SeaWiFS Ocean Color Reprocessing 2014.0

Introduction

The Ocean Biology Processing Group (OBPG) completed a full-mission reprocessing of the SeaWiFS ocean color dataset in December 2015. This reprocessing is part of a multi-mission effort to update common algorithms, product suites, and data formats across all supported missions. Sensor-independent changes are detailed in the R2014.0 Ocean Color Reprocessing General Description. Here we describe the SeaWiFS sensor-specific details of the reprocessing, and provide an assessment of data quality and impact relative to the previous SeaWiFS reprocessing version R2010.0 .

Sensor-specific Processing Details

Source Data

As in previous reprocessings, R2010.0 starts from the SeaWiFS Level-1A files. For SeaWiFS Global Area Coverage (GAC), the Level-1A source files are unchanged for this reprocessing. For the Local Area Coverage (LAC), some additional data has been acquired from various direct broadcast stations, thus expanding to a small extent the available spatial coverage of the merged LAC (MLAC) dataset.

Instrument Calibration

This reprocessing incorporates significant advancements in SeaWIFS calibration knowledge and implementation; specifically:

1. Improved computation and application of the dark offsets.

2. Updated Band 7 Gain 3 drift correction.

3. Incorporation of a time-dependent relative spectral response (RSR) for the NIR bands.

Details of each of these are described here.

Dark Offsets

The R2010.0 temporal anomalies for SeaWiFS showed a clear artifact in the second half of 2005 in the mid-wavelength Rrs (e.g., 555 nm). The Rrs appeared to be temporally stable before and after 2005. This artifact also appeared in dramatic fashion in the angstrom exponent.

The problem was traced to the calculation of the dark offset value. The dark offset is included in the SeaWiFS data as one 10-bit count per band and scan line. Shortly after launch, the application of the dark offset was implemented as the median of the dark counts for each band and granule, to address noise in the counts. This implementation was unchanged through R2010.0.

Examination of the dark count over the mission showed that the values had changed by less than 1 count in all of the bands. The use of the median filter caused the value applied to the data to jump by one count in some bands. This was particularly apparent in Band 8 (865 nm), where the mean dark count (dots) drifted above 20.5 counts in late 2005, causing the median (+) to jump from 20 to 21 counts.

This jump caused a corresponding decrease in the 865 nm radiance, resulting in the step increase in angstrom and the artifact in the Rrs. The visibility of the actual dark offset value was limited by the 10-bit resolution of the dark count, as shown by the noise in the mean counts. This was addressed by analyzing the intergain calibration (IGC) data, which provided samples of the dark count for each detector, reducing the effects of digitization, and also allowed the values to be computed for each gain. The resultant values were then averaged for each month of the mission, and implemented in the processing software as a look-up table. The final step was to re-analyze the lunar calibration data, which was also affected by the median filtering of the dark count. These improvements eliminated the step-change in angstrom and the artifact in the 555 nm Rrs.

Band 7 Gain 3 Drift Correction

The SeaWiFS lunar calibrations are performed using a different set of commanded gains for each band than the normal Earth-viewing data collection, in order to optimize the sensor signal for the calibration. As described for Reprocessing R2007.0, Band 7 (765 nm) has shown a mission-long drift between the Earth view gain (1) and lunar gain (3). This drift was incorporated into the lunar time series for Band 7.

In conjunction with the dark count analysis, the Band 7 gain drift was re-analyzed using the individual detector IGC data to reduce the effects of digitization. The result was a significant change in the computed gain drift. An updated fit to the gain drift values (black lines on the plot) has been incorporated into the Band 7 lunar time series for R2014.0.

Time-Dependent NIR Band RSR

A significant change in instrument response over time can change the effective bandpasses of remote sensing instruments, especially for those that have significant out-of-band signals in their spectral response functions. The SeaWiFS NIR bands have both shown significant reduction in radiometric response, and both have significant out-of-band response in the visible wavelengths. The end-of-mission (EOM) degradation in radiometric response vs. wavelength is shown here.

