Contaminant source identification using semi-supervised machine learning.

Area of Science:

• Hydrogeology
• Geochemistry
• Environmental Science

Background:

• Groundwater mixing and contamination source identification are complex hydrogeological challenges.
• Characterizing geochemical signatures in aquifers requires sophisticated inverse modeling.
• Existing methods often struggle with unknown numbers of sources and mixing ratios.

Purpose of the Study:

• To develop a novel approach for identifying original groundwater types and contaminant sources in geochemical mixtures.
• To address the challenge of unknown numbers of sources and their geochemical concentrations.
• To provide a method that does not require additional site-specific information.

Main Methods:

• Utilizes Non-negative Matrix Factorization (NMF) for Blind Source Separation (BSS).
• Incorporates a custom semi-supervised clustering algorithm, termed NMFk.
• Applies the method to geochemical data including concentrations, ratios, and delta notations.

Main Results:

• The NMFk methodology successfully identifies the unknown number of groundwater types.
• It accurately determines the original geochemical concentrations of contaminant sources.
• The approach was validated using both synthetic and real-world hydrogeological data.

Conclusions:

• NMFk offers a robust solution for deconvolving complex groundwater mixtures.
• This method enhances the ability to pinpoint contamination origins in aquifers.
• It provides a valuable tool for hydrogeologists and environmental scientists studying groundwater systems.