DEEP LEARNING APPROACH FOR UNDERSTANDING CO-OCCURRENCE AND SPATIAL VARIABILITY OF ARSENIC AND FLUORIDE IN PRECAMBRIAN BASEMENT AQUIFERS OF NORTH-WEST TANZANIA

Occurrence in elevated concentrations and lack of adequated data on potentially toxic elements (PTE) in groundwater are among challenges limiting access to safe drinking water and geochemical studies. For economic reasons and accessibility issues, most aquifers are less sampled, particularly in developing countries. Thus, many geochemical studies rely on the existing groundwater abstraction points, which usually follow the population distribution. For that matter, the detected PTEs exhibit a large scale of variation, which complicates the understanding of spatial processes controlling their spatial variability. In this study, we developed and used the deep learning (DL) approach to characterize the co-occurrence and spatial variability of arsenic (As) and fluoride (F) in the Precambrian basement aquifers of north-west Tanzania. The study was based on 130 groundwater samples from drinking water sources in 13 wards of the Geita district, south of Lake Victoria in the vicinity of small and large scale gold mining sites. Using data collected, we developed a spatial database comprising groundwater quality parameters and used robust exploratory spatial data analysis (ESDA) methods to study the spatial variability in As and F concentrations. Arsenic concentrations ranged between 0 and 290 ppb while that of F ranged between 0.09 and 2.17 mg/L. For both chemical elements, concentrations were clustering in space as indicated by positive and significant Moran’s I. As concentrations demonstrated high spatial variability in highlands whereas F demonstrated the highest spatial variability in the lowlands. In the highlands, groundwater was acidic with mean pH below 6.5 suggesting that As is mobilized through the oxidative dissolution of As-bearing minerals especially, pyrites, arsenopyrites, and chalcopyrite in rocks exposed to air by ongoing mining activities. The pH in the lowlands was slightly below 7.0 complicating the interpretation of the sources of elevated F in drinking water sources. Since the lowlands are dominated by human settlements with people engaging themselves in agriculture, anthropogenic sources could be contributing to the contamination of drinking water sources by fluoride.