1. Double differencing of IASI-A/B against Meteosat/SEVIRI IR Channels
Tim Hewison (EUMETSAT)
Introduction: EUMETSAT Calibration Team + GSICS IR Sub-Group + co-authors
Aims: To introduce GSICS and our activity to generate the IR Reference Sensor Traceability & Uncertainty Report
Also, consideration for the handling of processing changes to the reference sensors (such as IASI).

2. Overview Double-Difference MetopB/IASI - MetopA/IASI
IR Reference Sensor Traceability & Uncertainty Report
Possible strategy for migrating GSICS products from IASI-A to IASI-B

3. GSICS GEO-LEO IR Double Differences
Time series of Bias
in Meteosat-10/ SEVIRI IR13.4
wrt IASI-A
wrt IASI-B
For standard scene radiance (267K)
Over 1st 4 yr overlap
Biases vary
Ice contamination
Range -0.4 to -2.7K
Differences <0.1K
The SEVIRI imagers on MSG generally have much smaller biases, except the 13.4µm channel shown here
In fact, we can use the GSICS Corrections to estimate the bias wrt two reference instruments – IASI A & B
- These vary of their first 3 years overlap period
With small, systematic differences

4. GSICS GEO-LEO IR Double Differences
Time series of Bias
in Meteosat-10/ SEVIRI IR13.4
wrt IASI-A
wrt IASI-B
For standard scene radiance (267K)
Last 4 yr overlap
Biases vary
Ice contamination
Range -0.5 to -2.0K
Differences <0.1K
The SEVIRI imagers on MSG generally have much smaller biases, except the 13.4µm channel shown here
In fact, we can use the GSICS Corrections to estimate the bias wrt two reference instruments – IASI A & B
- These vary of their first 3 years overlap period
With small, systematic differences
Operational GSICS Products – Zoom on period Jan Nov 2017
- IASI-B step change -0.1K in August 2017

5. Time series of Double Differences (SEVIRI-IASIA)-(SEVIRI-IASIB)
No Obvious Trend in Any Channel! 
We can extend the analysis to calculate the double difference between IASI-B-A in each of SEVIRI’s 8 IR channels
No trend in any channel over this 4 year period
Good news – implies their calibration is stable – so we can combine whole period
However, there are small differences in the long-wave channels
Small differences in long-wave channels 

6. Statistics of Double Difference Time Series
(MSG3-IASIA)-(MSG3-IASIB) OPE RAC Std Bias over / :
Channel
Double Difference Trend [K/yr]
Mean Double Difference [K]
IR3.9
 -0.003 +
  0.004
-0.010
0.005
IR6.3
 -0.007
-0.002
0.004
IR7.4
  0.003
0.001
0.003
IR8.7
 -0.005
  0.002
-0.013
IR9.7
 0.003
  0.008
-0.048
0.007
IR10.8
 -0.008
-0.024
IR12.0
-0.029
IR13.4
 -0.006
-0.040
No statistically significant trend
in any channel
Within standard uncertainty of ~3mK/yr
Can combine entire period
Consistent results from other Meteosats
But larger uncertainties
No statistically significant difference
between IASI-A and -B
in Short- and Mid-bands
in any channel
Small, but significant difference
in long-wave band
Larger for colder scenes
Table just says the same thing in numbers
Highlighting the relative stability, which can be measured to ~3mK/yr
And the small size of the relative bias ~40mK

7. 2013-03/2017-03 (SEVIRI-IASIA)-(SEVIRI-IASIB) - Tb
Mean Difference dTb [K]
Channel
50% [cm-1]
200
210
220
230
240
250
260
270
280
290
300 K
IR13.4
714
782
-0.30
-0.24
-0.19
-0.15
-0.12
-0.09
-0.06
-0.04
-0.02
0.00
0.02
IR12.0
800
870
-0.16
-0.13
-0.11
-0.07
-0.05
-0.03
IR10.8
885
971
-0.27
-0.21
-0.10
-0.08
IR9.7
1018
1047
-0.14
-0.01
IR8.7
1124
1177
IR7.3
1316
1409
0.01
IR6.2
1493
1724
0.03
IR3.9
2385
2751
Meteosat-10/SEVIRI IR
But the magnitude of these double differences strongly varies with scene radiance
- It is much larger for colder scenes

