Using simple and easy water quality parameters to predict trihalomethane occurrence in tap water

Area of Science:

• Environmental Science
• Water Quality Monitoring
• Predictive Modeling

Background:

• Disinfection by-products (DBPs) monitoring in water is vital for public health but often laborious.
• Existing predictive models for DBPs face limitations due to improper datasets or reliance on complex parameters.
• Developing practical models using easily measurable parameters is essential for real-world application.

Purpose of the Study:

• To develop predictive models for trihalomethane (THMs) occurrence in tap water.
• To evaluate the efficacy of simple water quality parameters for THMs prediction.
• To compare the performance of linear regression models (LRM) and radial basis function artificial neural networks (RBF ANN).

Main Methods:

• Utilized four simple water quality parameters: temperature, pH, UVA254, and free chlorine (Cl2).
• Developed THMs predictive models using Linear/log linear regression models (LRM) and Radial Basis Function Artificial Neural Networks (RBF ANN).
• Trained and tested models using 64 tap water samples.

Main Results:

• LRMs showed limited prediction ability with only one or two parameters (testing datasets: N25 = 46-69%, rp = 0.334-0.459).
• RBF ANNs demonstrated significantly improved prediction accuracy by optimizing neuron number and Gaussian spread.
• Optimum RBF ANNs achieved high accuracy (N25 = 85-92%, rp = 0.813-0.886) for predicting total THMs, chloroform, and bromodichloromethane.

Conclusions:

• RBF ANNs, using simple parameters like temperature, pH, UVA254, and chlorine, provide an accurate and efficient method for THMs prediction.
• This approach offers an economic and convenient alternative for monitoring THMs in operational water supply systems.
• The study paves the way for improved, practical DBP monitoring strategies.