Sea Surface Salinity Forecasting with a Comparison Studying Case of GMM-VSG and FB-Prophet Model

Abstract

With the evolution of artificial intelligence, the utilization of machine learning algorithms for predicting hydrological data has gained popularity in scientific research, especially for the development and operational patterns of marine-related objects in coastal regions.  Salinity analysis plays a crucial role in evaluating the resilience and health of marine ecosystems. Traditional numerical models, although accurate, require significant computational resources. Therefore, this study assesses the effectiveness of GMM-VSG proposed by Shanghai University and FB-Prophet created by Meta (Facebook) as rapid alternatives for simulating the nonlinear relationships between salinity and various parameters, like tide-induced free surface elevation, river flows, and wind speed.  The algorithms were tested using an eight-year dataset collected at the MAREL buoy at the entrance to bay of Brest. Results indicate that, despite the simplicity of the input data, both algorithms successfully reproduced seasonal and semi-diurnal fluctuations in salinity. This underscores their potential as complementary tools for the ecological monitoring in estuarine environments.