Over the six months since the last reprocessing in February, the SeaWi FS Project has continued to improve the processing code. Also, many of you have notified us of problems you have observed in the derived products and we have tried to address these in preparation for a reprocessing. These improvements include modifications in the navigation, calibration, atmospheric correction, chlorophyll-a, and binning algorithms. As a result of these changes, the Project considers the resultant changes in the products significant enough to warrant another reprocessing. Over the past three months, in particular, much effort has been expended in evaluation and testing of these modifications and we now are satisfied that we have addressed all the issues we can identify to the best of our ability. I provide some details on the most significant enhancements below.
The reprocessing will begin immediately, although the distribution fro m the DAAC will not begin for a couple of weeks. This allows the DAAC time to make final preparations for the new products and also allows the Project one last opportunity to screen the products before making the complete transition to the new processing code. Our estimate of the number of scenes to be reprocessed are 4938 GAC files, 4853 LAC files, and 8841 HRPT files. We have completed a number of system upgrades and procedural enhancements over the past few months that will greatly expedite the reprocessing. Finally, a SeaDAS update that incorporates the algorithm changes will be released by the end of August.
A summary of the evaluation analyses associated with the reprocessing will be published in the September issue of Sea Technology. The Project Office will provide reprints to those who request them. Also, once the article is published, a more complete set of analysis products will be posted on the SeaWiFS homepage and we will notify you when those can be viewed.
1. Navigation
The major change in the SeaWiFS navigation processing is the
implementation of a new algorithm for determining the spacecraft
attitude using a Kalman filter combined with a simple dynamics model.
This method improves the yaw knowledge at the subsolar point and
reduces the scan-to-scan noise in the instrument pointing. Additional
changes include: refined attitude sensor alignments to improve overall
accuracy; a revised horizon scanner triggering height model; and better
detection/rejection of invalid telemetry data.
2. Prelaunch Calibration
During the spring of 1997, the SeaWiFS Project, NIST, and OSC conducted
a recalibration of the SeaWiFS sensor to determine if the sensor
responsivity had changed over the three years since it was originally
calibrated at Hughes/Santa Barbara. The prelaunch calibration table
that has been in use was based on a preliminary analysis of the 1997
calibration. That analysis has been updated and finalized and a new
prelaunch calibration table will be used in the reprocessing. The
magnitude of the changes is greatest in bands 6, 7, and 8 and is
roughly 3%. The SeaWiFS TM describing the recalibration has been
completed and is in the editing phase of the publication procedure.
3. Calibration of Bands 7 and 8
As I mentioned in the June SeaWiFS Project report, the sensitivities of
band 7 and 8, as observed in the solar and lunar calibration data, have
been slowly degrading since the mission went operational last September.
The degradation is greatest in band 8 and is presently 4-5%. Bands 7 and 8
are used for the aerosol radiance correction in the visible bands. Band 6
also shows indications of degradation, but the trend is very slight. The
most likely cause of the degradation is changes in the optical filters.
Corrections for these degradations were not incorporated in the February
reprocessing because the effects could not be clearly quantified at that
time and other improvements in the processing necessitated a reprocessing.
It has also become clear that the calibration drift has caused trends in
the derived products such as a steady increase in the global average
epsilon(765,865) values and a decrease in the global open ocean chlorophyll
values. We have developed a correction scheme using the solar and lunar
data which does remove the trends in the geophysical products.
4. Noise in the dark count adjustment
Examination of the dark count data showed that noise could produce
erroneous dark count adjustments in the calibration. The dark count offset
is applied on a line-by-line basis and noise can produce slight striations
in the data. A filter has been added to preprocess the dark count data.
5. Solar irradiance calculation
An additional higher order term in the Earth-Sun distance calculation has
been added. The magnitude of the correction is roughly 0.1%. Also, there
was a bug in the time of day calculation which introduced an error of about
1% in the solar irradiance estimate.
6. Diffuse transmittance in the Lwn calculation
Examination of the LAC data over the MOBY site indicated a scan angle
variation in the Lwn retrievals. This effect was removed by incorporating
a total diffuse transmittance correction to the Lwn calculation.
