Analyzing Spatio-Temporal Water Quality Dynamics for the River Thames using Superstatistical Methods and Machine Learning

By employing superstatistical methods and machine learning we analyse measured time series of water quality indicators of the River Thames. Particular emphasis is on the dynamics of dissolved oxygen. The probability density functions (PDFs) of dissolved oxygen, with trend substracted, exhibit heavy tails that are well-fitted by q-Gaussian distributions. We investigate how the entropic index q depends on the distance to the sea. Regression analysis reveals feature importances for oxygen concentration predictions, with temperature, pH, and time of the year playing a major role. We also forecast oxygen concentrations through the application of a state-of-the-art machine learning model, the Transformer. Within this context, the Informer model exhibits superior performance. The effectiveness of the Informer model is attributed to the self-attention mechanism, which emphasizes the half-life cycle of dissolved oxygen and its temporal dynamics during periods from morning to early afternoon and from late evening to early morning. Our research can help to inform policymakers in ecological health assessments, assisting in river water quality forecasting and helping to maintain healthy aquatic ecosystems.