Análise temporal da qualidade das águas do médio Paraíba do Sul no período (2012-2022)

Abstract

The aim of this study is to accurately predict the water quality at strategic monitoring points over a decade through the combined use of statistical methods and artificial intelligence. This interdisciplinary research stands out for the innovative application of General Regression Neural Networks (GRNN), implemented using the R statistical software, to forecast water quality at nine specific points along the Paraíba do Sul River. These points were strategically selected based on environmental relevance and availability of historical data. Once the monitoring points were defined, detailed and reliable hydrological and environmental data were obtained from the INEA (Instituto Estadual do Ambiente) database. The data underwent rigorous preprocessing and statistical treatment, employing methods such as Analysis of Variance (ANOVA) and Multiple Linear Regression to identify significant variables and trends. Subsequently, the data were fed into artificial neural network models, specifically the GRNN architecture, which was trained and validated using robust performance metrics, including RMSE, R², and MAPE, to assess predictive reliability. A distinguishing feature of this study is its integration of environmental science with advanced automation and data analysis techniques—fields in which the author has professional expertise. Drawing on extensive experience in industrial automation, instrumentation, and control systems, particularly in critical infrastructure sectors such as energy and petrochemicals, the study benefits from a systems-oriented approach. The design and implementation of the predictive model reflect engineering precision, rigorous data handling, and applied knowledge of process optimization. Additionally, the study considers the influence of diffuse pollution sources in the basin, such as pasture areas and cattle herds, which play a key role in altering water quality. These anthropogenic factors were incorporated into the analytical model, contributing to a more comprehensive understanding of the interactions between land use and river health. The results demonstrate the high predictive accuracy of the neural network model, confirmed by its performance when applied to real-world data. Beyond its technical contributions, this research offers valuable insights for environmental management, providing a solid foundation for decision-making processes related to water resource monitoring, conservation strategies, and public policy. It supports sustainable development goals by enhancing predictive capabilities and promoting proactive intervention in water quality degradation. In summary, this dissertation not only advances methodological innovation by combining statistical modeling and artificial intelligence, but also contributes practical solutions for the management and recovery of aquatic ecosystems in one of Brazil’s most important river basins.