Using generalized additive models for water quality assessments: A case study example from Australia

Abstract

Water quality management is an ongoing challenge for coastal managers. They are faced with disentangling the multiple determinants involved when assessing the utility of management interventions, understanding the processes behind historical trends and progress towards future water quality goals. Nonparametric statistical methods such as locally weighted scatterplot smoothing are often used in water quality assessments for this purpose while generalized additive models (GAMs) have been applied sparingly. Conversely, the extensive use of GAMs for air quality studies because of their reported ability to account for nonlinear confounding effects of seasonality, covariate trends and weather variables indicates that this is a statistical method that is well-suited to water quality studies. In this paper, we present a case-study application of GAMs in demonstrating the potential for this methodology to be used for trend analysis of water quality datasets. The case study is based upon an extensive water quality monitoring program that recently took place along the coastal region of the Gold Coast, Queensland, Australia. We use GAMs to uncover the functional relationships between a common water quality indicator (turbidity) and the suite of predictor variables that are expected to influence turbidity. The selection of suitable candidate predictors to trial in the model is based on expert judgement regarding the key determinants of turbidity in the case study area and is partially undertaken to minimise the effects of 'colinearity' in the predictor variables. Overall, the GAM approach performed well and provided insight into the drivers of turbidity for the case study area.