---

Analysis Definition

Questions 9 and 10 of the IOP Workshop are:

(9) How do the algorithms perform globally, processed from satellite Level-2 Rrs(λ)?
(10) How do the algorithms perform globally, processed from satellite Level-3 Rrs(λ)?

The NASA OBPG traditionally produces derived products such as IOPs by processing Level-2 (per satellite observation) remote sensing reflectance retrievals (Rrs) through an algorithm (e.g., IOP inversion model) to produce Level-2 derived products. These Level-2 products are than spatially and temporally binned and averaged to produce global Level-3 composites. An alternative approach to producing global composites of derived products is to start from Level-3 composites of normalized water-leaving radiances or Rrs and generate global derived products using these mean Rrs values as input to the product algorithm. i.e., the two methods of producing Level-3 global composites are:

1) [L1 observed radiances]->l2gen->[L2 IOPs]->l2bin->l3bin->[L3 IOPs]
2) [L1 observed radiances]->l2gen->[L2 Rrss]->l2bin->l3bin->[L3 RRss]->l3gen->[L3 IOPs]

The 2nd approach requires less computational resources and is therefore often used in research and application development activities external to NASA/OBPG, when non-standard product algorithms are to be evaluated or applied on global scales. Here we simply wish to assess any differences between these two binning strategies. To that end, the standard NASA Level-1 to Level-2 code (l2gen, a.k.a. msl12) was modified to read Level-3 binned products of nLw or Rrs and associated mean solar and path geometry and output derived products in the same Level-3 binned format. This new code, l3gen, shares all algorithm implementation and capabilities with l2gen, so any potential differences in algorithm application are minimized.

For each model/method, global SeaWiFS data was processed to 9.2-km monthly means for March, June, September, and December of 2005, and common bins were evaluated over several water classification subsets. i.e.:

• global: all bins
  
• deep: bins where water depth exceeds 1000 meters
  
• olig: oligotrophic bins (where chl < 0.1 mg m^3 on average, based on SeaWiFS mission mean)
  
• meso: mesotrophic bins (where 1 < chl < 0.1 mg/m^3 on average, based on SeaWiFS mission mean)
  
• eutr: eutrophic bins (where 10 < chl < 1 mg/m^3 on average, based on SeaWiFS mission mean)
  

Related Discussion

• Exact Rrs
  

Analysis Results

• GSM model, global histograms of monthly mean IOPs and histogram differences
  
• QAA model, global histograms of monthly mean IOPs and histogram differences
  
• PML model, global histograms of monthly mean IOPs and histogram differences
  

Observations and conclusions:

GSM (6-Band fit using Levenburg-Marquart)
Results are nearly identical for both compositing approaches. The exception is bb in eutrophic waters, where a fraction of retrievals in the 0.006-0.01 range (at 443-nm) appear to be shifted to higher values with binning strategy 2. The result is a more peaked distribution with similar mode but thicker high-bb tail. This is not yet understood. Note also that both binning strategies mask bb retrievals above 0.015 in any band, which is the reason for the cut-off on high-side tail.

QAA
In contrast, QAA shows largest differences in eutrophic absorption, with binning strategy 2 producing a higher peak with similar mode.

PML
The PML model shows small differences in all water class subsets, for both a and bb, with larger differences in eutrophic waters. This is likely a result of using mean radiant path geometry to derive the bidirectional corrections (Fresnel relfection/refraction, f/Q) to the mean Rrs, rather than correcting Rrs and Level-2 using the true radiant path geometry.

Summary
More analysis needs to be done to assess whether these differences are significant, but they are certainly less significant than the differences between models, as can be seen in the table below. This would suggest the binning strategy is not a primary concern at this time, and that we can evaluate the performance of IOP models and inversion methods on global Level-3 composites with relative confidence using either binning approach.

March 2005	GSM	QAA	PML
a_443
bb_443