부산대학교 도서관

Coccolithophores are one of the dominant algae in Arctic oceans and play an essential role in the carbon cycle given that they are the primary source of ocean particulate inorganic carbon (PIC). Ocean color remote sensing provides a powerful tool to observe the variation in coccolithophore blooms; however, heavy cloud cover prohibits satellite observation coverage and frequency in Arctic oceans, which causes uncertainties in characterizing the phenological features of coccolithophore blooms. In this study, a machine-learning-based empirical approach was developed to extend the quantity of existing standard PIC products from ocean color satellite observations for coccolithophore bloom waters under cloud cover conditions. Results showed that the machine-learning-based approach successfully recovered the PIC product from cloud cover conditions and filled the data gap generated by the default PIC algorithm. The new approach profoundly increased the frequency and coverage of ocean color satellite observations of PIC during coccolithophore blooms and provided detailed information in characterizing the phenology features of coccolithophore blooms. Further evaluation revealed that the machine-learning-based approach has great potential applicability to derive PIC under cloud cover conditions in other regions.