Influences of model structure and calibration data size on predicting chlorine residuals in water storage tanks

Pei Hua; Keila Roberta Ferreira de Oliveira; Peter Cheung; Fábio Veríssimo Gonçalves; Jin Zhang

doi:10.1016/j.scitotenv.2018.03.364

Influences of model structure and calibration data size on predicting chlorine residuals in water storage tanks

Sci Total Environ. 2018 Sep 1:634:705-714. doi: 10.1016/j.scitotenv.2018.03.364. Epub 2018 Apr 9.

Authors

Pei Hua¹, Keila Roberta Ferreira de Oliveira², Peter Cheung³, Fábio Veríssimo Gonçalves³, Jin Zhang⁴

Affiliations

¹ Technische Universität Dresden, Institute of Urban Water Management, 01062 Dresden, Germany.
² Fundação Universidade Federal de Mato Grosso do Sul, Faculdade de Engenharias, Arquitetura e Urbanismo e Geografia, Av. Costa e Silva, s/n° - Bairro Universitário, CEP: 79070-900 Campo Grande, MS, Brazil. Electronic address: keila.dassan@ufms.br.
³ Fundação Universidade Federal de Mato Grosso do Sul, Faculdade de Engenharias, Arquitetura e Urbanismo e Geografia, Av. Costa e Silva, s/n° - Bairro Universitário, CEP: 79070-900 Campo Grande, MS, Brazil.
⁴ Institute of Groundwater and Earth Sciences, Jinan University, 510632 Guangzhou, China.

PMID: 29649715
DOI: 10.1016/j.scitotenv.2018.03.364

Abstract

This study evaluated the influences of model structure and calibration data size on the modelling performance for the prediction of chlorine residuals in household drinking water storage tanks. The tank models, which consisted of two modules, i.e., hydraulic mixing and water quality modelling processes, were evaluated under identical calibration conditions. The hydraulic mixing modelling processes investigated included the continuously stirred tank reactor (CSTR) and multi-compartment (MC) methods, and the water quality modelling processes included first order (FO), single-reactant second order (SRSO), and variable reaction rate coefficients (VRRC) second order chlorine decay kinetics. Different combinations of these hydraulic mixing and water quality methods formed six tank models. Results show that by applying the same calibration datasets, the tank models that included the MC method for modelling the hydraulic mixing provided better predictions compared to the CSTR method. In terms of water quality modelling, VRRC kinetics showed better predictive abilities compared to FO and SRSO kinetics. It was also found that the overall tank model performance could be substantially improved when a proper method was chosen for the simulation of hydraulic mixing, i.e., the accuracy of the hydraulic mixing modelling plays a critical role in the accuracy of the tank model. Advances in water quality modelling improve the calibration process, i.e., the size of the datasets used for calibration could be reduced when a suitable kinetics method was applied. Although the accuracies of all six models increased with increasing calibration dataset size, the tank model that consisted of the MC and VRRC methods was the most suitable of the tank models as it could satisfactorily predict chlorine residuals in household tanks by using invariant parameters calibrated against the minimum dataset size.

Keywords: Calibration dataset size; Chlorine decay model; Drinking water storage tank; Model structure; Variable reaction rate coefficient.