A fit to the degradation vs. wavelength was convolved with the prelaunch RSR for each band to estimate the EOM RSR. The results for Band 8 showed an increase in the visible out-of-band response; the change for Band 7 was smaller, but still apparent.

The prelaunch and EOM RSRs were then convolved with both the blue ocean spectrum and the lunar spectrum to determine the difference in the effects of the RSR change for the Earth and lunar views. Since the temporal response is based on the lunar calibration, these differences will reside in the calibrated Earth view data. The differences were insignificant in the visible wavelengths, but significant in Bands 7 and 8, which are shown below.

Spectrum / Band	7	8
Blue Ocean Change	0.218%	0.840%
Lunar Change	0.122%	0.218%

The differences between the blue ocean and lunar results were used to adjust the temporal response functions for Bands 7 and 8, which were incorporated into the calibration for R2014.0.

Vicarious_Calibration

The OBPG applies an additional vicarious calibration to SeaWiFS during Level-2 processing (Franz et al. 2007). The 869nm band is assumed to be correctly calibrated from prelaunch measurements. The 748nm band is adjusted using match-ups from the South Pacific Gyre to force the aerosol type retrievals to agree, on average, with the aerosol type observed at the Tahiti AERONET site. The calibration of the visible bands (412nm to 678nm) is then adjusted to produce retrievals that match, on average, measurements from the Marine Optical Buoy (MOBy) near Lanai Hawaii. The vicarious calibration gains derived in this manner are:

Wavelength (nm)	412	443	490	510	555	670	765	865
Gain	0.9978	0.9911	0.9819	0.9860	0.9952	0.9701	0.9580	1.0
Std. Dev.	0.01059	0.01095	0.01068	0.01045	0.00947	0.00758	0.00047	0.00000

Level-2 Processing

The algorithms employed and products produced from SeaWiFS are as described in the R2014.0 Ocean Color Reprocessing General Description. Sensor-specific center wavelength values used in processing and product naming are as shown in the table above.

Impact and Quality Assessment

Impact of Reprocessing on Timeseries

To assess the impact of this R2014.0 reprocessing relative to the previous R2010.0 reprocessing, a comparative timeseries analysis was performed (see Franz (2009) for details on the evaluation approach). The impact of all calibration, algorithm, and ancillary data changes on remote sensing reflectance is shown in this comparison of monthly mean ocean color products relative to reprocessing 2010.0. As an example, the plot below shows the global deep water (> 1000m) Rrs averages as a function of time (dashed line is R2014.0, solid line is R2010.0), with the ratio shown in the right column. As can be seen, the changes to the Rrs time-series are subtle in comparison to R2010.0, but the ratio shows changes as large as 15% in the deep-water averages, with distinct shifts and trend changes. The shifts in the ratio time-series are showing places where the dark offsets were previously changing by one bit, primarily in the near infrared bands. The trend changes are due to a combination of the removal of these non-physical shifts, re-analysis of the lunar calibration data, adjustment of the band 7 commanded gain drift, and the temporal changes in the NIR spectral response functions. The effect on the retrieved aerosol angstrom exponent is more dramatic, as the mid-mission shift in mean aerosol type has now been removed. Also shown is a comparison of the deep-water chlorophyll trends for the two reprocessings, where the black line shows the chlorophyll for the new default OCI algorithm, and the blue line show the change when the old OC4 algorithm is compared. The impact of this reprocessing is an increased positive trend in chlorophyll of order 5% over the first half of the mission (1999-2005).

R2010.0 & R2014.0	R2014.0/R2010.0

It is not possible to verify that this trend change brings the SeaWiFS time-series closer to truth, as comparison with in situ time-series or with other sensors (e.g., MODIS-Aqua) are inconclusive at the 5% level. Comparison of the deep-water anomaly time-series (the same data as shown above, but with the mean seasonal cycle subtracted) does provide some further insight. The plots below show these anomaly time-series for chlorophyll and water-leaving reflectance (Rrs) for each spectral band. By reducing the chlorophyll in the first half of the mission, the R2014.0 chlorophyll time-series now shows a very clear and coherent trend over the mission lifespan. Similar observation can be made about the reflectance trends. Trends in the green and red are now much more stable over the mission for this clear-water dataset, with much of the the unnatural variability associated with dark-offset shifts largely removed (e.g., bump mid-2005).