8. Time series of Double Differences – IASIB-IASIA - Zoom 2016/7
Changed IASI-B on-board processing
Improved non-linear correction
Will update for IASI-A in 2018
And zooming in on the last 2 years highlights a change in the IASI-A/B double differences
Corresponding to the change in the onboard processing introduced on 2 Aug to improve the non-linear correction
Causing the previous small negative bias in IASI-B to become a small positive one of similar magnitude
This is expected to change again to ~zero when the counterpart change is rolled out to IASI-A in 2018 (which is smaller)

9. Overview Double-Difference MetopB/IASI - MetopA/IASI
IR Reference Sensor Traceability & Uncertainty Report
Possible strategy for migrating GSICS products from IASI-A to IASI-B

10. IR Reference Sensor Traceability & Uncertainty Report
Aims
To support choice of reference instruments for GSICS
To provide traceability between reference instruments (IASI, AIRS, CrIS)
To seek consensus on uncertainties in absolute calibration of reference sensors
By consolidating pre-launch test results and various in-flight comparisons
Limitations
No new results, just expressing results of existing comparisons in a common way,
reformatting where necessary, to allow easy comparisons
1. Error Budget & Traceability
Focus on radiometric and spectral calibration – for AIRS, IASI, CrIS
2. Inter-comparisons
Introduction: Pros and Cons of each method
Direct Comparisons: Polar SNOs, Tandem SNOs (AIRS+CrIS), Quasi-SNOs,
Double-Differencing: GEO-LEO, NWP+RTM, Aircraft campaigns
Other Methods: Regional Averages (“Massive Means”), Reference Sites (Dome-C..)
So within GSICS we have started to develop a report to …

11. Suomi-NPP CrIS Radiometric Uncertainty Estimates – D. Tobin
Differential error analysis of the calibration equation, aimed at providing a useful estimate of the absolute accuracy of the mean of a large ensemble of observations. Input parameter uncertainties are based on the design of the sensor and engineering estimates of the calibration parameters; i.e. no external information via external “Cal/Val” used.
Tobin et al. (2013), Suomi-NPP CrIS radiometric calibration uncertainty, J. Geophys. Res. Atmos.
Example 3-sigma RU estimates
for a typical warm, ~clear sky spectrum:
Simplified On-Orbit Radiometric Calibration Equation:
REarth = Re {(C’Earth – C’Space) /(C’ICT-C’Space)} RICT ‘ with:
Nonlinearity Correction: C’ = C  (1 + 2 a2 VDC)
ICT Predicted Radiance:
RICT = eICT B(TICT) + (1-eICT) [ 0.5 B(TICT, Refl, Measured) B(TICT, Refl, Modeled)]
Parameter
Nominal Values
3-s Uncertainty
TICT
280K
112.5 mK
eICT
0.03
TICT, Refl, Measured
1.5 K
TICT, Refl, Modeled
3 K
a2 LW band
0.01 – 0.03 V-1
V-1
a2 MW band
0.001 – 0.12 V-1
– V-1
A nice example of the first is the differential error analysis of S-NPP/CrIS conducted by Dave Tobin:
Where he takes the basic radiometer equation and estimates the uncertainty on each input term,
and propagates these through the equation to estimate the Radiometric Uncertainty (RU) on the calibrated radiance
Again, expressed here in BT – and as 3-sigma, which is generally <0.2K
and ~0.1K for warmer scenes and the mid- and short-wave channels
RU is generally 0.2K 3-sigma or less, and similar results for JPSS-1 CrIS
Currently working to include (relatively small) polarization effects into the calibration algorithm and corresponding RU estimates

