Latest Terra-MODIS Ocean Color Radiance Corrections

1
Latest Terra-MODISOcean Color Radiance
Corrections

• Inter-detector, Mirror Side,
• Cross-Scan, and Gain Adjustments
• Bob Evans, Ed Kearns, Kay Kilpatrick
• Rosenstiel School of Marine and Atmospheric
  Science
• University of Miami
• MODIS Science Team Meeting
• Baltimore, MD. July 2004

2
Examples of Instrument effects before and after
corrections Figures 1a and b show the effects
of mirror side and inter-detector banding. These
instrument artifacts are a result of incomplete
polarization correction (10 km wide mirror side
banding) and detector gain characterization or an
as yet unidentified source and introduces trends
across the focal plane (1km stripes and mirror
side trends). Mirror side difference, Figure 1a,
is AOI and time dependent, order 1.5 Lt Figure
1b shows the average detector-to-detector trends
in gain that are necessary to minimize
cross-focal plane trends and detector stripes,
order 0.2 Lt.
3
(No Transcript)
4
Typical Problems

• Single-pixel high stripes in 1km data
• Interdetector mismatches in bands
• 10-pixel high stripes
• Mirror-side differences
• Bias and RVS
• Edge-to-Edge of Swath discontinuities
• Unresolved RVS variations
• Temporal inconsistencies in nLw
• LUTs m1 values not precise enough
• Low-level changes

5
Basic Methodology

• Use Terra-MODIS time series at Hawaii
• gt1200 days/granules
• MOBY time series
• Break time series into epochs
• Accounts for instrument changes not well resolved
  in m1s

6
Interdetector balancing remove single-pixel
striping

• Correct on a per band, per epoch basis
• Use a modal analysis to determine peak response
  of each detector
• Requires multiple granules per epoch for adequate
  resolution
• Adjust modal peak for each detector to match that
  for detector 5
• Relatively easy to achieve
• Same technique can be done with x-scan dependence
  if need be.new polarization table has reduced
  the need somewhat

7
Detector balancing 412nm
8
Detector balancing 512nm
9
Cross Scan corrections

• Requires flat field assumption no significant,
  consistent zonal gradients near Hawaii
• Compute average cross-scan distribution per
  granule for each band, relative to the position
  at pixel 500 (west of nadir for Terra Day)
• Group the granules x-scan behavior within each
  epoch
• Avoid sun glint contamination!!
• Fit a 5th order polynomial in a least-squares
  sense to normalize the distribution
• Quantize this correction into 50 x-scan
  correction values. High AOI shows the most
  change. xscan carries bulk of the correction.

10
Shape of x-scan correctionfor each epoch 488 nm
11
Magnitude of cross scan correction 488nm
12
Shape of x-scan correctionfor each epoch 412nm
Low AOI
High AOI
13
Balance Mirror Side 2

• Add detector bias and cross-scan correction to
  2nd mirror side to match first mirror side
• Does not require a significant flat-field
  assumption, but assumes that the optical
  meridional variability in the ocean near Hawaii
  on the scale of 10km is small
• Compute average mirror side difference for each
  granule as a function of x-scan position
• Fit a 5th order polynomial to the mirror side
  differences for all granules within an epoch
• Quantize this as a 50 element correction vector

14
Net correction applied to Mirror 2 443nm
east
East high aoi
West low aoi
nadir
15
Net correction applied to Mirror 2 412nm
west low AOI
East high AOI
nadir
16
Example Day 2004 1131km L2 magnified

• nLw 412nm
• Hardest to do
• Note relatively good corrections right up to sun
  glint
• Works well over wide range of Lts

17

• Day 22, Year 2004
• nLw 412nm
• Note mirror-side striping
• only in upper right

