Deriving the diffuse attenuation coefficient in the Eastern Mediterranean Sea, using observational optical measurements and a multi-layer perceptron regression model

Abstract

The diffuse attenuation coefficient is an indicator of light availability in the surface layer, and is used in a broad range of applications, including numerical simulations, for the parameterization of the light transmission in the water column. In this study, a new dataset of the diffuse attenuation coefficient for the Eastern Mediterranean Sea test case is developed using an existing optical dataset of 2614 beam attenuation coefficient profiles. This method introduces a way of overcoming the difficulty of measuring the diffuse attenuation coefficient in-situ by utilizing the most routinely measured variable, the beam attenuation coefficient. The proposed approach uses existing semi-analytical relationships and a neural network. The neural network, a multi-layer perceptron regression model, is trained and validated with a dataset of 29398 concurrent bio-optical in-situ measurements from the PROSOPE cruise and remotely sensed surface variables. The model is applied to the Eastern Mediterranean dataset and the results are interpolated into a gridded gap-free field, with a grid resolution of 0.0416° x 0.0416°, which is assessed and compared with a satellite-derived product, investigating their significant differences. The resulting field's mean value is slightly reduced with respect to the satellite product, showing regions of higher turbidity, with the most prominent located in the northern Aegean Sea in regions of excess colored dissolved organic matter and around mesoscale features and in the Cretan and Levantine Sea in regions of higher mesoscale activity.