Modelling chlorine residuals in drinking water: a review

Abstract

World Health Organization’s guidelines on water quality limit concentrations of residual chlorine in drinking water to the range 0.2–5 mg/l. Modelling tends to be applied to understand how chlorine concentrations can be kept within the recommended limits. In this line, we reviewed 105 articles to show advances in modelling of chlorine residuals while focussing on both data-driven statistical models and process-based models. A total of 83 and 17% reviewed articles applied process-based models and statistical models, respectively. The most influential water parameters which were reported for chlorine decay were pH and temperature. For statistical models, modellers reported a wide range of sizes of training, testing, validation sub-samples, and number of neurons in the hidden layers of the network. Thus, the use of novel fitness function to concurrently seek for the most accurate and compact solution was recommended. Most studies applied coefficient of determination (despite its issues such as failure to quantify bias) to evaluate model performance. We recommended revised coefficient of determination and hydrological model skill score to be used as “goodness-of-fits” metrics since they can quantify model’s bias, and capacity to reproduce observed variability. We found that many modellers portrayed a common practice of not providing sufficient information (such as values of parameters) regarding their modelling results. For instance, 47% of the reviewed articles did not expressly specify the order of reaction in their chlorine decay modelling studies. The practice of not reporting sufficient pertinent information can affect reproducibility of results and hinder model improvement which would arise from possible follow-up studies.