Analyzing spatio-temporal dynamics of dissolved oxygen for the River Thames using superstatistical methods and...

Area of Science:

• Environmental Science
• Data Science
• Statistical Physics

Background:

• River water quality monitoring is crucial for ecological health.
• Time series analysis of water quality indicators presents complex dynamics.
• Superstatistics and machine learning offer advanced tools for analyzing such data.

Purpose of the Study:

• To analyze River Thames water quality time series data, focusing on dissolved oxygen dynamics.
• To model dissolved oxygen fluctuations using superstatistical methods and identify effective detrending techniques.
• To develop and evaluate machine learning models for predicting dissolved oxygen concentrations.

Main Methods:

• Superstatistical analysis employing q-Gaussian distributions.
• Multi-resolution analysis using multiplicative Empirical Mode Decomposition for detrending.
• Machine learning models including Light Gradient Boosting Machine and Transformer (Informer) for prediction.
• SHapley Additive exPlanations (SHAP) for feature importance analysis.

Main Results:

• Dissolved oxygen fluctuations exhibit heavy tails, well-modeled by q-Gaussian distributions.
• Multiplicative Empirical Mode Decomposition was the most effective detrending method.
• Light Gradient Boosting Machine performed best for same-time prediction, with temperature, pH, and time of year as key predictors.
• The Informer model achieved superior long-term forecasting performance, identifying daily dissolved oxygen cycles.

Conclusions:

• Geographical factors, like distance to the sea, influence water quality dynamics.
• Advanced machine learning models, particularly the Informer, are effective for long-term river water quality forecasting.
• Findings support policymakers in ecological health assessments and maintaining aquatic ecosystems.