Notes on the Southern Ocean chlorophyll algorithm

 

Mati Kahru, mkahru@ucsd.edu

 

·        Bio-optical data from the following cruises were used: NBP9711, REV9801 and REV9802, AMLR2000, AMLR2001, AMLR2004, LMG0402, AMLR2006. NBP0606 was not used.

·        The Chl-a was the “Chl-to-use” that Haili specified in table SPG_SO_algorithm in Algorithm_OCBAM.mdb (email on 9/16/2006).

·        Rrs was my type Rrs3 that is determined as the mean ratio of Lu/Ed in the top valid layer (specified by visual inspection of profiles), e.g. between 1-6 m depth.

·        The software interpolates Rrs to any wavelength and can switch between the bands of SeaWiFS, MODIS, GLI. The maximum band ratio (MBR) can use any of the bands. The standard plots use SeaWiFS standard bands  and pick the maximum between Lwn at 443, 490, and 510 nm bands as a ratio with Lwn555.

·        For each cruise or a set of cruises (NBP9711 and REV9801 and REV9802 were pooled) plots were made that show the standard OC4v4 versus SPGANTv3. (The SPG version 3 fit produced as a byproduct of the GLI project).

·        Separate versions of SPGANTv3 exist for SeaWiFS, MODIS and GLI.

·        The following plots compare the standard OC4v4 versus the SPGANTv3 for each of the cruises. For all of the cruises except AMLR2006 the SPGANTv3 produced better retrievals, i.e. smaller RMS1 error and larger r². The AMLR2006 had exceptionally high Chl-a in the AMLR region, however, the dataset used in the plots does not a dramatically different range. Still, AMLR2006 seems to be bio-optically different from the other cruises.

·        The first plot shows all the data together, the following plots show each cruise (cruise group) separately.

·        It is evident that, in general, retrievals of moderate level Chl-a concentrations (>0.1 and < 3 mg m^-3) by the NASA global OC4v4 algorithm under-predict Chl-a by approximately a factor of 2.

·        Compare the RMS1 values for both OC4v4 and SPGANTv3: for all the cruises except AML2006 the SPGANTv3 has significantly smaller RMS errors. For AMLR2006 the OC4v4 has slightly smaller RMS error.

·        Currently I don’t see a need to modify the SPGANTv3 as slight changes of the coefficients do not improve the algorithm performance.

 

 

NBP9711, REV9801, REV9802

 

 

 

AMLR2000

 

 

AMLR2001

 

 

AMLR2002

 

 

AMLR2004

 

 

LMG0402

 

 

AMLR2006