7. Whitecap correction
It was frequently observed that in very clear conditions, the Lt-Lr would
be negative. This was traced to the whitecap correction. The whitecap
radiance has been reduced to one-fourth of the original estimate.
8. Aerosol models
The aerosol correction is based on the selection of an aerosol model based
on an estimate of epsilon(765,865). It was found that 35-40% of the pixels
processed defaulted to a particular model because the computed epsilon was
lower than the minimum value the models would accommodate. Additional
aerosol models have been added to the algorithm to avoid this situation.
At present, roughly 15-20% of the pixels have epsilon values below the
range of the models.
9. Coccolithophore flag
The present version of the coccolithophore flag parameter settings was
based on a single bloom from the Bering Sea and has not been working
particularly well. The parameter settings have been revised based on tests
using several scenes from the Bering Sea and North Atlantic.
10. The chlorophyll-a algorithm
The OC-2 algorithm was based on a large, but limited bio-optical data set.
The primary limitation of that data set was the number of stations with
chlorophyll values above 5 mg/m3. It is now apparent that the algorithm
overestimates concentrations above concentrations of around 2 mg/m3. About
300 additional stations have been added to the data set and a revised
algorithm developed. The same formulation as OC-2 was used, but the
coefficients have been adjusted. The result is that values previously
below 0.03 mg/m3 will be increased by as much as 50% and values that were
pegged at the maximum value of 64 mg/m3 will be reduced by a factor of 3-4.
Values between 0.03 mg/m3 and 2 mg/m3 are essentially the same.
11. Space binning algorithm
Because bands 1 and 2 often go slightly negative in turbid or high
chlorophyll water, many pixels which have reasonable 490, 555, and
chlorophyll retrievals do not get included in the binned data. The
negative Lw flag is set if any of bands 1-5 are negative. This results in
a noticeable loss of coverage in the level-3 products. The reason is
because the level-3 products must all have the same number of valid values.
In order to circumvent this situation, the negative Lw flag will not be used as an exclusion criteria and negative values will be averaged as zero values.
There are additional issues which the Project will continue to pursue, but have no real solutions for at this time. We would greatly appreciate the assistance of the user community in addressing these problems and any others you can identify. These topics include the following:
1. Cloud masking algorithm
The present algorithm works reasonably well in general, but cloud edges can
be problematic because the 865 albedo can be similar to that of correctable
haze. To drop the albedo too low results in the masking of haze as well.
On the other hand, we have noticed that in some cases the albedo can be
raised resulting in a significant increase in coverage without compromising
the cloud masking. Those of you who are processing individual scenes
should vary the cloud albedo threshold to obtain the most optimum result.
If you elevate the cloud threshold, you may also need to switch off the
"high total radiance" mask which terminates processing of data above the
"knee" in the bilinear gain.
2. Atmospheric correction over turbid water
Several of you have noticed that in highly turbid water, the atmospheric
correction yields noticeably nonuniform aerosol radiances, epsilons, etc.
which impact the pigment products. This is not surprising as the NIR
reflectance in these waters may deviate from zero. I can envision
iterative schemes which either correct for the NIR reflectance somehow or
extrapolate the atmospheric correction parameters from nearby non-turbid
waters, but the Project does not have a methodology in place for addressing
this situation.
3. Eolian dust
Airborne dust such as from the Gobi and Saharan deserts is often observed
in SeaWiFS data and can cover large expanses of the ocean. Dust does
selectively absorb in the visible resulting in abnormally low Lw's. The
present set of atmospheric models does not include any models for dust
primarily because the phase function and indices of refraction are not well
known. High concentrations of dust triggers the cloud mask, but low
concentrations are processed leading to erroneous derived products, e.g.,
low Lw and high pigment estimates. We do not have a method for flagging
dust contaminated pixels, let alone correcting for it.
4. "Speckling" in the derived products
Scattered throughout the data are random values that are inconsistent with
the surrounding pixels. This is quite obvious in the chlorophyll fields.
This noise is not related to bit flips evident in some HRPT data (Gene
Feldman will release a separate message on that problem later). The noise
causes both high and low values which are not always associated with other
known problems such as cloud edge effects. We will continue to analyze
data to try and determine the source, but we could use some assistance.
Chuck McClain
SeaWiFS Project Manager
