Implementation of the Instrument Only Correction

Size: px

Start display at page:

Download "Implementation of the Instrument Only Correction"

Marvin Simon
5 years ago
Views:

1 Implementation of the Instrument Only Correction Sidharth Misra and Shannon Brown Jet Propulsion Laboratory, California Institute of Technology 03/29/2016 Copyright 2016 California Institute of Technology Government sponsorship acknowledged

Wiggle Correction VFC locking and Aquarius Ca = antenna counts, Cr = reference load counts, G = counts/k, Tr = reference temperature, Ta (K) Tr (K) Va/r VFC Fa/r Counter Ca/r (1) VFC has a propensity

2 Wiggle Correction VFC locking and Aquarius Ca = antenna counts, Cr = reference load counts, G = counts/k, Tr = reference temperature, Ta (K) Tr (K) Va/r VFC Fa/r Counter Ca/r (1) VFC has a propensity to lock up at frequencies (2) Causing locked frequency outputs even though V is changing (3) Causing locked counts even though T is changing Histogram of counts " TA = C A C $ R # G % '+T R & Reference load wiggles Systematic antenna bias (4) Resulting in noisy reference counts to be biased towards a locked count when averaged 2

3 How is the correction implemented? Step 1: Derive correction using difference between consecutive reference load count samples Step 2: Normalize mean correction to zero (this ensures that wiggle correction is not impacted by gain changes in the system and does not add its own systematic drift) Step 3: Provide wiggle correction table with respect to reference load count to GSFC Step 4: GSFC takes the wiggle correction table which is a function of counts to generate a timeseries of the correction for each channel Step 5: Correct reference load counts ONLY to calculate TA " TA' old = C A C R % $ '+T R # G & " TA' new = C A C R f C R $ # G Step 6: Process nominally ( ( )) % ' +T R & G L C Ta Ta APC Tb O 3

4 Reference Count Correction All channels 4

5 Wiggle Correction Conclusions Wiggle correction has a zero bias with respect to reference load This ensures NO long-term drift is introduced due the correction The wiggle correction is different for all six channels and is separately applied This indicates that any annual signature is not due to the wiggle correction The wiggle correction is applied ONLY to the reference load This indicates that seasonal or local changes are probably due to some other anomaly V3 and H2 show anomalous behavior at the end of the mission Wiggle correction table suggests that this behavior might be a separate calibration issue We are still looking into this 5

6 Looking Ahead Drift calibration It is possible to achieve drift calibration without HYCOM Is it necessary though? Systematic bias in ocean measurements As shown before, the VFC phenomenon does impact ocean counts The impact is seasonal, local and gain drift related BUT systematic Probably <0.1K Do we worry about this? ( C f ( C )) ( C f ( C )) A A R R TA final = + G T R Unearthed calibration issues Going to an instrument only correction has revealed previously hidden calibration issues Annual signatures, long-term drift, mission-end drift, seasonal structures We should revisit complete counts à TA à TB calibration Talk from Shannon will discuss many of these 6

7 Backup 7

8 How is the correction implemented? Step 1: Derive correction using difference between consecutive reference load count samples Step 2: Normalize mean correction to zero (this ensures that wiggle correction is not impacted by gain changes in the system and does not add its own systematic drift) Step 3: Provide wiggle correction table with respect to reference load count to GSFC Step 4: Correct reference load counts ONLY to calculate TA " TA = C A C R % $ '+T R # G & 8

9 Reference Count Correction V3 and H2 9

10 Aquarius calibration issues Drift: All six channels of the Aquarius radiometer have indicated a drift with respect to ocean model Exponential correction applied to noise-diode temperature to correct drift Wiggles Pseudo-periodic oscillation in the data that are different for all six channels Root cause backend Voltage to Frequency Converter (VFC) locking issue impacts reference load counts Systematic pseudo-random bias There is a potential that the same phenomenon that causes wiggles also causes bias in the measured antenna temperature that would appear pseudo-random in nature but is in fact systematic (T a T exp ) Aquarius beam 2 V pol Correction and status of all three discussed here 10

11 (2) Wiggle Correction Antenna Counts Histogram " TA = C C % A R $ '+T R # G & 11

12 (2) Wiggle Correction Correction Derivation Reference load V3 Reference load diff V Reference load diff binned with counts V3 TA correction V

13 (2) Wiggle Correction TA correction Examples shown for 2 out 6 channels Wiggles can be corrected Currently in v4.1 Residuals remaining

