Wastewater quality estimation through spectrophotometry-based statistical models

Abstract

Local administrations are increasingly demanding real-time continuous monitoring of pollution in the sanitation system to improve and optimize its operation, to comply with EU environmental policies and to reach European Green Deal targets. The present work shows a full-scale Wastewater Treatment Plant field-sampling campaign to estimate COD, BOD5, TSS, P, TN and NO3−N in both influent and effluent, in the absence of pre-treatment or chemicals addition to the samples, resulting in a reduction of the duration and cost of analysis. Different regression models were developed to estimate the pollution load of sewage systems from the spectral response of wastewater samples measured at 380–700 nm through multivariate linear regressions and machine learning genetic algorithms. The tests carried out concluded that the models calculated by means of genetic algorithms can estimate the levels of five of the pollutants under study (COD, BOD5, TSS, TN and NO3−N), including both raw and treated wastewater, with an error rate below 4%. In the case of the multilinear regression models, these are limited to raw water and the estimate is limited to COD and TSS, with less than a 0.5% error rate.