12. IR Reference Sensor Inter-Comparisons
Mean Difference dTb [K]
Pseudo Channel [cm-1]
Min Freq [cm-1]
Max Freq [cm-1]
200
220
240
260
280
300
Start date
655
650
660
-0.11
-0.10
-0.08
NaN
End date
665
670
-0.09
Start time
21:53
675
680
-0.06
-0.07
End time
23:21
685
690
-0.05
Min Latitude [°]
-74.31
695
700
-0.04
Max Latitude [°]
74.31
705
710
-0.03
Min Longitude [°]
715
720
Max Longitude [°]
180
725
730
-0.02
Min Scan Angle [°]
735
740
-0.01
Max Scan Angle [°]
3.5
745
750
-0.12
755
760
Collocation method:
Big circle SNO
765
770
Collocation dist [km] 50
775
780
-0.50
Collocation time [s]
1200
785
790
-0.15
Collocation sec(theta)
795
800
Filtering applied
see ref
805
810
-0.24
815
820
-0.17
Algorithm Ref
JGR Tobin 2016
825
830
-0.18
Dataset Ref
835
840
Monitored Instrument
IASI-A L1C
845
850
-0.16
Processing Version
latest
855
860
CLASS
865
870
Reference Instrumemt
CrIS NSR SDRs
875
880
-0.13
CSPP
885
890
895
900
Form consensus on relative calibration
Re-binning results of existing comparisons
Biases with respect to Metop-A/IASI
With standard uncertainties (k=1)
At full spectral resolution
In 10cm-1 bins within AIRS bands
Or average results over broad-band channels
Converted into Brightness Temperatures
For specific radiance scenes
i.e. 200K, 210K, … 300K
For all viewing angles
and/or for specific ranges - e.g. nadir ±10°
Over specific period
Common 4-year period from IASI-B start
Meteosat-10/SEVIRI IR /
Himawari-8/AHI IR /
For the second part we have set up a common spreadsheet to allow us to compare a variety of inter-comparisons between hyperspectral reference sensors
On a common scale to form a consensus on their relative calibration …
For example results are shown here for the double-differences for all the IR channels of Met-10 and Himawari-8
Again as a function of the scene TB
Together with the uncertainties of the comparisons
Both of which follow similar patterns

13. IR Reference Sensor Traceability & Uncertainty Report
Conclusions
IASI Changes
IASI-A/B double difference was stable over >4yr
Small differences in long-wave band – addressed by changes to non-linearity
Gradual roll-out allows characterization of impact
Cannot be completely corrected in re-processing
IR Reference Sensor Traceability & Uncertainty Report
supports choice of GSICS reference instruments
provides traceability between reference instruments
consolidates pre-launch test results and various in-flight comparisons
forms consensus on uncertainties in absolute calibration of reference sensors
So to conclude that: While GSICS now provides operational calibration corrections for the IR channels …
… We are now establishing routine comparison of results to monitor relative calibration
We will use these as part of a report on the Traceability and Uncertainty of our IR reference instruments, which …
These comparisons can also highlight any changes in the references instruments’ calibration
For example, IASI-A/B, where the double difference was stable over >4yr,
With Small differences in long-wave band, which are being addressed by introducing changes to non-linearity
Important that the gradual roll-out of these changes allows characterization of impact
However, they cannot be completely corrected in re-processing and may require an empirical approximation to support the generation of FCDRs

14. Possible strategy for migrating GSICS products from IASI-A to IASI-B
Tim Hewison (EUMETSAT)
Introduction: EUMETSAT Calibration Team + GSICS IR Sub-Group + co-authors
Aims: To introduce GSICS and our activity to generate the IR Reference Sensor Traceability & Uncertainty Report
Also, consideration for the handling of processing changes to the reference sensors (such as IASI).

15. Options for NRTC Generation
Just switch RefA->RefB
Single reference w/jumps
Switch RefA->RefB+∆
After period to test stability
Single reference, no jumps
Merge all Refn+∆n
Multiple refs, no jumps
Prime GSICS Corrections
Just average all* Refs
Multiple refs, with jumps
Assign uncertainty to cover
RefA
∆AB
RefB
∆BC
RefC
Time

16. Pragmatic Approach for NRT GSICS Corrections
IRRefUTable analysis of reference sensors’ performance:
Ascertain which reference instruments consistent
With consensus (of all references)
Within given uncertainty, uLref(λ, Tb)
Accept all qualified references
Quote absolute uncertainty of GSICS Corrections
As uLref(λ, Tb)
Simple!
But acceptable?

17. Options for FCDR Generation
Define ∆ Corrections
Based on stable overlaps
Apply to all* references
* at least all with stable references
Merge together
Applying delta corrections
“Prime GSICS Correction”
Version changes:
Or Split time series
When reprocessing impossible (e.g. onboard changes)
Reprocess whole dataset
Where possible
Need to regenerate FCDR
When any component changes
Introduce review cycles
RefAv1
RefAv2
∆A1B1
∆A2B2
RefBv1
RefBv2
Time