Machine-Learning Approach for Identifying Arsenic-Contamination Hot Spots: The Search for the Needle in the Haystack

In the 40 years since the relation between arsenic (As) toxicity and groundwater extraction was first documented from the Holocene alluvial basin of West Bengal, India, (1) we have become more aware that groundwater contamination with naturally occurring (geogenic) As poses a serious health threat of global proportions. (2) With the aim of implementing effective and sustainable mitigation strategies, research into the occurrence and location of toxic As levels in drinking and irrigation water and in the food chain provided insight into all aspects of the As-contamination issue, including (a) geogenic As provenance in volcanic and metamorphic rocks, hydrothermal additions to groundwater and hot springs, and weathering of rocks in orogenic mountain belts, (b) its accumulation in sedimentary-basin aquifers, (c) the mobilization and transport of the contaminant into the groundwater, and (d) the associated health risks of sustained As ingestion for >200 million people around the world. (3,4) A wide range of potential As-mitigation measures have been proposed over the years, ranging from in situ chemical and biological oxidative processes for immobilizing As to subsequent filtration methods and social awareness programs for the affected population. (5−7)