Water Quality Model Structure Identification Using Dynamic Linear Modeling: River Cam Case Study Revisited

This paper investigates the use of dynamic linear modeling and maximum likelihood estimator for water quality model structure identi cation. In addition to the posterior trajectories of model's parameters, the proposed method also examines the trajectory of the estimated prediction error variance. The premise is that the model predictability should be improved as we move down in a time series. If absurd variation in either the trajectories of model's parameter or the trajectory of the model's prediction error variance is observed, the adequacy of the candidate model should be questioned. This method is applied to three candidate models using the time series data from the River Cam, and it is shown that both the trajectories of model's parameters and the trajectory of prediction standard deviation are important in exposing the structural weakness of a candidate model.