Latest Terra-MODISLatest Terra-MODISOcean Color Radiance Corrections Ocean Color Radiance Corrections

Inter-detector, Mirror Side, Inter-detector, Mirror Side, Cross-Scan, and Gain AdjustmentsCross-Scan, and Gain Adjustments

Bob Evans, Ed Kearns, Kay KilpatrickBob Evans, Ed Kearns, Kay KilpatrickRosenstiel School of Marine and Atmospheric Rosenstiel School of Marine and Atmospheric

ScienceScienceUniversity of MiamiUniversity of Miami

MODIS Science Team MeetingMODIS Science Team MeetingBaltimore, MD. July 2004Baltimore, MD. July 2004

Examples of Instrument effects before and after corrections – Figures 1a and b show the effects of 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 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 polarization correction (10 km wide mirror side banding) and detector gain characterization or an as yet unidentified sourceyet unidentified source and introduces trends across the focal plane (1km stripes and mirror side 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 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 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.trends and detector stripes, order ±0.2% Lt.

Figure 1a. Mirror Side Banding, nLw 412November 1,2000 v3.0.1 Level 1b nocorrections applied beyond MCST

Figure 1b. Detector Corrections tominimize detector to detectorbanding by spectral band

Figure 4B: No corrections-Southeastern U.S. Gulf stream May 9, 2000 nLw412 nocorrections or oceans calibration V3.0.1 level 1b

Figure 5- - With corrections -Southeastern U.S. Gulf stream May 9, 2000 nLw412reprocessed using level 1b v3.0.1 and Oceans corrections and calibrations. RadcorV12_56

Typical ProblemsTypical Problems Single-pixel high stripes in 1km dataSingle-pixel high stripes in 1km data

Interdetector mismatches in bandsInterdetector mismatches in bands 10-pixel high stripes10-pixel high stripes

Mirror-side differencesMirror-side differences Bias and RVSBias and RVS

Edge-to-Edge of Swath discontinuitiesEdge-to-Edge of Swath discontinuities Unresolved RVS variationsUnresolved RVS variations

Temporal inconsistencies in nLwTemporal inconsistencies in nLw LUT’s m1 values not precise enoughLUT’s m1 values not precise enough Low-level changesLow-level changes

Basic MethodologyBasic Methodology Use Terra-MODIS time series at HawaiiUse Terra-MODIS time series at Hawaii

>1200 days/granules>1200 days/granules MOBY time seriesMOBY time series

Break time series into “epochs”Break time series into “epochs” Accounts for instrument changes not well Accounts for instrument changes not well

resolved in m1’sresolved in m1’s

Interdetector balancing: Interdetector balancing: remove “single-pixel” stripingremove “single-pixel” striping Correct on a per band, per epoch basisCorrect on a per band, per epoch basis Use a modal analysis to determine peak response Use a modal analysis to determine peak response

of each detectorof each detector Requires multiple granules per epoch for adequate Requires multiple granules per epoch for adequate

resolutionresolution Adjust modal peak for each detector to match that Adjust modal peak for each detector to match that

for detector 5for detector 5 Relatively easy to achieveRelatively easy to achieve Same technique can be done with x-scan Same technique can be done with x-scan

dependence if need be….new polarization table dependence if need be….new polarization table has reduced the need somewhathas reduced the need somewhat

Detector balancing 412nmDetector balancing 412nm

Detector balancing 512nmDetector balancing 512nm

Cross Scan correctionsCross Scan corrections Requires “flat field” assumption: no significant, Requires “flat field” assumption: no significant,

consistent zonal gradients near Hawaiiconsistent zonal gradients near Hawaii Compute average cross-scan distribution per granule for Compute average cross-scan distribution per granule for

each band, relative to the position at pixel 500 (west of each band, relative to the position at pixel 500 (west of nadir for Terra Day)nadir for Terra Day)

Group the granules’ x-scan behavior within each epochGroup the granules’ x-scan behavior within each epoch Avoid sun glint contamination!!Avoid sun glint contamination!! Fit a 5th order polynomial in a least-squares sense to Fit a 5th order polynomial in a least-squares sense to

normalize the distribution normalize the distribution Quantize this correction into 50 x-scan correction values. Quantize this correction into 50 x-scan correction values.

High AOI shows the most change. xscan carries bulk of High AOI shows the most change. xscan carries bulk of the correction.the correction.

