Machine Learning Boosts Robust Ocean Color Retrieval in Complex Atmosphere and Ocean Systems: Taking Coastal Forel–Ule Index Mapping as an Example

Accurate satellite ocean color retrieval in complex atmosphere-ocean systems faces significant challenges. Traditional algorithms relying on remote sensing reflectance (Rrs(λ)) are hindered by considerable uncertainty or invalid values due to imperfections in or the failure of atmospheric correction. This study addresses these limitations by leveraging Machine Learning (ML) to explore Rayleigh-corrected reflectance (Rrc(λ)) data towards accurate Forel-Ule Index (FUI) retrieval in the complex atmosphere-ocean environment of the China adjacent seas. First, satellite-ground matchups were constructed with 1,005 in-situ FUI data and quasi-synchronized Moderate-resolution Imaging Spectroradiometer (MODIS) Rrc(λ) images (0rc(λ) images. The new product shows a 10% (7%) increase in percentage of valid data for daily (monthly) time series, with notable improvements of 17% (24%) in coastal waters (within 20 km from coastline), when compared to Rrs(λ)-derived FUI products. Both the high accuracy and complete spatio-temporal coverage help revealing short-term FUI variability after a heavy rainfall and long-term FUI changes missed by Rrs(λ)-derived products. The proposed approach may provide some insights for retrieving other ocean color parameters and contribute to high-accuracy remote sensing retrievals in complex atmosphere-ocean systems.