A Machine Learning Approach for Spatial Mapping of the Health Risk Associated with Arsenic-Contaminated Groundwater...

Area of Science:

• Environmental Science
• Hydrogeology
• Data Science

Background:

• Groundwater in Taiwan's Lanyang Plain is vital but faces arsenic (As) contamination exceeding safe drinking water standards.
• Significant spatial variability in groundwater As concentrations poses regional human health risks.
• Accurate spatial mapping is crucial for risk assessment and water resource management.

Purpose of the Study:

• To compare the spatial mapping accuracy of a back-propagation neural network (BPNN) against ordinary kriging (OK) for groundwater arsenic (As).
• To develop a reliable spatial map of As concentrations for assessing human health risks and water suitability.
• To evaluate the effectiveness of BPNN in identifying areas with high As contamination.

Main Methods:

• Employed a back-propagation neural network (BPNN) for spatial mapping of groundwater arsenic (As) concentrations.
• Utilized geostatistical ordinary kriging (OK) as a comparative method for spatial prediction.
• Performed cross-validation using monitoring data to assess prediction performance (R^2 and RMSE).

Main Results:

• BPNN demonstrated superior prediction performance with a higher average determination coefficient (R^2 = 0.55) compared to OK (R^2 = 0.49).
• BPNN achieved a lower average root mean square error (RMSE = 0.49) than OK (RMSE = 0.54), indicating greater accuracy.
• The study successfully generated spatial maps of As concentrations and associated health risks (HQ, TR).

Conclusions:

• BPNN is recommended as a more reliable tool for spatial mapping of groundwater arsenic (As) concentrations.
• The developed maps effectively delineate high-risk areas for human health and identify suitable water uses for irrigation and aquaculture.
• This approach aids in prioritizing areas for intensive groundwater monitoring and informing water resource management decisions.