Machine Learning-Based Models for Water Quality Prediction in Large River Systems

Ensuring optimal water quality in large-scale river systems remains a critical environmental challenge. Poor water quality can significantly affect aquatic ecosystems, leading to increased fish mortality, impaired health, and disruptions in biodiversity. While traditional methods like the Water Quality Index (WQI) and Water Quality Classification (WQC) are constrained by their reliance on periodic laboratory analyses, recent studies have shifted towards using sensor data combined with machine learning (ML) models for more accurate and real-time monitoring. This study enhances water quality monitoring by accurately predicting Dissolved Oxygen (DO) levels, which are a key indicator of water quality, in the rivers of Flanders, Belgium, through the application of machine learning models. The results showed that the Long Short-Term Memory (LSTM) model outperformed other models in capturing the intricate temporal patterns of dissolved oxygen (DO) variations. It demonstrated robust performance in both single-step and multi-step predictions, particularly in detecting critical DO levels (<6 mg/L), which are indicative of poor water quality. Additionally, the incorporation of confidence intervals into the predictions provided a more reliable assessment of forecasting performance. The findings of this study establish a robust predictive framework for large-scale water quality monitoring, providing valuable insights for the protection and mitigation of aquatic ecosystems.