(1) Drift correction " TA = C A C $ R # C RND C R % 'T ND +T R & Drift potentially caused by

+ C off ( ) + C off C A = G T A +T RX C R = G T R +T RX 6 unknowns 5 equations ( ) + C off C

C off Impossible to calibrate the calibrator using instrument only parameters Need an external

14 (1) Drift correction " TA = C A C $ R # C RND C R % 'T ND +T R & Drift potentially caused by out-gassing during the first couple of months Directly impacts noise-diode of all channels ( ) + C off ( ) + C off C A = G T A +T RX C R = G T R +T RX 6 unknowns 5 equations ( ) + C off C RND = G T R +T ND +T RX ( ) + C off C AND = G T A +T ND +T RX C ACND = G( T A +T CND +T RX ) + C off Impossible to calibrate the calibrator using instrument only parameters Need an external constraining source Prefer NOT use Ta ocean model Antarctic model used for relative calibration 14

(1) Drift correction Ice model Coupled thermodynamic/radiative transfer model MEMLS model (Wiesmann and Matzler, 1999) used to compute upwelling TB Heat transport equation

function of density (Paterson, 2000) Depth (m) Summer Ice Temperature Winter Summer -76 o, 45 o Tuned using multi-frequency AMSR- E TBs and in situ surface temperature

15 (1) Drift correction Ice model Coupled thermodynamic/radiative transfer model MEMLS model (Wiesmann and Matzler, 1999) used to compute upwelling TB Heat transport equation solved for ice T(z,t) profile Surface temperature values obtained from near by AWS station (JASE) used as top boundary condition Thermal diffusivity increases as a function of density (Paterson, 2000) Depth (m) Summer Ice Temperature Winter Summer -76 o, 45 o Tuned using multi-frequency AMSR- E TBs and in situ surface temperature data Generated random snow layer structures to find a realization that gave best fit 6-37 GHz V&H-pol TBs Ice dielectric model from Tiuri et al., (1984) gave best fit AMSR-E data 15

16 (1) Drift correction Vicarious Drift Correction + Double Difference T A,v3_ice T Amodel,v3_ice Solve for exponential fit on ice T A,v3_ice - T Amodel,v3_ice Calculate Tnd correction from fit apply to ocean T A,v3_ocn,new Exponential fit derived over ice for V3 once scaled by the ratio Tocean/Tice fits ocean ΔTA exactly bias A,v3_ocn = T A,v3_ocn,old - T A,v3_ocn,new bias A,pb_ocn = [T Amodel,v3_ocn T amodel,pb_ocn ] [T A,v3_ocn,old -bias A,v3_ocn T A,pb_ocn,old ] Solve for exponential fit for other channels using bias A,pb_ocn

17 (1) Drift correction Vicarious Drift Correction + Double Difference We can use this vicarious-double difference to correct drift without HYCOM Plots above give an example of drift correction derived off the ice vicariously fit to the other channels and compared to the ocean model

18 Identifying Lock Points 18

19 Deriving a distribution based correction Aquarius antenna counts ultimately require the lock-point spikes in the distribution to reduce and it s neighbors to increase A temporal behavior can not be directly derived from a distribution correction 19

20 (3) Pseudo-random ocean bias correction Antenna Counts Histogram " TA = C C % A R $ '+T R # G & 20

21 (3) Pseudo-random ocean bias correction Impact on Antenna Counts Antenna counts have a much larger dynamic range (from land to ocean) than the reference load counts. The offset error introduced due VFC locking on the Antenna counts varies as a function of, Brightness temperature scene changes Seasonal temperature changes Salinity changes Counts drift Due to multiple factors involved, the offset error would look random in nature Even though the noise looks random, it introduces systematic errors in the antenna temperature measurements locally the antenna counts exhibit a similar stability as the reference load counts with additional varying factors This locally introduces a non-random systematic bias to the science measurements that is also temporal in nature. 21

22 (3) Pseudo-random ocean bias correction Identifying Lock Points Black and White For Easier Visualization Counts vs. Days Lock points Aquarius 6 receivers Days (1-600) Counts Counts Counts Days (1-600) Counts Counts Counts 22

23 (3) Pseudo-random ocean bias correction TA impact Above figures show a simulated example of bias introduced to mean ocean TAs due to backend VFC locking Errors in general less than 0.05K (higher for H-pol) Local bias larger during first couple of months of drift 23

