A Consensus Calibration Based on TMI and Windsat
Tom Wilheit1,Wes Berg2, Linwood Jones3, Rachael Kroodsma4, Darren McKague4, Chris Ruf4, Matt Sapiano2 Texas A&M University, (2) Colorado State University, (3) University
of Central Florida, (4) University of Michigan
GPM Intersatellite Working Group (aka X-CAL)
Make Radiances from Constellation Radiometers Physically ConsistentDifferences of Frequencies/Incidence Angles
Clean Up Other Problems
Variety of Methods To Generate Two Point RecalibrationsUnified set of Deltas
Tbnew = A* Tbold + B
Could Be More Complex Where Necessary.
Currently Have Consensus Calibration based on Windsat and TMIUse TMI as Transfer Standard. (CC_1.1) (75% Windsat—25% TMI)
In Process of Applying to AMSR-E and SSM/I (F13,14,15)
Working on CC_1.2 Not Very Different but More Defensible
X-CAL Status OverviewAccuracy: Relative/ To What Degree can We Make the Sensors Alike? Looks Like We are Meeting our 0.1K Goal We Continue to Refine for Sake of Credibility Absolute Difficult to Estimate/ Standards are Inadequate NIST is Working on Standards/ May Impact Satellites in Future
Double Differences Where Possible We Use Double Differences to Minimize Model Sensitivity Corresponding Channels e.g. TMI 19.35V/ Windsat 18.7V
S = Source Sensor T= Target Sensor
DD = (TBTobs – TBTcalc) - (TBSobs – TBScalc) = (TBTobs – TBSobs) - (TBTcalc – TBScalc)
DATA SETJuly 2005-June 2006TMI Version 7 (Includes reflector temperature correction)Windsat
TAMU, UCF, CSU methods use coincident(±1h) observations binned in 1° boxesTAMU solves for SST, WS, PW and CLW using source sensor then computes targetCSU similar to TAMU but external SST and covariancesUCF computes both sets of TBs from GDAS analyses
Univ. Michigan cold end method uses global histograms of TBs and finds limiting values.
Warm End : U. Michigan uses method analogous to TAMU and CSU over Amazon Rain Forest. JPL and UCF use same method but less mature implementations
Unified Deltas
Common TBs 163. 85. 188. 109. 200. 206. 135. Extrapolate method 1 &2 deltas to common TBsTAMU, UCF, CSU MethodsAll Data 0.31 -1.66 -0.61 -3.20 -1.50 -3.24 -2.41Lowest 0.18 -1.71 -0.76 -3.08 -1.89 -3.25 -2.42 Warm End (U. Mich Method) 281. 280. 285. 284. 284. 281. 281.K -0.76 -0.92 -1.20 -1.43 -3.37 -3.17 -3.16K
UNIFIED DELTAS
Standard Deviations (K) Uncertainties in the Means (K) ALL LOW ALL LOW10V 0.061 0.032 0.035 0.01910H 0.066 0.078 0.038 0.04519V 0.151 0.221 0.087 0.12819H 0.179 0.189 0.103 0.10921V 0.329 0.241 0.190 0.13937V 0.010 0.059 0.006 0.03437H 0.020 0.101 0.012 0.058
Courtesy of Darren McKague
Consensus Calibration
Warm End Variance TMI a little more than twice as large as WS (K**2)Cold End Variance TMI a little more than three times as large as WS
Keep the numbers simple and round Windsat Gets 3 times the weight of TMI (i.e. 75%WS/25%TMI)
Consensus Calibration 1.1
75% of Unified Deltas
TMI_CC_1.110V 10H 19V 19H 21V 37V 37H0.23K -1.25 -0.46 -2.40 -1.42 -2.43 -1.81
@ 163K 85 188 109 200 206 135-0.57K -0.69 -0.90 -1.07 -2.53 -2.38 2.37
@ 281K 280 285 284 284 281 281
Negative #’s indicate TMI cold relative to Windsat
Plans for CC_1.2
Weights for Unified Deltas based on Error Models
Common Partitioning of Cold End Values (Quartiles)
Examine Updated Radiative Transfer Models
Earth Incidence Angle Issue
From Steve Bilanow’s Presentationat March 2011 X-CAL meeting
TMI Pointing Uncertainty Effects
• “Prelaunch measure of TMI 10 V and10H boresight alignment offsets from a 49 degrees scan cone were reported at 0.555 and 0.185 degrees respectively*.
* Memorandum from Jim Shiue, 12/11/97”This corresponds to ~ 1.3 K and -0.2 K bias shifts.
When you do the trigonometry, this translates to increases of the Earth Incidence Angle for the two 10.7 GHz channels of 0.649° and 0.216°. (OK, a few too many significant figures)
Is it real?Does it matter?
Use TAMU Model to Calculate TMI-WS DifferencesWarm End Differences < 0.05K /Ignore
Deltas Computed with TAMU Model
10V 10H 19V 19H 21V 37V 37H0.34 -1.71 -0.86 -2.98 -1.73 -3.20 -2.49-1.20 -1.48 Including Jim Shiue’s Angles
@ 171K 89 202 137 200 216 156-0.76 -0.92 -1.20 -1.43 -3.37 -3.17 -3.16
@ 281 280 285 284 284 281 281
Deltas are more self-consistent using Jim Shiue’s angles.
