POSTER Session 1

Estimating Inherent Optical Properties using Remote-Sensing Reflectance based on machine learning

Accurate estimation of Inherent optical properties (IOPs) is crucial for effective monitoring and management of marine environments. Utilizing satellite data has proven highly efficient for this purpose. Over the years, various models have been developed, employing diverse mathematical formulations and optical theories, to delineate the relationship between remote-sensing reflectance (Rrs) and IOPs. This study focuses on estimating IOPs for the 2nd Geostationary Ocean Color Imager (GOCI-II) using a machine learning approach. Due to the lack of in-situ data, simulation datasets from the HydroLight were used for the training that are form Rrs at 412, 443, 490, 510, 555, 620, and 660 nm and their spectral ratios to the absorption coefficient of phytoplankton (aph) (443 nm), the combination of detritus and gelbstoff (adg) (443 nm), and the backscatter coefficient of particles (bbp) (555 nm). The results of the machine learning models were compared with other primary methods that are the quasi-analytic algorithm (QAA) and the Generated Inherent Optical Properties (GIOP) models based on the simulation datasets. The machine learning models demonstrated more accurate results than both the QAA and GIOP models for all three IOP parameters especially for optically complex turbid waters. The result implies the potential of machine learning in deriving IOPs from satellite data, suggesting a promising avenue for enhancing marine environmental monitoring.

Eunna Jang, Korea Ocean Satellite Center, Korea Institute of Ocean Science and Technology (KIOST), Republic of Korea, https://orcid.org/0000-0002-4474-631X

Jae-Hyun Ahn, Korea Ocean Satellite Center, Korea Institute of Ocean Science and Technology (KIOST), Republic of Korea

Code of Conduct
Privacy Policy