The data-driven paradigm is widely used in ocean science and includes the statistical modeling of various phenomena in coastal marine environments and data assimilation in numerical models. One of the most important challenges in the data-driven paradigm is finding the model that best approximates the underlying mechanism of a phenomenon with measurement data. In this paper, we propose a Bayesian approach to modeling coastal marine environments using ocean observational data, and we apply it to the Saemangeum coast. There are two main advantages to the Bayesian method: domain knowledge can be encoded to prior probability, and Markov-chain Monte Carlo simulation can be used in model estimation and inference. We apply the method to estimate model parameters and predict coastal water quality and sea current for maintaining optimal coastal water quality. The threshold quantity of sea current is computed to ensure sustainable coastal development. One of the interesting results we have obtained is a flat plateau relationship between the sea current for water exchange and the level of improvement of coastal water quality. This means that coastal water quality is not always being improved, even if the amount of water exchange is increased. The computed results are in good agreement with oceanographic theory, while showing a valid difference compared to the results using the frequentist approach and probabilistic inference using the probabilistic graphical model. These results will be helpful in coastal water quality management, ultimately contributing to sustainable coastal development. (C) 2016 Elsevier Ltd. All rights reserved.