Shape of x-scan correctionShape of x-scan correctionfor each epoch 488 nmfor each epoch 488 nm

Magnitude of cross scan Magnitude of cross scan correction 488nmcorrection 488nm

Shape of x-scan correctionShape of x-scan correctionfor each epoch 412nmfor each epoch 412nm

Low AOI High AOI

Balance Mirror Side 2Balance Mirror Side 2 Add detector bias and cross-scan correction to Add detector bias and cross-scan correction to

2nd mirror side to match first mirror side 2nd mirror side to match first mirror side Does not require a significant flat-field Does not require a significant flat-field

assumption, but assumes that the optical assumption, but assumes that the optical meridional variability in the ocean near Hawaii on meridional variability in the ocean near Hawaii on the scale of 10km is smallthe scale of 10km is small

Compute average mirror side difference for each Compute average mirror side difference for each granule as a function of x-scan positiongranule as a function of x-scan position

Fit a 5th order polynomial to the mirror side Fit a 5th order polynomial to the mirror side differences for all granules within an epochdifferences for all granules within an epoch

Quantize this as a 50 element correction vectorQuantize this as a 50 element correction vector

Net correction applied to Net correction applied to Mirror 2 443nmMirror 2 443nm

eastEast high aoi

West low aoi

nadir

Net correction applied to Net correction applied to Mirror 2 412nmMirror 2 412nm

west low AOI

East high AOI

nadir

Example: Day 2004 113Example: Day 2004 1131km L2 1km L2 magnifiedmagnified

nLw 412nmnLw 412nm Hardest to doHardest to do Note relatively Note relatively

good corrections good corrections right up to sun glintright up to sun glint

Works well over Works well over wide range of Lt’s wide range of Lt’s

Day 22, Year 2004Day 22, Year 2004 nLw 412nmnLw 412nm Note mirror-side Note mirror-side

stripingstriping only in upper rightonly in upper right

Setting the final gainsSetting the final gainsnLw to Lt to nLw to Lt…nLw to Lt to nLw to Lt…

Radiance deficits and anomalies are detected Radiance deficits and anomalies are detected and measured in nLw space (L2)and measured in nLw space (L2)

Corrections are applied to Lt values (L1B)Corrections are applied to Lt values (L1B) Requires a recursive methodRequires a recursive method

Estimate nLw radiance errorEstimate nLw radiance error Scale to LtScale to Lt

• Per band Per band • Systematically underestimate to account for variabilitySystematically underestimate to account for variability

Repeat until solution convergesRepeat until solution converges

Gain correctionGain correction Implement an overall exponential correction for Implement an overall exponential correction for

blue Bands 8 and 9 (based on mean MOBY blue Bands 8 and 9 (based on mean MOBY behavior), behavior), notnot per epoch per epoch

Adjust Bands 10, 11, 12 on a per epoch basisAdjust Bands 10, 11, 12 on a per epoch basis Adjust MODIS (Band X / Band 9) to match MOBY Adjust MODIS (Band X / Band 9) to match MOBY

(Band X / Band 9) ratio(Band X / Band 9) ratio Want ratios (products) to be stableWant ratios (products) to be stable Aim for low RMSAim for low RMS Neglect MOBY v. MODIS individual band biasNeglect MOBY v. MODIS individual band bias

Adjust 748nm band (linear/exponential) to remove Adjust 748nm band (linear/exponential) to remove long-term trends in mean epsilon fieldslong-term trends in mean epsilon fields

Modal time seriesModal time series Use a time series of the mode of each Use a time series of the mode of each

MODIS band’s data in a small concentric MODIS band’s data in a small concentric circles (30, 10, and 3 km) around MOBBY circles (30, 10, and 3 km) around MOBBY with increasing weightswith increasing weights

Compare to filtered MOBY data time Compare to filtered MOBY data time seriesseries

Time series 443Time series 443 Modal time Modal time

seriesseries

% difference % difference from MOBYfrom MOBY

(red points matchup pairs, blue bar (red points matchup pairs, blue bar epoch average)epoch average)

ratio ratio mode/measuredmode/measured

Time series 551nmTime series 551nm Modal time Modal time

series series 551nm551nm

% difference % difference from MOBYfrom MOBY

(red points matchup pairs, blue bar (red points matchup pairs, blue bar epoch average)epoch average)

