Remote sensing of absorption and scattering coefficient using neural network model: Development, validation, and application

The total absorption (a(λ)) and backscattering (bb(λ)) coefficients of natural waters are the most significant factors affecting light propagation within water columns, and thus play indispensable roles in the estimation of aquatic biomass, primary production, and carbon pools. Despite its importance, no accurate retrieval model has been specifically developed for both oceanic and coastal waters, but significant efforts have been made in regard to oceanic inversion models. The objectives of the present study are to evaluate the applicability of the quasi-analytical algorithm (QAA) in deriving a(λ) and bb(λ) from oceanic and coastal waters, and to improve it using a neural network-based semi-analytical algorithm (NNSAA). Based on a comparison of the a(λ) and bb(λ) predicted by these models with field measurements taken from the national aeronautics and space administration bio-optical marine algorithm dataset (NOMAD), the Yellow Sea and China East Sea, it is shown that the NNSAA model (R2>0.82 and mean relative error, MRE=20.6–35.5%) provides a stronger performance than the QAA model (R20.77 and MRE