Predicting water turbidity in a macro-tidal coastal bay using machine learning approaches

Abstract

Water turbidity is of particular importance for diffusion and migration of nutrients and contaminants, biological production, and ecosystem health in coastal turbid areas. The estimation of water turbidity is therefore significant for studies of coastal dynamics. Many factors influence turbidity in complex and nonlinear ways, making accurate estimations of turbidity a challenging task. In this study, three machine learning models, Artificial Neural Networks (ANN), Genetic Programming (GP), and Support Vector Machine (SVM) are developed for better estimation and prediction of the tidally-averaged sea surface turbidity. The observational data of tides and waves at a macro-tidal coastal bay, Jiangsu coast, China are used as model inputs. Through data reduction, it is found that tidal average sea surface turbidity is most determined by the average tidal range of the two preceding tidal cycles (2 and 3 tidal periods before the present one, respectively) and the tidal average significant wave height of the present tidal cycle of turbidity. These three machine learning models all show successful estimations of turbidity, and comparisons of the optimized models indicate that ANN shows the best performance and GP helps to provide physically meaningful predictors. This study provides an example of developing a predictive machine learning algorithm with a limited dataset (94 tidal cycles). The generality of the present predictors can be reinforced with much more data from a variety of coastal environments.