Water Quality Prediction in Freshwater Reservoirs under Climatic Variability Using Bayesian Networks

Water quality sampling in freshwater reservoirs is economically and logistically demanding, limiting monitoring frequency. This study evaluates water quality dynamics and identifies key controlling variables in two Mexican reservoirs: Necaxa and Manuel Ávila Camacho (Valsequillo). Using multivariate statistical techniques and Bayesian Networks, a probabilistic machine learning framework was implemented to quantify variable influence and explore scenario-based responses. Results reveal significant spatiotemporal variability driven by local conditions and seasonal factors primarily precipitation and temperature, which exacerbate pollution levels during warm and rainy months. Critical parameters identified include Total Suspended Solids, Total Organic Carbon, Chemical Oxygen Demand, and Escherichia coli, which compromise agricultural use and aquatic ecosystem sustainability. Based on these findings, an optimized scheme of eight key variables is proposed for each reservoir according to its specific use. This research demonstrates that combining Bayesian Networks with multivariate analysis provides a robust decision-support tool for contexts with limited data, promoting efficient, adaptive, and sustainable water resource management strategies.