O efeito das variações sazonais nos parâmetros de qualidade de uma estação de tratamento de água no Ceará

Abstract

At a Water Treatment Plant (WTP) in the Metropolitan Region of Fortaleza (MRF), water quality parameters (both raw and treated) were correlated with rainfall indices, treatment efficiency was evaluated, and chemical dosages were predicted using machine learning between 2020 and 2023. The monitored parameters included color, turbidity, free residual chlorine, total coliforms, and Escherichia coli. Machine learning models, specifically polynomial regression and XGBoost, were tested and evaluated using determination coefficients (R²) and average absolute relative deviation (AARD%) to determine which model was more suitable for predicting chemical dosages. The graphical evaluation shows a profile with random variations, with parameter increases during both rainy and dry seasons, caused by surface runoff from rainfall and intermittent sewage discharges. Despite this, the parameter values remained below the maximum limits allowed by Brazilian regulations. The removal efficiency for color and turbidity at the treatment output exceeded 90% in most years. Few cases of total coliform contamination were identified post-treatment, which may be due to isolated post-treatment contamination events, unrelated to the treatment process. Regarding chemical dosage predictions, the XGBoost model proved to be the most efficient, with determination coefficients (R²) above 0.85 for most chemicals and low average absolute relative deviation (AARD%). High concentrations of the parameters were observed during some rainy and dry periods, which may result from surface runoff carrying sediments and the discharge of domestic and industrial sewage into the water body. While these natural and human activities impact water quality parameters, their levels remain well below the maximum limits set by Brazilian legislation. Excellent performance in removing color and turbidity was observed after disinfection, confirming the treatment’s efficiency. Although total coliform contamination was identified, likely due to isolated incidents, the treatment was generally effective in removing microorganisms. Based on the metrics, the application of machine learning algorithms, such as XGBoost, proved to be effective, delivering satisfactory results for most chemical dosages.