1. Ocean Optics XXI, Glasgow, Scotland, October 8-12, 2012 David Antoine; André Morel Laboratoire d’Océanographie de Villefranche (LOV), CNRS and Université Pierre et Marie Curie, Paris 06, UMR 7093, Villefranche-sur-Mer, France Stéphane Maritorena; David A Siegel; Norm B Nelson Earth Research Institute (ERI), University of California at Santa Barbara (UCSB), Santa Barbara, California, United States of America. Department of Geography, University of California at Santa Barbara (UCSB), Santa Barbara, California, United States of America Hubert Loisel; David Dessailly Laboratoire d’Océanologie et de Géosciences (LOG), CNRS and Université du littoral, côte d’opale, Wimereux, France Evaluation of Particulate Backscattering Inversion Algorithms in Clear Oceanic Case 1 Waters

2. Ocean Optics XXI, Glasgow, Scotland, October 8-12, 2012 Rationale (1/3) A number of “b bp algorithms”, i.e., inversion of radiometric quantities in terms of a and b b, exist (see, e.g., IOCCG report N°5, 2006) Validation of the b bp retrieval from these algorithms is still quite limited, because of 1 – A lack of b b measurements (although the situation improves now) 2 – b b measurements are still dominated by either coastal waters or open ocean waters with relatively high [Chl], i.e., b bp (550) > ~0.002 m -1 Therefore, we don’t really know how these algorithms perform for waters that represent more than half of the global ocean, i.e., waters with b bp (550) < ~0.001-0.002 m -1 See, e.g., recent paper by Dall’Olmo et al., 2012, Optics Express 20(19)

3. Ocean Optics XXI, Glasgow, Scotland, October 8-12, 2012 Figs. 5A in Kostadinov et al., 2009, J. Geophys. Res., 114, C09015, doi:10.1029/2009JC005303 Average global repartition of b bp at 550 nm

4. Ocean Optics XXI, Glasgow, Scotland, October 8-12, 2012 Rationale (2/3) This is important in particular in view of reducing uncertainty on the behaviour of particulate backscattering for low-Chl waters Fig. 1 in Behrenfeld et al., 2005, Global Biogeochem. Cycles, 19, GB1006, doi10.1029 / 2004GB002299 Fig. 5a in Antoine et al., 2011, L&O, 56(3), 955–973 Fig. 1A in Huot et al., 2008, Biogeosciences, 5, 495–507

5. Ocean Optics XXI, Glasgow, Scotland, October 8-12, 2012 Rationale (3/3) A number of applications exist that use “satellite b bp ”  How uncertainties in b bp retrievals for clear waters may affect these results? Fig. 3 in Loisel et al., 2006, J. Geophys. Res., 111, C09024, doi:10.1029 / 2005JC003367 Fig. 4C in Behrenfeld et al., 2005, Global Biogeochem. Cycles, 19, GB1006, doi10.1029 / 2004GB002299 Figs. 5A and 7A in Kostadinov et al., 2009, J. Geophys. Res., 114, C09015, doi:10.1029/2009JC005303

6. Ocean Optics XXI, Glasgow, Scotland, October 8-12, 2012 Objectives Use inversion algorithms of quite different nature, and apply them to in situ data sets in order to evaluate uncertainty in the final b bp product for clear waters. The goal is not to perform an inter-comparison of algorithms with the idea of ranking algorithms Application of the algorithms is performed in different configurations in order to evaluate robustness to: 1 – Loss of some information when using a single input quantity (R rs ) instead of both R and K d 2 – Behaviour when fed with satellite R rs, which includes additional errors from atmospheric correction

7. Ocean Optics XXI, Glasgow, Scotland, October 8-12, 2012 Data sets (1/2) BOUSSOLE: Med. Sea clear waters, Chl from ~0.05-5 mg m -3, b bp (555): 0.0005-0.005 m -1 PnB Stations 2-6: More coastal, still Case 1, Chl from ~0.5-10 mg m -3, b bp (555): 0.0007-0.01 m -1 BIOSOPE: SE Pacific gyre, the most oligotrophic waters in the World ocean Chl from ~0.02-5 mg m -3, b bp (555): 0.0002-0.005 m -1

8. Ocean Optics XXI, Glasgow, Scotland, October 8-12, 2012 Data sets (2/2) Data from Hobilabs Hydroscat-4 (BOUSSOLE), Hydroscat-6 (PNB) sensors (Maffione and Dana,1997. Appl. Opt. 36: 6057–6067), Wetlabs EcoBB3 (BIOSOPE) Measurements & data analysis protocols: Antoine et al. (2011) L&O, 56(3), 955–973 for BOUSSOLE Kostadinov et al., 2007, J. Geophys. Res. 112 for PNB Twardowski et al. 2007, Biogeosciences 4, 1041-1058 for BIOSOPE

9. Ocean Optics XXI, Glasgow, Scotland, October 8-12, 2012 Algorithms The GSM Model (Garver & Siegel, 1997; Maritorena et al., 2002) Non-water components of absorption and scattering are expressed as known shape functions with unknown magnitudes New version of the Loisel and Stramski method (Loisel and Stramski, 2000; Loisel et al., 2001). No assumptions about spectral shapes for absorption and scattering (“No Spectral Assumption Algorithm” or “NSAA”) LOV: (Morel et al., 2006, Deep-Sea Res. I, 53, 1439-1559) Just based on 2 equations: K d ( ) = 1.0395 [a( ) + b b ( )] /  d (Gordon, 1989, L&O 34) R( ) = f' b b ( ) / [a( ) + b b ( )] LUTs are used for  d and f’ (Morel & Gentili, 2004, J. Geophys. Res.) For the 3 algorithms, b bw ( ) is computed following Zhang et al. (2009, Opt. Exp. 17: 5698-5710) and Zhang and Hu (2009, Opt. Exp. 17: 1671-1678), as a function of temperature and salinity

