Generalized linear mixed modeling for spatiotemporal data outlier detection of emerging contaminants: A multi-stage strategy

• Spatiotemporal outlier detection of emerging contaminants has been challenging • A generalized linear mixed modeling was adopted to address this challenge • A multi-stage strategy was designed to enhance ability of the Bayesian-type modeling • The effectiveness of this strategy was validated by two real-world datasets • A decision-support tool was provided for data outlier analyses of emerging contaminants Accurate data are critical for managing emerging contaminants. However, they are challenged by complex dependencies, multidimensional uncertainties, and the need to integrate diverse ecological drivers. Traditional data outlier detection methods often struggle to simultaneously account for spatial and temporal autocorrelation and effectively addressing uncertainties. To address them, this study proposed a multi-stage strategy based on spatiotemporal Bayesian generalized linear mixed model construction and selection. Within the Bayesian-type modeling framework, this strategy accommodated diverse spatiotemporal data structures through flexible combinations of its components. Following model fitting, a multi-stage and multi-criteria selection process was developed to identify the optimal model. The effectiveness of the strategy was demonstrated by two case studies related to emerging contaminants in natural aquatic environments. These applications showed that the strategy can identify well-supported models that not only capture the inherent spatiotemporal associations but also effectively pinpoint outlier sampling points, with practical supporting evidence. The multi-stage strategy provided a decision support tool for analyzing the environment management and public health prevention and control system of emerging contaminants.