Self-organizing map improves understanding on the hydrochemical processes in aquifer systems

Area of Science:

• Environmental Science
• Hydrogeology
• Data Science

Background:

• Understanding hydrochemical features is vital for water resource management.
• Complex regions present challenges in analyzing hydrochemical changes due to multiple influencing factors.

Purpose of the Study:

• To characterize hydrochemical processes in a complex catchment using machine learning.
• To compare the effectiveness of the self-organizing map (SOM) against traditional classification methods.

Main Methods:

• Collected 208 groundwater samples from Kumamoto, Japan.
• Applied the self-organizing map (SOM) technique to analyze groundwater chemistry data (major cations and anions).
• Compared SOM results with Stiff diagrams, cluster analysis, and principal component analysis (PCA).

Main Results:

• The SOM, with integrated clustering, identified 11 distinct clusters in the complex region.
• SOM provided more detailed insights into hydrochemical and contamination processes than traditional methods.
• Z-transformation normalization yielded the best results, indicated by lower topographic error (TE), ensuring better spatial matching of clusters.

Conclusions:

• The self-organizing map (SOM) is a valuable tool for understanding regional-scale hydrochemical processes.
• This machine learning approach can significantly enhance groundwater resource management strategies.
• Careful application and normalization selection are crucial for accurate SOM results.