10. Ocean Optics XXI, Glasgow, Scotland, October 8-12, 2012 The test steps 1 st step (ideal case): For algorithms using R and K d as inputs: both R and K d are independently derived from field radiometry measurements For algorithms using R rs as input: R rs derived from field radiometry measurements 2 nd step (algorithms using R and K d as inputs): R is still derived from field radiometry measurements, but K d is now modelled (either from R rs or Chl) 3 rd step : satellite R rs are used for all algorithms (for algorithms using R and K d as inputs: both R and K d are derived from satellite R rs )

11. Ocean Optics XXI, Glasgow, Scotland, October 8-12, 2012 Results using R-K d when feasible, and R rs otherwise PNB BOUSSOLE BIOSOPE 1 10 -4 2 10 -2

12. Ocean Optics XXI, Glasgow, Scotland, October 8-12, 2012 PNB BOUSSOLE BIOSOPE Results using R-K d but with K d =f(Chl) and R rs otherwise

13. Ocean Optics XXI, Glasgow, Scotland, October 8-12, 2012 Results using SeaWiFS R rs PNB BOUSSOLE

14. Ocean Optics XXI, Glasgow, Scotland, October 8-12, 2012 Results using SeaWiFS R rs All bands pooled together PNB BOUSSOLE

15. Ocean Optics XXI, Glasgow, Scotland, October 8-12, 2012 Respective importance of seawater / particles In determining total b b (solid lines) and a (dashed lines) 443 nm 550 nm Using Morel & Maritorena (2001)

16. Ocean Optics XXI, Glasgow, Scotland, October 8-12, 2012 Getting accurate b bp in the field for clear waters Effect of accounting for field determinations of dark currents on the b bp spectral slope (  ) (blue: with dark records included) BOUSSOLE data (Hydroscat-IV measurements)

17. Ocean Optics XXI, Glasgow, Scotland, October 8-12, 2012 to conclude  Still very difficult to get accurate b bp from radiometry in clear waters  Still difficult as well to get accurate b bp field measurements in clear waters. The b bp values < ~0.001 m -1 determined in situ using currently available instrumentation have to be carefully considered  Overall the best results are obtained in the green, around =550 nm  Seems illusory (at least difficult) to get b bp in the blue ( ~440nm) by using only blue bands. Need methods that constrain to some extent the b bp derivation using more bands (e.g., GSM)  Degradation of results when using a modelled K d instead of the measured one is not so dramatic  Which K d is to be used? Field determination or modelled value from R rs or Chl ? Using a model could actually decrease the noise of the inversion. The best thing to do might be to improve our determination of K d from field radiometry.

18. Ocean Optics XXI, Glasgow, Scotland, October 8-12, 2012 Thank you for your attention Questions?

19. Ocean Optics XXI, Glasgow, Scotland, October 8-12, 2012 Algorithms (1/3) The GSM Model (Garver & Siegel, 1997; Maritorena et al., 2002) Gordon et al. (1988) Non-water components of absorption and scattering are expressed as known shape functions with unknown magnitudes (=unknowns) : a ph (λ) = Chl a ph *(λ) a cdm (λ) = a cdm (443) exp(-S(λ -443)) b bp (λ) = b bp (443) (λ /443) -η a ph *(λ), S and η were optimized for global applications using a large in situ data set. Unknowns and their confidence intervals are retrieved by fitting the model to the observed R rs using a non-linear least-square technique.

20. Ocean Optics XXI, Glasgow, Scotland, October 8-12, 2012 Algorithms (2/3) New version of the Loisel and Stramski method (Loisel and Stramski, 2000; Loisel et al., 2001). No assumptions about spectral shapes for absorption and scattering (“No Spectral Assumption Algorithm” or “NSAA”)  R rs is used as input parameter instead of R(0 - )  New formulation between : b b and (R rs, K d,  w,  )  K d ( ) is retrieved from NN (Jamet et al., 2011)  More realistic  and b/a combinations How ? Radiative transfer simulations with no spectral assumptions (a=1, b/a [0.02 to 30],  [0.01-0.20 %] Performance using the synthetic (error free) IOCCG data set with the true K d

21. Ocean Optics XXI, Glasgow, Scotland, October 8-12, 2012 Algorithms (3/3) LOV: (Morel et al., 2006, Deep-Sea Res. I, 53, 1439-1559) Just based on 2 equations: K d ( ) = 1.0395 [a( ) + b b ( )] /  d (Gordon, 1989, Limnol. Oceanogr. 34: 1389-1409) R( ) = f' b b ( ) / [a( ) + b b ( )] from which a( ) = 0.962 K d ( )  d (,  s, Chl) x [1 – R( ) / f'(,  s, Chl)] b b ( ) = 0.962 K d ( )  d (,  s, Chl) x [R( ) / f'(,  s, Chl)] LUTs are used for  d and f’ (Morel & Gentili, 2004, J. Geophys. Res., 109, C6) For the 3 algorithms, b bw ( ) is computed following Zhang et al. (2009, Opt. Exp. 17: 5698-5710) and Zhang and Hu (2009, Opt. Exp. 17: 1671-1678), as a function of temperature and salinity