Analysis of Nonlinearity Correction for CrIS SDR April 25, 2012 Chunming Wang NGAS Comparisons Between V32 and V33 Engineering Packets.

Analysis of Nonlinearity Correction

for CrIS SDR

April 25, 2012

Chunming Wang NGAS

Comparisons Between V32 and V33 Engineering Packets

2

Expected Linearity Improvement Using v33 Engineering Packet Parameters is Confirmed

• Detailed analyses of residual nonlinearity were performed using the Golden Days data and data from April 15, 2012

– Convergence of statistics were examined

– Distribution of scene brightness temperature, FOV to FOV differences in brightness temperatures were examined

• Stratification of statistics using mean brightness temperature for each FOR provided valuable information on linearity of the detectors

– Change in the magnitude of nonlinearity as a function of mean brightness temperature relative to ICT were analyzed

– Sensitivity of brightness temperature to small radiance variation for low temperature scene were taken into consideration

• Expected improvement in linearity using v33 parameters is confirmed– Independent processing of RDR using NGAS off-line code provided additional

confirmation

Updated Parameters Substantially Improves Linearity of CrIS SDR

3

IDPS Generated SDR Products for April 15 Were Used in the Analyses

February 24 April 15

• Standard IDPS SDR products showed stable quality– No obvious anomalous radiances were detected; small data gap is due to delay in

data delivery to NGAS

– Expected warming in Northern hemisphere and cooling in Southern hemisphere were visible

4

Differences in Brightness Temperatures of LWIR FOVs from FOR Mean Were Reduced

February 24 April 15

• Ensemble averages of brightness temperature difference of each FOV to the FOR mean were substantially reduced

– All Earth scenes were used without rejection by variation in brightness temperatures among 9 FOVs

– Standard deviations of the differences due to geometric effects were unchanged

FOV5 FOV5

Side FOVs Side FOVs

Corner FOVs Corner FOVs

5

Meam Differences in Brightness Temperatures Among MWIR FOVs Were Greatly Reduced

• Substantial improvement for FOV7 and FOV8 were observed– FOV7 and FOV8 are now in family with the rest of FOVs

– Residual differences are at similar magnitude as the difference between FOV9 and FOV6 which were shown to be basically linear during TVAC tests

February 24 April 15

6

Statistics of SWIR FOVs Were Unchanged Due to Identical Processing Parameters

February 24 April 15

• The brightness differences from FOV to FOV were substantial– In-depth analysis of the distribution of these differences show the detectors are

basically linear

– Brightness temperature differences seem to be linked to geometry

Analyses Methodology

8

Key Issues Concerning the Analysis Methodology Were Investigated

• Convergence of statistics is achieved using one day of data – One or two orbits data may not be sufficient

– Convergence in average brightness temperature is slower than average differences from FOR mean

• Effect of scene brightness relative to ICT is taken into consideration– When scene brightness if very close to that of ICT nonlinearity effect is minimized

– At very low temperature scene brightness temperature is sensitive to radiance uncertainty

• Separation of nonlinearity from other sources of errors– Identify signatures of nonlinearity

– Independent processing of RDR using NGAS off-line code provided additional confirmation

Confidence in Conclusion is Gained by the Validation of Methodology

9

Using Spectrally Averaged Channel Brightness Temperature Reduces Effects of ILS Errors

• Spectral resampling helps reduce effects of spectral calibration uncertainties

– Averaging in brightness temperatures space is preferred because of the flatness of Earth scene spectra in brightness temperature

• Nonlinearity is an effect on the broad spectrum

– Overall nonlinearity is a function of the radiance energy over the entire band

– Spectral resampling does not affect dynamic range of spectra

10

Convergence of Brightness Difference Requires Averaging Over 3 Orbits of Data

• Convergence of mean brightness temperature is slow due to bi-modal distribution of radiances

– Mean brightness temperatures for all FOV changes simultaneously

– It requires more than 3 orbits of data to bring the average FOV to FOV difference to within 10% of its final value

