GLI calibration results

1
GLI calibration results for ocean color channels
GLI calibration Group
M. Yoshidac, K. Tanakaa, H. Murakamia, S.
Kuriharaa , J. Inouea, Y. Okamuraa, K. Isonoa, J.
Niekea, T. Hashimotoa, H. Yamamotoa, I. Asanumaa,
K. Iwafuneb, Y.Aokib, H. Yatagaib, H. Kuriharab,,
N. Sekioc, Y. Mitomic, and Y. Sengad, aJapan
Aerospace Exploration Agency (JAXA), bFujitsu
Ltd, cRESTEC, dTokai Univ.
2
Contents

• GLI system overview
• On-board calibration
• Lamp calibration
• Solar calibration
• Stripe noise correction
• Vicarious calibration
• Vicarious calibration by global data set
• Vicarious calibration by ground observation data
• Final selection of the calibration coefficient

3
GLI system overview
Scan mirror A,B
scan the earth surface by mirror rotation
12 detector (1km resolution channels)
For on-board calibrations
4
GLI system overview
Characteristics of GLI channels
Short-wavelength Infrared (SWIR) 4channels
(1km)
Visible and Near Infrared (VNIR) 19 channels
(1km)
1 Knee points of the piece-wise linear gain
channels 4, 5, 7, and 8. 2 Channels 28 and 29
are re-sampled for each 2-km (1/8) on board and
stored in the 1-km level-1B product. 3 Maximum
radiance for linear response. 4 Channel center
and width are derived from GLI spectral
response. 5 SNR at the standard input level
(W/m2/sr/?m) was measured in pre-launch
evaluation tests.
Middle and Thermal Infrared (MTIR) 7channels
(1km)
5
On-board calibration

• Schematic view of the solar lamp calibrator

Lamp Calibration Calibrator halogen
lamp Channel ch1-29 (visible short
infrared) Frequency1/8days
Solar Calibration Calibrator Sun light Channel
ch1-29 (VN,SW) (visible short
infrared) Frequency1/orbit
6
On-board calibration (? Lamp calibration)

• Lamp calibration raw data trend from 1st lamp
  calibration in January

• Raw data of mirror side A increased (especially
  at longer wavelength), and mirror side B
  decreased.
• The change is caused by also the change of
  lamp(calibrator) radiance.

Correct GLI raw data by lamp monitor data
7
On-board calibration (? Lamp calibration)

• Corrected data trend
• corrected data
• GLI spectral integrated data divided by monitor

Lamp A

• Corrected data increased in both mirror sides.

8
On-board calibration (? Lamp calibration)

• Output ratio of mirror side B to A

B/A
Pre-launch

• The ratio increased after launch, and decreased
  gradually on orbit.
• We consider that the reflectance of side A
  decreased due to the influence of launch, but
  recovered since then.

9
On-board calibration (? Lamp calibration)

• Lamp calibration Summary
• Raw data level from mirror side A increased, and
  mirror side B decreased.
• Data corrected by monitor diode is decreased in
  both mirror sides, so decrease of raw data level
  in mirror side B is considered to be caused by
  lowering of lamp radiance.
• For mirror side B, increase of corrected data
  level is small as 1, so we guess the reflectance
  remains on orbit.
• The ratio of side B to A increased after launch,
  so the reflectance of side A is thought to be
  decreased (about 4) due to the influence of
  launch, but recovered since then.

10
On-board calibration (? Solar calibration)

• GLI output trend
• after correction of diffuser incident angle and
  sun-earth distance
• (scan-mirror side B, EL15deg, normalized by 06
  Apr. 2003).

ch1027
ch19

• Solar calibration outputs dropped in channels
  under 700nm(ch13).
• Other channels output is stable in 2.

11
On-board calibration (? Solar calibration)

• Comparison of Solar calibration with Lamp
  calibration

• Solar calibration outputs dropped in channels
  under 700nm.
• Lamp calibration outputs did not drop.
• ? We consider the drop is caused by degradation
  of diffuser.