“TMI_CC_1.1” based on TAMU model only @ Cold End, U. Mich at Warm End.75% Weight for Windsat, 25% TMI
New angles result in slightly more consistent set of deltas.
Conclusions/Questions
We have a Consensus Calibration that allows us to move forwardWill be updated
TMI 10 GHz Angle Issue is RealIt Matters (a little)
Needs to be Accounted for Where TMI is being used as a Standard
Similar Problems will recur throughout the ConstellationAccounted for when known Won’t always be KnownRecalibration Will Absorb this Sort of Problem.
Goal of X-CAL is 0.1K consistency.Not all Sensors Will be That GoodStill Not Good Enough For Climate PurposesAlgorithm Team Needs to Think about Another Bias Removal Layer
Average Precipitation in Washington DC is of order 0.1mm/h
Absolute Accuracy is Another StoryHesitate to state an error bar
Spare Slides
TAMU ALGORITHMCOMPONENTS OF UNCERTAINTY
(All Cold End Data) 10V 10H 19V 19H 21V 37V 37H @TB 172 90 206 143 231 219 160 EOF1 -.02 .02 -.04 -.26 EOF2 .02 -.09 EOF3 .16EOF4 .10EOF5 .05EOF6 all < 0.02RH .03 .02 .05 .34 CLHT .02 .03LR .10NET .03 .04 .06 .36NOTES: ALL VALUES IN KELVINSVALUES LESS THAN 0.02K LEFT BLANK BUT INCLUDED IN SUMS
TAMU ALGORITHM UNCERTAINTYAll Cold End Data
10V 10H 19V 19H 21V 37V 37H @TB 172 90 206 143 231 219 160 Mod .03 .04 .06 .36Stat/M .02 .03 .06 .05 .04 .06 .03Net/M .04 .03 .07 .08 .36 .06 .03
Stat/QR all below 0.02Net/QR .03 .02* .04 .06 .36 .02* .02*
NOTES: ALL VALUES IN KELVINSVALUES LESS THAN 0.02K LEFT BLANKUse Red Values, if < 0.02 use 0.02
TAMU ALGORITHMCOMPONENTS OF UNCERTAINTY
(Lowest Quartile Only) 10V 10H 19V 19H 21V 37V 37H @TB 170 88 198 131 219 213 151 EOF1 -.02 .02 -.16 EOF2 -.04 EOF3 .09EOF4 -.06EOF5 .03EOF6 all < 0.02RH .02 .03 .20 CLHT -.02 .05 .06LR -.06NET .03 .06 .06 .20NOTES: ALL VALUES IN KELVINS/VALUES LESS THAN 0.02K LEFT BLANK BUT INCLUDED IN SUMS
TAMU MODEL UNCERTAINTYLowest Quartile Only
10V 10H 19V 19H 21V 37V 37H @TB 170 88 198 131 219 213 151 Mod .03 .06 .06 .20
Stat/M .07 .05 .04 .04 .05 .06 .03Net/M .08 .05 .07 .07 .21 .06 .03
Stat/QR all < 0.02Net/QR .03 .06 .07 .20 NOTES: ALL VALUES IN KELVINSVALUES LESS THAN 0.02K LEFT BLANKUse Red Value
Relationship to PPSWhen tasks become routine we pass them over to PPSMatchup/Monitoring softwareData Set support (1C and “Base” files)
Consensus Calibration EffortIn general, can we get a better calibration using multiple sensors?Try with TMI/Windsat combination
SoundersWe are in the process of organizing a methods comparison for sounders
AMSR-EWe have a preliminary analysis of JAXA AMSR-E data setUpdated data set availableNeed additional data that PPS can provide
SSM/INext up/ CSU “Base” files available soon
CC_1.1 Improves Self-Consistency of TMI DrasticallyCC_1.X Improves Self-Consistency of WS Significantly
Updates to Consensus Calibration
CC_1.2McKague suggested more rigorous method of unifying teams’ results.
Requires uncertainty estimates for each team’s numbersWe now have a second warm end estimate (UCF)
Updates can come fairly often before launch of Core.Improved techniques/Additional InstrumentsWe’re still figuring out the details of what we’re doing
GMI should have significant weight in Post Launch CC
Soon after Core launch, stability in CC will become much more important.Update approval mechanism will be needed.
Basics of the TAMU Model
3 Reasons to Present: Collaboration/ Example/Tool Used Here
Source Sensor (e.g. WS) Surface & AtmosphereTarget (e.g. TMI)
Adjust Surface & Atmosphere for best fit to Source RadiancesIterate until Adjustments are < 0.01K
For corresponding channels DTb is the double difference DTb = (Tsource – Ttarget)obs - (Tsource – Ttarget)calc
Surface = Elsaesser Model (Wilheit/Kohn model has been used for comparison)Modified to allow for negative windspeedsAdjust SST & WS
Atmosphere RT Models from RosenkranzModified to allow for Negative Cloud Liquid Water Cloud @ 4-5kmFixed RH Profile/Lapse RateAdjust CLW and Temperature at bottom of Atmosphere (PW follows)
Uncertainties in the TAMU Model
Key Assumptions in the TAMU Model
Cloud Location 4 to 5km Insignificant Contributor to Uncertainty
Lapse Rate From Co-located GDAS: 6.26 ± 0.30 K/km Minor Contribution to Uncertainty
Relative Humidity Profile Mean and Covariance Matrix from Co-located GDAS Compute EOFs for uncertainty Primary Source of Uncertainty