GSA Connects 2021 in Portland, Oregon

Paper No. 40-3
Presentation Time: 2:05 PM


PODGORSKI, Joel, ARAYA, Dahyann and BERG, Michael, Water Resources and Drinking Water, Eawag, Swiss Federal Institute of Aquatic Science and Technology, Überlandstrasse 133, Dübendorf, 8600, Switzerland

Groundwater contamination by geogenic arsenic, fluoride or manganese affects aquifers around the world and can present significant hazards to human health in the concentrations frequently found in nature. Due to such contaminants generally not being measured in groundwater quality analyses and the fact that their health effects can be similar to those stemming from other causes, their presence can go undetected for a long time. In order to better assess the extent of geogenic contaminants in groundwater, we have used statistical learning methods such as logistic regression and random forest to create prediction maps of the probability of a contaminant exceeding a specific drinking water guideline at country to global scales. As these maps provide the risk of exposure to a contaminant, they can be used by water managers and other government and non-government agencies to prioritize areas for groundwater quality testing and raise awareness. Furthermore, they help work toward Sustainable Development Goal (SDG) 6 with regard to the provision of safe drinking water.

In this paper, we will present some of our recent work on global prediction models of arsenic and fluoride in groundwater as well as country- to regional-scale models of arsenic, fluoride and manganese. In addition, we will discuss some of the common considerations with respect to modeling. Regardless of the contaminant being modeled or the spatial extent of the model, the measurements of concentrations in groundwater must be compiled along with an appropriate set of potential predictor variables that subsequently get associated with each measurement location and later get used to generate a probability map. As well, different machine learning methods exist as do various statistics for evaluating the performance of a model. As such, we will discuss some of the more important methods and considerations to keep in mind from our experience. A brief overview of how to conduct such statistical modeling, for example, with the R programing language or the Groundwater Assessment Platform, will also be given.