain)es to iden)fy persistent biases between two sensors.
A sample SNO showing CrIS and AIRS footprints within 100 km of the SNO loca)on.
Colloca)on difference distribu)ons for a large ensemble of SNOs for various
ranges of spa)al variability.
0
50
100
21:53:59 UTC
AIRSCrIS
700 800 900 1000 11000
2
4
6
wavenumber (cm−1)
0
50
100
Mea
n (m
W/(m
2 sr.
cm−1
))
22:31:24 UTC
700 800 900 1000 11000
2
4
6
wavenumber (cm−1)
STD
EV (m
W/(m
2 sr.
cm−1
))
LW mean and standard devia)on spectra for two example SNOs collected on 20120816.
830-‐840 cm-‐1
SNO Datasets
22
CrIS/AIRS: 1.2M “Big Circle” SNOs collected to date (March 2012 to Nov 2013); 20 minute window; -‐30 to 30 deg scan angle, <=2 deg scan angle diff.
AIRS V5 L1B; CrIS ADL/CSPP SDR_1.4b_NLC_ILS
835 cm-‐1 CrIS/AIRS SNO BT Diffs
CrIS/IASI-‐A: 5270 “Big Circle” SNOs collected to date (March 2012 to Nov 2013); 20 minute window; nadir. ~20 days of coincidences, ~30 day gaps,
~half at +72.4 deg, ~half at -‐72.4 deg. IASI_xxx_1C_M02; CrIS ADL/CSPP SDR_1.4b_NLC_ILS
2510 cm-‐1 CrIS/AIRS SNO BTs
CrIS/IASI SNO loca1ons
CrIS/IASI Northern SNOs Hamming apodiza)on
Ø Results shown for IDPS processing and with the reprocessed dataset including the NLC refinements presented earlier.
Ø Differences of ~0.2K or less
Ø NLC refinements: • Improved agreement in the
LW band. • Negligible changes in the
the MW band (as expected).
mean CrIS mean IASI
Mean Difference ± Uncertainty
−0.4
−0.2
0
0.2
0.4
BT
Dif
(K)
−0.4
−0.2
0
0.2
0.4
BT
Dif
(K)
200 220 240 260 280 300 320
−0.4
−0.2
0
0.2
0.4
BT (K)
BT
Dif
(K)
200 220 240 260 280 300 320BT (K)
672-‐682 cm-‐1 830-‐840 cm-‐1
1382-‐1408 cm-‐1 1585-‐1600 cm-‐1
2360-‐2370 cm-‐1 2500-‐2520 cm-‐1
24
Summary of SNO results for 6 representa1ve spectral regions,
and VIIRS/CrIS comparisons:
VIIRS–CrIS, M15 VIIRS–CrIS, M16 VIIRS–CrIS, M13
600 700 800 900 1000 1100220
240
260
280
300
BT (K
)
1200 1300 1400 1500 1600 1700 1800220
240
260
280
300
BT (K
)
1800 2000 2200 2400 2600 2800220
240
260
280
300
wavenumber
BT (K
)
VIIRS M16
VIIRS M15
VIIRS M13
672-‐682 cm-‐1 830-‐840 cm-‐1
1382-‐1408 cm-‐1 1585-‐1600 cm-‐1
2360-‐2370 cm-‐1 2500-‐2520 cm-‐1
IASI-‐CrIS AIRS-‐CrIS
Ø LW differences display only small dependence on scene BT for both IASI and AIRS SNOs.
Ø MW differences are rela)vely independent of scene BT for IASI and for AIRS at 1382-‐1408 cm-‐1; Differences for AIRS at 1585-‐1600 cm-‐1 range from ~+0.3K at 200K to -‐0.1K at 265K.
Ø SW differences are rela)vely flat above ~240K; Below ~230K larger differences between all three sensors are observed.
Ø Consistent with SNO results shown in L. Strow presenta)on, and reported by L. Wang et al. at NOAA STAR.
25
800 1000 1200 1400 1600 1800 2000 2200 2400 2600−1.5
−1
−0.5
0
0.5
1
1.5
Wavenumber (cm−1)
Ham
min
g O
bs−
Calc
in K
ECMWF IFS 38r1
ECMWF IFS 38r2
Difference
1000 1500 2000 2500−0.5
0
0.5
1
1.5
2
2.5
Wavenumber (cm−1)
Sta
nd
ard
De
via
tion
of
Ob
s−C
alc
(K
)
ECMWF IFS 38r1
ECMWF IFS 38r2
Difference X10
c/o Yong Chen and Yong Han, NOAA STAR:
c/o Larrabee Strow, UMBC:
Clear Sky Obs-‐Calc Analyses
Ø Behavior of mean biases and standard devia)on of obs-‐calcs are consistent with forward model and atmospheric state uncertain)es and imply very good radiometric performance for CrIS.