12
On-board calibration (? Solar calibration)

• Output ratio of mirror side B to A

• The ratio increased after launch, and decreased
  gradually on orbit.
• ? Similar to lamp calibration trend

13
On-board calibration (? Solar calibration)

• Solar calibration Summary
• Solar calibration outputs dropped in channels
  under 700nm, and we consider this is caused by
  degradation of diffuser.
• Other channels output is stable in 2.
• The ratio of mirror side B to A increase after
  launch, and decreased gradually on orbit. It is
  considered that reflectance of side A decreased
  after launch by some reason, and recovered
  gradually.

14
Stripe noise correction

• Method

The de-striping coefficients ak,l,j are applied
to radiometric corrected (as Level-1B processing
except for the de-striping) but not geometric
corrected radiance, Lk,l,j .
Lk,l,jcorrected ak,l,j Lk,l,j , (1) k
pixels, l lines, j bands. We assume ak,l,j are
related to input radiance Lk,l,j and scan-mirror
incident angle ?k,l, i.e, ak,l,jb0i,jb1i,j?
Lk,l,jb2i,j / Lk,l,jb3i,j? Lk,l,j2b4i,j? ?k,l
b5i,j? ?k,l2, (2) i detectors (1-12 or 48
? scan-mirror sides), imod(l-1, 24 or
96)1. We derive b0-b5 by statistical analysis.
15
Stripe noise correction

• Temporal change of the coefficients a

An example of a in Feb, May, Aug, and Oct 2003
(based on det6 Mirror B). Side B (solid) ? ratio
of the detector sensitivity Side A (dotted) ?
ratio of the mirror reflectance (B/A)
CH18 (865nm high)
CH19 (865nm low)
Oct
Aug

• Detector coefficients look stable.
• temporal change of mirror-side difference in the
  high-gain bands ? Stray light?

B/A changes from Feb to Oct
May
Feb
CH24 (1050nm)
CH35 (11?m)

• ltb0?b5gt
• Ver.1 Constant
• Ver.2 function of time

16
Stripe noise correction

• Example of the correction

ltCH19(865nm)gt
ltCH1 (380nm)gt
De-stripe
L1B original
L1B original
De-stripe

• Mirror-side noise remains sometimes in dark
  areas surrounded by bright areas especially in
  the left side of the L1B image
• ?Stray light ?

• Corrected well

17
Stripe noise correction

• Stripe noise correction Summary
• GLI VN/SW channels have two kind of
  stripe-pattern noises
• (1) Detector sensitivity normalization error
• This is caused by calibration coefficients
  derived in the pre-launch evaluation tests or
  some reasons on orbit. It relatively smaller than
  (2), and stable during the mission period except
  for CH1-3.
• (2) Mirror reflectance normalization error
• This is striking sometimes, and it is caused by
  the accuracy of pre-launch A/B-side difference
  table or some reasons on orbit maybe stray
  light. Stray light seems to influence on the
  side-A more than side-B. The A/B-side difference
  looks decreasing during the mission period. The
  amplitude is different from on-board calibration
  results, but the temporal change looks consistent.

18
Vicarious calibration(? Vicarious calibration by
global data set )

• Operation flow

Processed 16 days 02/06, 03/20, 04/06, 04/22,
05/08, 05/24, 06/09, 06/25, 07/11, 07/27, 08/12,
08/28, 09/13, 09/29, 10/15, 10/24 in 2003
Fix two NIR bands (CH13-19) s calibration
coefficients to 1.
LUTs
1-day L1B LTOA and geometry
Aerosol model selection by GLI atmos corr. Look
Up Table (?aM ??a?ma)
SeaWiFS 8-day binned nLw interpolated to GLI
bands through an in-water model (Tanaka et
al.) Pressure (and water vapor) by JMA objective
analysis Column ozone by TOMS
?m?w
Simulated LTOA
Derive correction coefficients by comparison
between GLI L1B and simulated LTOA.
19
Vicarious calibration by global data set

• Results (example)

Simulated and observed GLI CH03 LTOA (412nm) on
2003/10/15 (by CH13-19B)
GLI LTOA
Simulation by GLI NIR RSTAR SeaWiFS nLw
Compare ? Vicarious coefficients
GLI/Simulation
20
Vicarious calibration by global data set
4. Results

• Results (Kvc vicarious coefficients)

Coefficients based on CH13-19B (global average)
temporal change of Mselect
color obs time
GLI channels
Ch1-9
Ch10-19
Ch24-29
color channels
time
time
time

• About 5 change (increase) in CH01.
• Unrealistic large scatter in SWIR channels (due
  to simulation sensitivity)

21
Vicarious calibration by global data set

• Results (Mirror-angle dependency)

Scan-mirror incident angle dependency based on
CH13-19B

• CH1-3 show angle dependencies.