R2010.0	R2014.0

Validation with In Situ Measurements

Validation of the remote sensing reflectance retrievals was performed relative to all available Rrs match-ups from SeaBASS holdings and additional match-up analysis with level 2 data from the Aerosol Robotic Network - Ocean Color (AERONET-OC). Statistical analysis of the satellite to in situ match-ups as of this writing is provided below. Also shown are scatter plots and frequency distribution comparisons. These statistics are effectively unchanged from previous reprocessing 2010.0.

Validation of Rrs relative to SeaBASSin situ matchups

Product Name	SeaWIFS Range	In situ Range	#	Best Fit Slope	Best Fit Intercept	R2	Median Ratio	Abs % Difference	RMSE
Rrs412	-0.00223, 0.01754	0.00021, 0.02150	715	1.06926	-0.00076	0.79518	0.90373	24.66709	0.00192
Rrs443	-0.00085, 0.01978	0.00034, 0.02227	958	1.03754	-0.00032	0.79568	0.96207	18.64380	0.00142
Rrs490	0.00032, 0.02676	0.00067, 0.03020	976	0.94667	-0.00004	0.81467	0.93405	14.98170	0.00120
Rrs510	0.00064, 0.02685	0.00065, 0.03023	785	0.95200	-0.00004	0.85479	0.93326	15.24788	0.00102
Rrs555	0.00088, 0.02563	0.00029, 0.03052	712	0.91538	0.00006	0.87378	0.93482	17.34178	0.00114
Rrs670	-0.00035, 0.01195	0.00002, 0.01090	455	0.91949	-0.00003	0.84475	0.92326	36.09421	0.00054

Validation of Rrs relative to AERONET-OC in situ match-ups

Product Name	SeaWIFS Range	In situ Range	#	Best Fit Slope	Best Fit Intercept	R2	Median Ratio	Abs % Difference	RMSE
Rrs412	-0.00264, 0.01614	0.00001, 0.01728	966	1.22422	-0.00091	0.73501	0.92645	29.17420	0.00151
Rrs443	-0.00062, 0.01924	0.00007, 0.02090	972	1.13060	-0.00027	0.81951	1.04554	22.45560	0.00126
Rrs490	0.00017, 0.02152	0.00039, 0.02755	558	0.88051	0.00007	0.87926	0.89149	16.84298	0.00127
Rrs555	0.00114, 0.02392	0.00115, 0.02313	972	0.92678	-0.00005	0.84843	0.91380	12.83385	0.00137
Rrs670	-0.00041, 0.00664	0.00002, 0.00588	833	1.05333	-0.00018	0.73990	0.83510	26.84098	0.00047

Validation Results: SeaBASS chlorophyll retrievals

Product Name	SeaWIFS Range	In situ Range	#	Best Fit Slope*	Best Fit Intercept*	R2*	Median Ratio	Abs % Difference	RMSE*
chlor_a	0.02818, 49.29538	0.02400, 138.04700	1739	0.91926	0.02513	0.85076	1.06927	36.27090	0.29030

References

Franz, B. A., Bailey, S. W., Werdell, P. J., & McClain, C. R. (2007). Sensor-independent approach to the vicarious calibration of satellite ocean color radiometry. Appl. Opt., 46(22), 5068. https://dx.doi.org/10.1364/ao.46.005068

Franz, B. A. (2009). Methods for assessing the quality and consistency of ocean color products. NASA Goddard Space Flight Center, Ocean Biology Processing Group. https://oceancolor.gsfc.nasa.gov/docs/methods/sensor_analysis_methods