The inherent and apparent optical properties (IOPs and AOPs) of seawater limit the performance of free-space optical (FSO), underwater wireless optical communication (UWOC), and imaging systems. Absorption, scattering, and downwelling irradiance are three such properties that influence system performance and often evolve independently. In situ measurements of multiple IOPs and AOPs would provide environmental sensing for fielded optical systems, but such comprehensive measurements are typically expensive or impractical. This effort analyzed existing oceanographic data sets to uncover wavelength-dependent correlations between IOPs, AOPs, test depths, and ocean depths. We then employed machine learning (ML) methods to predict the optical properties of diffuse attenuation (Kd) and backscatter (bb) using beam attenuation (c) and compared these results to ground-truth values. Predicted values of Kd and bb were well matched to their ground-truth data. Furthermore, we demonstrate ML-based Jerlov optical water type classification using beam attenuation as the optical data input. With our methods validated, we collected new optical data sets and processed them using our ML models. Results are promising and indicate future in situ classification and prediction capability. To highlight one practical application, we present a preliminary ML-enabled performance estimator for a generic FSO or UWOC system.
|