24 (3) Pseudo-random ocean bias correction Correction techniques The impact on TA can be corrected in the following ways 1.Introduce random-noise to antenna counts to wash out impact of locking points on the antenna counts Pros: Applied to all antenna counts Pros: Does not introduce its own systematic bias Cons: Increases the noise in the derived salinity data (have margin) 2.Apply correction based off similar shapes derived from wiggle correction by applying constrained probability theory (backup) Pros: Does not increase white noise of the system Cons: Correction might introduce its own systematic bias due to improper assumption Cons: Very hard to verify 3.Alternate recommendation: Ignore systematic bias Current simulated bias is 0.05K to 0.08K for both channels which is below Aquarius requirements 24

25 1. Wiggle Correction Summary Correction currently being implemented at GSFC Will be evaluated over next couple of weeks Secondary calibration issues previously hidden might come through 2. Drift Correction Current initial drift correction based off HYCOM exponential fit We ve demonstrated that an exponential fit using Antarctic Ice model scales with respect to ocean drift for V-pol beam 3 This correction can be vicariously applied to other channels and polarizations 3. Pseudo-random ocean bias There does exist a systematic offset bias that is dependent on salinity, surface temperature, instrument count drift and varies over time Correction of such bias is possible but not trivial Initial simulations of bias generally <0.08K and this TA impact can potentially be ignored 25

26 Removing the Lock Points x new ( t lock ) = " $ # %$ x old ( t lock ) 1, if slp(t lock ) > p x old ( t lock ) +1, if slp(t lock ) < p We calculate the slope of the time domain signal at every locking point If the signal is rising, chances are the signal is locked higher (left figure) and vice-versa (right figure) We optimize for the slope value p to redistribute the histogram 26

27 Histogram Lock Points Detected Pre-correction Post-correction 27

28 Resulting systematic biases Initial Analysis <0.1K peak to peak bias depending on channel being observed 28

29 Challenges Ahead Still ways to go before this correction can be implemented in Aquarius v5.0 Verification is a big challenge We do not want to add systematic bias of our own We can not compare with the HYCOM model since we are correcting localized variations Needs to be compared with localized ARGO regions Is this even a problem? Lock point identification not complete Channel 1 Beam 2 has the lowest ocean counts, making it very difficult to retrieve the locking points using above method At certain locations due to high density of locking points, some lock points get missed Locking spikes magnitude not completely equal 29

30 What is locking? (or flat-spot ) TBà (antenna)à TAà (radiometer front end+detector)à Và (VFC)à Fà (counter)à C VFCs are responsible for converting voltage proportional to the Tb measurement to counts VFC (Voltage to Frequency Converter) can lock on certain frequencies due to the presence of an interfering clock signal Spike translates to a flat-spot in VFC response. The signal gets locked on to a particular count value (voltage) After launch, odd feature in histogram of Jason-2 AMR 34 GHz TB observed Traced to VFC flat spot issue proven by lab test with AMR spare hardware VFC locked onto 9001 counts due to interference with another clock in the system Biased noise diode measurements over ocean Jarnot et al., MLS calibration report 30

Simulated Locking Error on Antenna Temperature As an example, we derived simulated antenna counts from the model antenna temperature and measured gain and offset of the radiometers We added an offset

31 Simulated Locking Error on Antenna Temperature As an example, we derived simulated antenna counts from the model antenna temperature and measured gain and offset of the radiometers We added an offset error to the simulated counts similar to the wiggle errors observed on the reference counts We re-derived the antenna temperature values and subtracted the original antenna temperature samples Latitude 31

32 Summary Aquarius Voltage to Frequency Converters (VFC) get locked at certain frequency locations and this impact is clearly observed in the histogram of the uncalibrated counts We have already applied a correction to the reference load counts causing TA wiggles Locking counts at the antenna scene counts causes non-random systematic temporal biases in the ocean salinity retrieval that is hard to detect We have developed a preliminary method to identify these counts and apply a correction from the counts distribution to temporal samples Method needs to be verified and tweaked further 32

Microwave Radiometer (MWR) Counts to Tb (Brightness Temperature) Algorithm Development (Version 6.0) and On-Orbit Validation

Microwave Radiometer (MWR) Counts to Tb (Brightness Temperature) Algorithm Development (Version 6.0) and On-Orbit Validation Zoubair Ghazi CFRSL Central Florida Remote Sensing Lab Dissertation Defense