Time series Time series blue/green blue/green ratioratio

Net gains in timeNet gains in time Gains are computed per band for each Gains are computed per band for each

epochepoch

Multiply the gain by the xscan correction Multiply the gain by the xscan correction for west, center, and east side of scan for for west, center, and east side of scan for each epocheach epoch

443nm net correction443nm net correctiongain * xscan for each epochgain * xscan for each epoch

west low AOI

East high AOI

nadir

Top panel - radcor gain vs time, purple line is band 8 gain adjustmentTop panel - radcor gain vs time, purple line is band 8 gain adjustmentBottom panel - Bottom panel - deviations of the measured m1 coefficients from the deviations of the measured m1 coefficients from the

linear trends linear trends band 8, mirror side 0 (corrected) [from Gerhard Meister]band 8, mirror side 0 (corrected) [from Gerhard Meister] Radcor oscillations ~ factor 5 large than SD M1 trend residualsRadcor oscillations ~ factor 5 large than SD M1 trend residuals

Note: Radcor time resolution ~60 days, M1 two weeksNote: Radcor time resolution ~60 days, M1 two weeks

551nm net correction551nm net correctiongain * xscan for each epochgain * xscan for each epoch

west low AOI

East high AOI

nadir

Gain corrections for MODIS-TerraGain corrections for MODIS-TerraRed - 667, 678nm, Green - 551nm, Yellow - 531nm, Blue - 443nm, Purple Red - 667, 678nm, Green - 551nm, Yellow - 531nm, Blue - 443nm, Purple

- 412nm- 412nm

Days since Jan 1, 2000 to Dec 31, 2003

Calibration adjustmentsdeveloped for currentMODIS ocean reprocessing

Blue bands fit to winterMOBY nLw observationsto minimize sun-glint andpolarization issues.

Green bands adjusted tomatch MODIS to MOBYblue/green band ratios.

TERRA 041 reprocessinggains

Smoothing net corrections Smoothing net corrections exponential least square regressionexponential least square regression

551nm551nm

Example:Nadir xscan

Improvements in terra_v24_49Improvements in terra_v24_49 Smoother gain evolutionsSmoother gain evolutions

No seasonal oscillationsNo seasonal oscillations No need for additional smoothingNo need for additional smoothing

More predictable x-scan and mirror sideMore predictable x-scan and mirror side Overall we have been able to remove much of Overall we have been able to remove much of

the problems in the l1b data that were present in the problems in the l1b data that were present in the transition between 2003 and 2004.the transition between 2003 and 2004.

Bias and RMS terra_v24_49Bias and RMS terra_v24_49BandBand BiasBias RMSRMS

412nm412nm 1%1% 7%7%

443nm443nm 7%7% 5%5%

488nm488nm 9%9% 7%7%

531nm531nm 8%8% 9%9%

551nm551nm 8%8% 12%12%

N = 15N = 15

Remaining ProblemsRemaining Problems Remaining sun glint contamination (x-scan)?Remaining sun glint contamination (x-scan)? Still some odd behavior, variable in time & spaceStill some odd behavior, variable in time & space No other guidance for red bands (fluorescence)No other guidance for red bands (fluorescence)

Requires set balancing of 667/678nm ratioRequires set balancing of 667/678nm ratio Some per-detector xscan behavior remains (though less Some per-detector xscan behavior remains (though less

than before). The time trend in the cross scan correction than before). The time trend in the cross scan correction at high AOI likely indicates a change in the mirror at high AOI likely indicates a change in the mirror polarization. polarization.

Still data and time intensiveStill data and time intensive Terra_v24_50 still has a few issues to be fixedTerra_v24_50 still has a few issues to be fixed

Red gains still maladjusted for 2 or 3 epochsRed gains still maladjusted for 2 or 3 epochs Inter-detector balance off for 2 epochs for Band 12Inter-detector balance off for 2 epochs for Band 12

Smoothing and prediction Smoothing and prediction (experimental)(experimental)

Fit bicubic splines under tension to net Fit bicubic splines under tension to net cross-scan behaviorcross-scan behavior

Caveats: avoid summer times (residual Caveats: avoid summer times (residual glint)? Edge of scan?glint)? Edge of scan?

Produces look-up tables and functional Produces look-up tables and functional relationships for correctionsrelationships for corrections

Limited predictive capabilities (good Limited predictive capabilities (good enough for short term?)enough for short term?)

Day of year 2000xscan pixel

xscan pixel Day of year 2000