WATER SECURITY IN SOUTH ASIA: THE POTENTIAL ROLE OF ARTIFICIAL INTELLIGENCE IN SUPPORTING THE SELECTION OF REMEDIATION APPROACHES FOR GROUNDWATER ARSENIC (Invited Presentation)

Access to safe, sustainable drinking water supplies is fundamental to economic development and public health. Groundwater arsenic is a major public health challenge disproportionately impacting populations in South Asia, and appropriate selection and management of optimal remediation strategies remains challenging, particularly in decentralized areas. Artificial intelligence including machine learning approaches have recently been used for geospatial modelling of groundwater contaminant distribution in India [1, 2]. Here, we explore the potential application of machine learning to inform the selection of optimal remediation strategies based on geochemical aspects associated with source water chemistry. Using random forest models based on secondary data of groundwater composition across Bangladesh [3], we investigate the geospatial distribution of groundwater arsenic, iron, phosphorus and the ratio of ([Fe] – 1.8[P])/[As]. The ratio ([Fe] – 1.8[P])/[As] was observed to have a statistically significant association with the arsenic removal efficiency of small-scale remediation units in Bihar [4], consistent with source water-dependent competition of arsenic and phosphate for sorption sites particular in absorption-based remediation processes [5]. This would enable first generation mapping of areas which, for example, are more geochemically favourable to higher remediation efficiency, or inform where additional iron may facilitate increased efficiency, particularly for adsorption-based technologies. Limitations relate to the recommendation for site-specific testing, local heterogeneity, other groundwater matrix influences, and factors beyond geochemistry which may impact remediation efficiency. Notwithstanding, our exploratory investigation of machine learning to inform groundwater remediation selection in Bangladesh offers substantial opportunity for further development as a groundwater remediation decision support tool.

Acknowledgements: We acknowledge a UoM-KTH-SU seedcorn award, a Dame Kathleen Ollerenshaw Fellowship & the Indo-UK project FAR-GANGA (NE/R003386/1 & DST/TM/INDO-UK/2K17/55(C) & 55(G); www.farganga.org).

References: [1] Podgorski et al IJERPH 2020; [2] Wu et al EGH 2021; [3] BGS & DPHE 2001; [4] Richards et al STOTEN 2022; [5] Hug et al EST 2008.