Performance evaluation of non-water absorption coefficient partitioning algorithms in optically complex coastal waters of Kochi and Goa, India

Absorption coefficient partitioning algorithms (APAs) were developed to partition the total absorption coefficient (a(λ)) or total non-water absorption coefficient (anw(λ)) into the absorption subcomponents, i.e., absorption due to phytoplankton aph(λ), colored dissolved organic matter (CDOM) aƍ(λ) and non-algal particulate matter ad(λ), λ is the wavelength. Absorption coefficients of CDOM and non-algal particulate matter are generally combined due to a similarity in exhibited spectral shape and represented as colored detrital matter (CDM) absorption coefficient, adƍ(λ). This study focuses on the applicability of five APAs Schofield's, Lin's, Zhang's, Stacked Constraints Model (SCM) and Generalized Stacked Constraints Model (GSCM), in deriving the absorption subcomponents from anw(λ) in optically complex coastal waters of Kochi and Goa, India. The average spectral Mean Absolute Percentage Errors (MAPE) obtained for all models in the retrieval of aph(λ), ad(λ), aƍ(λ) and adƍ(λ) lie in the ranges of 26-44%, 37-45%, 34-65% and 42-56%. Slopes of adƍ(λ), aƍ(λ) and ad(λ) as indicated by Sdƍ, Sƍ and Sd are derivable from GSCM, Schofield and Lin's models only. GSCM model exhibited good retrieval capability of Sd with MAPE values of 22% and a correlation coefficient of 0.74. In retrieval of Sƍ parameter, none of the models demonstrated satisfactory performance. Overall, the GSCM and Schofield's models demonstrated good performance in the retrieval of absorption subcomponents, aph(λ), adƍ(λ), ad(λ) and Sd. Effect of applying baseline correction to ad(λ) on model performance is studied. Tuning with in situ data can further improve the absorption subcomponent and slope parameter retrieval capability of the models.