Optimal regression model for the prediction of inactivation rate of E. coli in solar disinfection (SODIS)

Abstract

Solar disinfection (SODIS) is a simple and cost-effective method for disinfecting drinking water with questionable microbial quality by exposing it to sunlight in transparent containers. The effectiveness of SODIS relies on various parameters such as UV intensity (I), water temperature (T), and turbidity (Tu). This study aimed to identify the optimal regression model for predicting the inactivation rate constant of E. coli among 28 possible regression equations using I, T, and Tu as predictors. The 28 regression equations were derived from seven combinations of predictor variables (I, T, & Tu; I & T; I & Tu; T & Tu; I; T; and Tu; and Tu) utilizing four trends (linear, logarithmic, exponential, and power). The proposed models were calibrated using data collected from 33 SODIS experiments conducted over a five-month period from April to August 2021. Based on rankings from the Taylor diagram, the regression equation combining the linear trend and I & Tu as predictors demonstrated the best predictive performance. Residual analysis indicated that square root transformation was necessary to improve normality and homogeneity of residuals. Notably, turbidity within the range of 1 – 30 NTU, previously considered nonsignificant, became significant after the square root transformation. This study underscores the importance of an exhaustive approach that considers all possible combinations of predictor variables and trends, allowing the data to reveal patterns and correlations without premature restrictions, thereby ensuring no potentially valuable insights are overlooked.