Co

nve

rge

nce

of

Sp

ect

raC

on

verg

en

ce o

f D

istr

ibu

tion

of

Diff

ere

nce

to

FO

R m

ea

n

Co

nve

rge

nce

of

Dis

trib

utio

n o

f E

art

h S

cen

e B

righ

tne

ss

11

Convergence of Brightness Difference Requires Averaging Over 3 Orbits of Data

• Convergence for MWIR seems faster than LWIR band

– More than 2 orbits of data is required to bring the average FOV 2 FOV differences in brightness temperature to within 10% of its final value

Co

nve

rge

nce

of

Sp

ect

raC

on

verg

en

ce o

f D

istr

ibu

tion

of

Diff

ere

nce

to

FO

R m

ea

n

Co

nve

rge

nce

of

Dis

trib

utio

n o

f E

art

h S

cen

e B

righ

tne

ss

12

Brightness Temperature Error Due to Nonlinearity Depends on Scene Brightness

BT Range, smoothed channels

BT RangeDesignatedWindow channels

ICT TemperatureMin,Max

Mean BT

• Earth scene spectrum has different brightness temperature for all channels

• Warmest channels carry most of photon energy

– A subset of window channels is selected for each band to represent the brightness of the scenes

– Average of all FOVs is used to classify the brightness of a scene

13

Each Earth Scene (FOR) is Classified into one of 50 Groups According to Its Brightness

• Bi-modal distribution of the Earth scene brightness is consistent with channel brightness statistics

– Large number of Earth scenes are warmer than ICT

– Since Earth scene spectrum is not constant in brightness the total energy is lower than black body at the same brightness

• ICT temperature varies over a very small range

14

FOV-to-FOV Brightness Temperature Differences Depend on Scene Temperature

High Temperature Scenes Low Temperature Scenes

LWIR

MWIR

SWIR

FOV6-FOV9

FOV6-FOV9

15

Examination of the Joint Probability Distribution Reveals Scene Dependence of BT Differences

February 24LWIR932.5 cm-1

Sce

ne

Brig

htn

ess

BT DifferenceFrom FORMean

16

Wider Spread of Distributions in BT Difference for Cooler Scene is Due to Higher Sensitivity

• Constant perturbation in radiance space leads to larger changes in brightness temperature for cooler scenes

– Wider spread of difference in brightness temperature among FOVs is due in part of this sensitivity

• Very warm scenes are also more likely to be cloud free

– Cloud free scene may be more uniform than cloudy scenes

17

Examination of Joint Probability Distribution for MWIR FOV Helps Us Recognize Nonlinearity

Nonlinear FOV

Linear FOV

February 24,2012MWIR1275 cm-1

Large Difference Away

from Calibration Points

18

Correction with v33 Engineering Parameters Nearly Completely Removed Nonlinearity

April, 152012MWIR1275 cm-1

19

Residual Nonlinearity for LWIR Are Significantly Reduced for FOV9 with v33 Parameters

April 15LWIR932.5 cm-1

20

Examination of the Joint Probability Distribution Shows SWIR Detectors Are Mostly Linear

February 24SWIR2535 cm-1

21

Statistical Results for SWIR Band Are Highly Consistent for Two Focus Days

April 15SWIR2535 cm-1

22

Empirical Data from Two Days Seem to Suggest Geometric Trend in BT Bias for SWIR

• Brightness temperature biases seem to be linked to the position of the FOVs

– Both days of data show the similar trend

• More in-depth analyses are needed to determine the cause of these biases

– Analyses of DS and ICT raw spectra are needed

FOV2FOV1 FOV3

FOV5FOV4 FOV6

FOV8FOV7 FOV9

23

Conclusion

• Residual nonlinearity for all detectors are very small– Joint probability distribution of the Earth scene brightness and brightness

difference is very useful in identifying nonlinearity

– SWIR detectors are all linear

• SWIR band FOV-to-FOV biases may be caused by non-uniformity of the calibration targets

– More analyses are on-going

• Methodology can be used to monitor nonlinearity