Differentiating estuarine dissolved organic matter composition by unsupervised and supervised machine learning.

Area of Science:

• Environmental Chemistry
• Biogeochemistry
• Water Quality Analysis

Background:

• Differentiating Dissolved Organic Matter (DOM) composition in estuaries is crucial for understanding biogeochemical processes and trophic cycling.
• Tracing the spatiotemporal variations of estuarine DOM is complex due to multiple sources and transformation processes.
• Human impacts from industrialization and urbanization further complicate estuarine DOM dynamics.

Purpose of the Study:

• To investigate the dynamics of estuarine DOM by analyzing its optical properties.
• To capture DOM variability using machine learning and explainable artificial intelligence.
• To differentiate DOM composition and elucidate involved processes in a human-impacted estuary.

Main Methods:

• Collected 249 sub-surface water samples from the Seine Estuary under contrasting hydrological conditions.
• Analyzed DOM optical properties using UV-Visible absorbance and Excitation-Emission Matrix (EEM) fluorescence spectroscopy.
• Applied unsupervised (K-means clustering) and supervised (Light Gradient Boosted Machine) machine learning, with SHapley Additive exPlanations (SHAP) for analysis.

Main Results:

• Unsupervised machine learning identified three distinct estuarine zones based on spatial variations in DOM optical parameters.
• Supervised machine learning validated the defined zonation, and SHAP analysis revealed zone-specific DOM characteristics.
• Zone I (upper estuary) showed high molecular weight and autochthonous DOM; Zone II (mid-estuary) had autochthonous, aromatic, and degradation products; Zone III (lower estuary) featured aromatic, low molecular weight, and degradation products.

Conclusions:

• A workflow was presented for differentiating DOM composition and tracing its variability along the land-to-sea continuum.
• The study successfully elucidated DOM dynamics and involved processes in the Seine Estuary.
• The developed approach has significant implications for environmental management and can be adapted to other estuarine systems.