Kim, J. and Choi, J, 2016. Data-driven modeling of coastal water quality using the Bayesian method for coastal management. In: Vila-Concejo, A.; Bruce, E.; Kennedy, D.M., and McCarroll, R.J. (eds.), Proceedings of the 14th International Coastal Symposium (Sydney, Australia). Journal of Coastal Research, Special Issue, No. 75, pp. 647–651. Coconut Creek (Florida), ISSN 0749-0208.
To understand and a predict a coastal water quality system, a data-driven statistical model has been proposed using the Bayesian method and applied to the Saemangeum tidal lake. To describe a coastal water quality system, a multivariate statistical model was derived by determining observed variables and their interrelationships such as sea surface temperature, salinity, Chl-a, DO, pH, TN, TP, COD, NH4N, NO2N, NO3N, PO4O, and SiO2Si for parameters of coastal marine environments, coastal water quality, and nutrients using observed field data. To estimate this statistical model, a Bayesian approach using Markov chain Monte Carlo method was applied to identify an optimal data-driven model. There are no limitations of statistical assumptions for samples using the Bayesian method, which is required in a frequentist approach, such as the maximum likelihood method. The Saemangeum tidal lake's coastal water quality system was quantitatively described and assessed by interpreting coefficients of model parameters with relation among variables from a derived structural equation model. Moreover, a prediction for coastal management was possible by Bayesian inference. Thus, there are new findings on the salinity threshold necessary to maintain optimal water by improving degraded water quality. Based on the findings, a quantity of water mixing (exchaning fresh water through sluice gates) can be applied while continuing construction of land reclamation.