18
Setting the final gainsnLw to Lt to nLw to Lt

• Radiance deficits and anomalies are detected and
  measured in nLw space (L2)
• Corrections are applied to Lt values (L1B)
• Requires a recursive method
• Estimate nLw radiance error
• Scale to Lt
• Per band
• Systematically underestimate to account for
  variability
• Repeat until solution converges

19
Gain correction

• Implement an overall exponential correction for
  blue Bands 8 and 9 (based on mean MOBY behavior),
  not per epoch
• Adjust Bands 10, 11, 12 on a per epoch basis
• Adjust MODIS (Band X / Band 9) to match MOBY
  (Band X / Band 9) ratio
• Want ratios (products) to be stable
• Aim for low RMS
• Neglect MOBY v. MODIS individual band bias
• Adjust 748nm band (linear/exponential) to remove
  long-term trends in mean epsilon fields

20
Modal time series

• Use a time series of the mode of each MODIS
  bands data in a small concentric circles (30,
  10, and 3 km) around MOBBY with increasing
  weights
• Compare to filtered MOBY data time series

21
Time series 443

• Modal time series
• difference from MOBY
• (red points matchup pairs, blue bar epoch
  average)
• ratio mode/measured

22
Time series 551nm

• Modal time series 551nm
• difference from MOBY
• (red points matchup pairs, blue bar epoch
  average)
• Time series blue/green ratio

23
Net gains in time

• Gains are computed per band for each epoch
• Multiply the gain by the xscan correction for
  west, center, and east side of scan for each epoch

24
443nm net correctiongain xscan for each epoch
west low AOI
East high AOI
nadir
25
Top panel - radcor gain vs time, purple line is
band 8 gain adjustmentBottom panel - deviations
of the measured m1 coefficients from the linear
trends band 8, mirror side 0 (corrected) from
Gerhard Meister Radcor oscillations factor 5
large than SD M1 trend residualsNote Radcor
time resolution 60 days, M1 two weeks
26
551nm net correctiongain xscan for each epoch
west low AOI
East high AOI
nadir
27
Gain corrections for MODIS-TerraRed - 667,
678nm, Green - 551nm, Yellow - 531nm, Blue -
443nm, Purple - 412nm
TERRA 041 reprocessing gains
Calibration adjustments developed for
current MODIS ocean reprocessing Blue bands fit
to winter MOBY nLw observations to minimize
sun-glint and polarization issues. Green bands
adjusted to match MODIS to MOBY blue/green band
ratios.
Days since Jan 1, 2000 to Dec 31, 2003
28
Smoothing net corrections exponential least
square regression551nm
Example Nadir xscan
29
Improvements in terra_v24_49

• Smoother gain evolutions
• No seasonal oscillations
• No need for additional smoothing
• More predictable x-scan and mirror side
• Overall we have been able to remove much of the
  problems in the l1b data that were present in the
  transition between 2003 and 2004.

30
Bias and RMS terra_v24_49
Band Bias RMS
412nm 1 7
443nm 7 5
488nm 9 7
531nm 8 9
551nm 8 12
N 15
31
Remaining Problems

• Remaining sun glint contamination (x-scan)?
• Still some odd behavior, variable in time space
• No other guidance for red bands (fluorescence)
• Requires set balancing of 667/678nm ratio
• Some per-detector xscan behavior remains (though
  less than before). The time trend in the cross
  scan correction at high AOI likely indicates a
  change in the mirror polarization.
• Still data and time intensive
• Terra_v24_50 still has a few issues to be fixed
• Red gains still maladjusted for 2 or 3 epochs
• Inter-detector balance off for 2 epochs for Band
  12

32
Smoothing and prediction (experimental)

• Fit bicubic splines under tension to net
  cross-scan behavior
• Caveats avoid summer times (residual glint)?
  Edge of scan?
• Produces look-up tables and functional
  relationships for corrections
• Limited predictive capabilities (good enough for
  short term?)

33
(No Transcript)
34
Day of year 2000
xscan pixel
35
(No Transcript)
36
xscan pixel
Day of year 2000