GOCI-II based sea surface salinity estimation using machine learning for the first-year summer

ABSTRACT

Estimation of sea surface salinity (SSS) using the Geostationary Ocean Color Imager-II (GOCI-II) measurements in the East China Sea (ECS) was conducted from July to September of 2021, when the first-year observations are available after GOCI-II launch in 2020. The SSS in the ECS is mainly affected by the Changjiang River plume, which varies from under 20–35 psu, and the discharged freshwater disperses from the river mouth towards Jeju Island, Korea. For the SSS estimation, a multi-layer perceptron neural network (MPNN) was employed to train the nonlinear processes of GOCI-II spectral measurements as inputs and the SSS of Soil Moisture Active Passive (SMAP) as the target. Because GOCI-II has four additional spectral bands (380, 510, 620 and 709 nm) compared to the bands in the first generation of GOCI (413, 443, 490, 555, 660 and 680 nm), we developed a new MPNN algorithm and analysed (1) how much these new spectral measurements increase SSS accuracy, and (2) how the enhanced spatial and temporal resolution of GOCI-II make SSS features different from GOCI. The first results showed that the root mean square error (RMSE) and coefficient of determination (R²) were 0.68 psu and 0.92, respectively. Furthermore, R², when compared with in-situ measurements at the Ieodo Ocean Research Station (I-ORS), increased as much as 0.23 (0.20–0.43) for the 10-band model, which performed much better than previous 6-band model. The second result suggests how the improved spatial features at more frequent SSS measurements can be utilized to avoid low salinity intrusion into the aquafarming sites off the coast of Korea. More importantly, we elucidated why the MPNN algorithm performs better than conventional SSS estimation methods by comparing various optical properties with SSS variation; thus, the newly developed model can provide SSS not only in ECS but also low salinity water near southwest Korean coasts at hourly spatial resolution of 250 m.

KEYWORDS:

• GOCI-II application
• SSS estimation
• Changjiang diluted water
• Machine learning application

Acknowledgements

The authors thank the data providers in this article, including the Korea Hydrographic and Oceanographic Agency (KHOA), the Remote Sensing Systems, and the National Aeronautics and Space Administration (NASA).

Disclosure statement

No potential conflict of interest was reported by the author(s).

Correction Statement

This article has been corrected with minor changes. These changes do not impact the academic content of the article.

Additional information

Funding

The work was supported by the National Research Foundation of Korea [2020R1A6A3A13075125]; Ministry of Oceans and Fisheries, Korea [20180456]