22
Vicarious calibration by global data set

• Summary
• We could derive vicarious coefficients, which
  have enough accuracy for Level-2 processing, (for
  dark target) using global data sets and radiative
  transfer model.
• The coefficients show,
• Band characteristics for all VNIR and SWIR
  channels except for strong absorption channels.
• Scan-angle dependency and its temporal change for
  CH01-03.

23
Vicarious calibration(? Vicarious calibration by
ground observation data )

• Method

rT Top of Atmospheric Reflectance rTA
Atmospheric Reflectance NrW Normalized Water
Leaving Reflectance t0 Diffuse
Transmittance(sun ?sea surface) t Diffuse
Transmittance (sensor ?sea surface)
over ocean
rT rTA tNrWt0
(Gordan and Wang, 1994)
rTA
Atmosphere
Aerosol Optical property
rw
Sea
24
Vicarious calibration by ground observation data

• Results (vicarious coefficients)

Over dark target (ocean)
GLI overestimates the radiance in NIR channels
over dark target.
25
Vicarious calibration by ground observation data

• Results (vicarious coefficients)

Over bright target (land, snow)
Over dark target (ocean)
GLI overestimates the radiance in NIR channels
over dark target. ? The coefficients are about
1 over bright target.
Observed radiance has offset ?
26
Vicarious calibration by ground observation data

• Results (investigate offset possibility
• Do the coefficients depend on the radiance
  over MOBY ? )

Ratio of the observed to simulated radiance (i.e.
inverse of the vicarious coefficients)
ltshorter wavelengthgt Radiance is high enough not
to be influenced by the offset
412nm
443nm
460nm
ltlonger wavelengthgt Radiance is low enough to be
significantly influenced by the offset
749nm
865nm
866nm
- The ratios ltshorter wavelengthgt do not depend
on the radiance. ltlonger wavelengthgt increase
with decreasing simulated radiance.
Support the suggestion that GLI has an offset.
27
Vicarious calibration by ground observation data

• Results (Offset estimation)

Multi-scattering radiance from the reflection by
land and cloud around the calibration site.
865nm
Radiance over dark target (MOBY)
Slope (coefficient over bright targets)
Estimated offset
The estimated offset is about 0.4W/m2/um/sr in
channel 19 (866nm).
28
Vicarious calibration by ground observation data

• Summary
• We derived vicarious calibration coefficients
  using MOBY measurement.
• It is suggested that GLI overestimates the
  radiance in NIR channels.
• The comparison of the results with the results
  over bright targets would suggest that the GLI
  observed radiance has offset versus the simulated
  radiance.
• The estimated offset is about 0.4W/m2/um/sr in
  channel 19 (866nm).

29
Final selection of the calibration coefficients

• Candidates of vicarious calibration coefficients

30
Final selection of the calibration coefficients

• Final decision on GLI Version2 processing

• Vicarious calibration and de-striping will be
  based on scan mirror side-B and detector 6. 
• Level-1B processing will include stripe noise
  correction considered temporal change.
• Level-1B data will not include the vicarious
  coefficients, because we found different
  characteristics between bright and dark targets.
• Calibration team recommends several candidates of
  vicarious coefficients considered temporal
  change and scan-mirror incident angle dependency
  for each application, so users can use a set
  of coefficients among the candidates or all 1.0
  as their own choice in Level-2 processing.

Level-2 (geophysical parameter)
Level-1B (calibrated radiance data)
Level-1A (radiance data)
Users choise
De-stripe
Candidates of coefficients