Jha, Achla Om Prakash (2020-04). Predicting Private Well Water Quality Impacted by Hurricane Harvey Using Principal Component Analysis and GIS-Based Kriging. Master's Thesis. Thesis uri icon


  • When Hurricane Harvey struck Texas in 2017, bringing an unprecedented 102-155 cm of rain to the central and upper Gulf Coast areas and causing widespread flooding, an opportunity developed to conduct a citizen science campaign to investigate water quality in wells impacted during a two-month period following the storm. This study evaluated inorganic contamination in private wells potentially impacted by Hurricane Harvey by 1) characterizing levels of inorganics and fecal indicator bacteria (FIB) in private wells and comparing concentrations to the USEPA drinking water standards for public water systems and 2) assessing the geographic risk variation of contaminants through application of Kriging geospatial interpolations. More than 400 well users participated in the seven free well water sampling events offered in 15 hurricane-impacted counties. Arsenic was detected (>1 microgram/L) in 80.9% of wells sampled, with 3.4% exceeding 10 microgram/L. Lead concentrations exceeded the action level of 15 microgram/L in 3.4% of samples and was detected (>1 microgram/L) in 25.3% of wells. Iron exceeded 300 microgram/L in 22.9% of wells and was detected (>10 microgram/L) in 71.0%. Twenty-three percent of the samples exceeded 50 microgram/L of manganese and 69.0% of samples had detectable concentrations. The median contaminant concentrations were higher in shallow wells than in deep wells. Also, median contaminant concentrations were higher in wells that had submerged well heads during flooding than wells which did not have submerged well heads. Principal component analysis for elements with secondary drinking water standards with three principal components yielded a cumulative variance of 69.3% in comparison to two principal components with a cumulative variance of 57.5% for elements with primary drinking water standards. Ordinary kriging, universal kriging and empirical Bayesian kriging models were used to interpolate inorganics concentrations and the principal component scores. Empirical Bayesian kriging gave the lowest root mean square error for most variables and was selected as the optimal method. Kriging could have improved if applied to a smaller study area. Study results indicated that about 43.9% of the wells exceeded at least one EPA drinking water standard. Well system characteristics and especially, well head flooding status also likely affected concentrations of inorganic elements. For three kriging methodologies tested, Empirical Bayesian kriging was most accurate.

publication date

  • April 2020