Seawater intrusion pattern recognition supported by unsupervised learning: A systematic review and application.

Area of Science:

• Hydrogeology
• Environmental Science
• Data Science

Background:

• Seawater intrusion is a major global groundwater contaminant, threatening water resources, agriculture, and ecosystems.
• Multivariate analysis and unsupervised learning are established tools for analyzing water quality data and identifying patterns.

Purpose of the Study:

• To systematically review and bibliometrically analyze unsupervised learning techniques for identifying seawater intrusion.
• To provide practical recommendations for data preprocessing, research, and validation in hydrogeological studies.

Main Methods:

• Systematic literature review following PRISMA guidelines.
• Bibliometric analysis of 102 coastal hydrogeological studies.
• Application and explanation of principal components analysis (PCA), hierarchical clustering, K-means, and self-organizing maps using R software.

Main Results:

• 74% of studies using dimensional reduction (e.g., PCA) linked variance to salinization.
• 77% of studies using clustering identified seawater intrusion influence in at least one water sample cluster.
• The review highlights the efficacy of these methods in seawater intrusion assessment.

Conclusions:

• Unsupervised learning techniques, particularly PCA and clustering, are valuable tools for detecting and characterizing seawater intrusion.
• Standardized data preprocessing and transparent reporting are crucial for study replication and validation.
• Further research opportunities exist in refining these methods for more accurate groundwater management.