An integrated data-driven framework for surface water quality anomaly detection and early warning

The surface water quality anomaly detection for rapid early warning is essential to prevent potential harmful compounds, resulting from river environmental spills or intentional releases, from dispersing in large scale. In this study, an effective data-driven framework for surface water quality anomaly detection is developed to provide early warnings for dealing with river environmental pollution in advance. The developed framework is constructed by an integration of Bayesian autoregressive (BAR) model for water quality variation prediction and Isolation Forest (IF) algorithm for water quality anomaly detection. First, an autoregressive method based on Bayesian inference is used to forecast the tendencies of water quality variations. Second, an IF algorithm is applied to identify the features of water quality anomalies using the prediction residuals obtained in the previous stage. The integration framework is then applied to analyze and detect the surface water quality variations and anomalies in Potomac River of West Virginia, USA, comparing with prediction-based anomaly detection method, classification-based anomaly detection method, and different scenarios. The results demonstrate that the developed integration framework not only could enhance water quality anomaly detection accuracy, but also effectively provide early warning for emergency operations in a quick response.