Solomon White ^1* Tiago Silva ² Laurent Amoudry ³ Evangelos Spyrakos ⁴ Adrien Martin ⁵ Encarni Medina-Lopez ⁶

• ¹ School of Engineering, University of Edinburgh, Edinburgh, United Kingdom
• ² Centre for Environment, Fisheries and Aquaculture Science (CEFAS), Lowestoft, United Kingdom
• ³ Proudman Oceanographic Laboratory, National Oceanography Centre, University of Southampton, Liverpool, Hampshire, United Kingdom
• ⁴ University of Stirling, Stirling, Scotland, United Kingdom
• ⁵ Noveltis (France), Labège, France
• ⁶ University of Edinburgh, Edinburgh, Scotland, United Kingdom

Understanding and monitoring sea surface salinity (SSS) and temperature (SST) is vital for assessing ocean health. Interconnections among the ocean, atmosphere, seabed, and land create a complex environment with diverse spatial and temporal scales. Climate change exacerbates marine heat waves, eutrophication, and acidification, impacting biodiversity and coastal communities. Satellite-derived ocean colour data offers enhanced spatial coverage and resolution compared to traditional methods, enabling estimation of SST and SSS. This paper presents a methodology to extract these properties using machine learning algorithms trained with in-situ and multispectral satellite data. Our global neural network model achieves an R2 of 0.83 for temperature and 0.65 for salinity. In the specific case study in the Gulf of Mexico, root mean square error (RMSE) for temperature was 0.83°C for test cases, and 1.69°C for validation, outperforming previous methods in coastal dynamic environments. Feature importance is analysed using Shapley values. These reveal key spectral features influencing SST and SSS estimation, highlighting the significance of factors like solar azimuth angle and specific bands. Infrared bands play a crucial role in SST prediction, while blue/ green colour bands are more influential for SSS. Our approach addresses the "black box" nature of machine learning models, providing insight into the relative importance of spectral bands.