A machine learning approach for spatial mapping of the health risk associated with arsenic-contaminated groundwater in taiwan’s lanyang plain

Ching Ping Liang, Chi Chien Sun, Heejun Suk, Sheng Wei Wang, Jui Sheng Chen

Research output: Contribution to journalArticlepeer-review

2 Scopus citations

Abstract

Groundwater resources are abundant and widely used in Taiwan’s Lanyang Plain. How-ever, in some places the groundwater arsenic (As) concentrations far exceed the World Health Organization’s standards for drinking water quality. Measurements of the As concentrations in groundwater show considerable spatial variability, which means that the associated risk to human health would also vary from region to region. This study aims to adapt a back-propagation neural network (BPNN) method to carry out more reliable spatial mapping of the As concentrations in the groundwater for comparison with the geostatistical ordinary kriging (OK) method results. Cross validation is performed to evaluate the prediction performance by dividing the As monitoring data into three sets. The cross-validation results show that the average determination coefficients (R2) for the As concentrations obtained with BPNN and OK are 0.55 and 0.49, whereas the average root mean square errors (RMSE) are 0.49 and 0.54, respectively. Given the better prediction performance of the BPNN, it is recommended as a more reliable tool for the spatial mapping of the groundwater As concentration. Subsequently, the As concentrations estimated obtained using the BPNN are applied to develop a spatial map illustrating the risk to human health associated with the ingestion of As-containing groundwater based on the noncarcinogenic hazard quotient (HQ) and carcinogenic target risk (TR) standards established by the U.S. Environmental Protection Agency. Such maps can be used to demarcate the areas where residents are at higher risk due to the ingestion of As-containing groundwater, and prioritize the areas where more intensive monitoring of groundwater quality is required. The spatial mapping of As concentrations from the BPNN was also used to demarcate the regions where the groundwater is suitable for farmland and fishponds based on the water quality standards for As for irrigation and aquaculture.

Original languageEnglish
Article number11385
JournalInternational Journal of Environmental Research and Public Health
Volume18
Issue number21
DOIs
StatePublished - 1 Nov 2021

Keywords

  • Back-propagation neural network
  • Groundwater arsenic contamination
  • Hazard quotient
  • Ordinary kriging
  • Target risk

Fingerprint

Dive into the research topics of 'A machine learning approach for spatial mapping of the health risk associated with arsenic-contaminated groundwater in taiwan’s lanyang plain'. Together they form a unique fingerprint.

Cite this