Before 7/25/13 (ECMWF IFS 38r1) AFer 7/25/13 (ECMWF IFS 38r2) Difference
Before 7/25/13 (ECMWF IFS 38r1) AFer 7/25/13 (ECMWF IFS 38r2) Difference
Con1nuing Work: Calibra1on and RU Refinements
• Spectral Ringing • Polariza)on • SW band biases
26
Spectral Ringing
• Considered to be part of the RU budget, but is covered separately in Dan Mooney’s talk
• For the large majority of spectral channels, the associated ar)facts are very small, and with apodiza)on are negligible everywhere.
27
0 10 20 30 40 50 60 70 80 90−0.1
−0.05
0
0.05
0.1
CrIS cross track scan index
VIIR
S −
CrIS
(K)
200−210 K210−220 K220−230 K230−240 K240−250 K250−260 K260−270 K270−280 K280−290 K290−300 K300−310 K310−320 K320−330 K
Error from Scene Mirror Induced Polariza1on • CrIS uses a 45° gold scene mirror that provides low sensi)vity to polariza)on; no
correc)on is included in the SDR algorithm/processing. • However, it seems almost certain that CrIS should have polariza)on effects of ~0.1 K for
especially warm and cold brightness temperatures in some spectral regions. • A correc)on should be developed based on CrIS characteriza)on tests yet to be
conducted (measurements of scene mirror degree of polariza)on, pr, and interferometer polariza)on sensi)vity, pt)
• Radiance error dependence ~2prpt (N – BICT) • Sugges)ons of this type of behavior in CrIS/VIIRS comparisons vs. scan angle:
28 * Biases removed for mean of INDICES 1-‐10 & 80-‐90
Scan Angle and Scene BT dependence of VIIRS/CrIS Comparisons at 10.8µm*
29
200
250
300
BT (K
)
200
250
300
BT (K
)
2150 2200 2250 2300 2350 2400 2450 2500 25500
0.05
0.1
wavenumber
BT (K
)
SW Band Biases • FOV-‐2-‐FOV analyses and differences with
respect to other sensors suggest small ar)facts in the SW band, both in Mean biases and FOV-‐2-‐FOV differences. E.g. Differences with respect to IASI
• Mechanisms inves)gated to date: – Spectral shiF – Thermal SP view contamina)on – Solar SP view contamina)on – Noise – Polariza)on Ø Low level Nonlinearity
• Displays FOV dependent behavior • Has plausible spectral and scene level
dependencies
0.20
0.20.40.60.8
FOV 1
−0.20
0.20.40.60.8
FOV 2
−0.20
0.20.40.60.8
FOV 3
0.20
0.20.40.60.8
FOV 4
−0.20
0.20.40.60.8
FOV 5
−0.20
0.20.40.60.8
FOV 6
22002300240025000.2
00.20.40.60.8
FOV 7
2200230024002500−0.2
00.20.40.60.8
FOV 8
2200230024002500−0.2
00.20.40.60.8
FOV 9
CrIS-‐IA
SI (K
) CrIS-‐IA
SI (K
) CrIS-‐IA
SI (K
)
wavenumber wavenumber wavenumber
Mean for all FOVs Difference for this FOV
Sample spectra (top) and Effect of small NLC (bomom)
Summary and Conclusions
30
Ø On-‐orbit Radiometric Uncertainty (RU) characterized: – Based on careful es)ma)on of 3-‐sigma uncertain)es of the primary calibra)on parameters and
perturba)on of the radiometric calibra)on equa)on – Overall RU es)mates are < 0.3K (LW), < 0.15K (MW), < 0.15 K (SW)
Ø Nonlinearity correc)on algorithm and coefficients refined – Provides improved traceability of the nonlinearity coefficients to the TVAC External Calibra)on Target – Refinements reduce overall RU in LW band and reduce FOV dependence in LW and MW bands – SDRs reprocessed using NLC (and ILS) refinements and distributed
Ø CrIS SDRs and RU es)mates have undergone extensive verifica)on/evalua)on over a range of representa)ve condi)ons – Periodic aircraF underflights provide high quality, SI traceable verifica)on of the CrIS RU – CrIS/VIIRS comparisons imply excellent stability of both sensors, and scene BT dependence further
characterized and diagnosed – SNOs of CrIS/AIRS and CrIS/IASI show excellent mean agreement and behavior with scene temperature – Clear sky obs-‐calcs imply very good CrIS radiometric performance
Ø Areas of further refinement have been iden)fied and are under inves)ga)on – Spectral Ringing – Polariza)on – Possible low level SW band nonlinearity