Drinking Water Source Monitoring Using Early Warning Systems Based on Data Mining Techniques

Abstract

Improving drinking water source monitoring is crucial for efficiently managing the drinking water treatment process and ensuring the delivery of safe water. Data mining techniques could prove useful to help forecast source water quality. In this study, two approaches were used to forecast turbidity mean levels and peaks in the main drinking water source of the city of Québec, Canada. Trend analysis was applied for the prediction of significant turbidity events (>99th percentile of data distribution). Artificial neural networks using antecedent moisture conditions as input parameters (all turbidity peaks) served to forecast daily turbidity time series. Results show that trend analyses help anticipate the timing of turbidity peaks ― with differences between the cold season (fall and winter) and the warm season (spring and summer) and mean anticipations between 45 and 85 min and 25 and 45 min, respectively ― and the magnitude of the peak. The artificial neural network model was developed and proven capable of predicting the mean levels of turbidity at the drinking water intake of the investigated catchment. These early warning systems could be applied to source water system forecasting and provide a framework for adjusting drinking water treatment operations.