Decision-Making Improvement and Modelling Tools for Drinking Water Treatment during Rainfall Events.

The presence of particles and natural organic matter (NOM) in raw water (RW) is undesirable and require their removal by drinking water treatment plants (DWTPs). To ensure effective treatment, DWTPs usually monitor surrogate parameters such as turbidity (for particles) and UV254 absorbance (for NOM). Rainfall and subsequent river flow events in watersheds can lead to RW quality degradation, requiring adjustments to chemicals dosages at DWTPs. However, changes in RW quality may occur hours or days after a rainfall, which can complicate decision-making for DWTPs operations. This study aims to develop a procedure for selecting input variables and modelling RW quality after rainfall and river flow peak events. Cross-correlation analyses were conducted on several both rain gauge and flow rate time series in a watershed, along with RW data collected at the water intake of a DWTP. The analyses revealed that RW turbidity and UV254 increased at different time lags after rainfalls and flow peaks. The input variables with the highest correlations between timelagged rainfall and river flow were used to predict turbidity and UV254 using two machine learning models. The study found that the maximum correlation coefficient between flow peaks and turbidity was observed after a few hours, while for UV absorbance, it was observed after a few days. This difference in behavior adds complexity to drinking water treatment practices, which must consider these factors in operational schemes, such as adjusting chemical dosages. The results of the present study will help in developing more effective chemical dosing strategies to remove key contaminants.