Determination of CDOM and NPPM absorption coefficient spectra from coastal water remote sensing reflectance

Multilayer perceptron (MLP) neural network algorithms were developed to retrieve the absorption coefficient spectra of the colored dissolved organic matter (CDOM) and nonpigmented particulate matter (NPPM) from the remote sensing reflectance R_rs of optically complex waters. The two MLP algorithms, consisting of one hidden layer with ten neurons and requiring R_rs at 412, 490, and 665 nm as inputs, were trained with a comprehensive experimental dataset of the Northern Adriatic Sea coastal waters. The products of the proposed regional MLP algorithms showed higher accuracies than regional band-ratio algorithms, and exhibited average uncertainties of 20% and 25% in the determination of CDOM and NPPM absorption coefficients at 412 nm, respectively.