Unfolding the spatial heterogeneity of the natural background level of arsenic in groundwater at the meso-scale

Geogenic arsenic (As) contamination in groundwater represents a significant environmental and public health concern worldwide. Modelling its spatial distribution is crucial for effective management and protection of water resources. However, it remains a methodological challenge, particularly at the meso-scale, between the more commonly investigated regional and local scales. This doctoral research presents a comprehensive characterization of geogenic arsenic in the shallow aquifer of the Ferrara province (eastern Po Plain, Italy), a hydrogeologically complex and anthropized setting. The overarching goal was to develop a reproducible methodological framework for Natural Background Level (NBL) estimation and spatial modelling at the meso-scale, enabling advancements in groundwater quality assessment. To address groundwater data scarcity, the research introduces a novel approach that repurposes hydrochemical data from sites under remediation, aggregating heterogeneous datasets and systematically excluding anthropogenic influences. Multivariate and machine learning (ML) analyses were then employed to delineate hydrochemical zones and identify the main factors controlling As mobilization. Results revealed widespread reducing conditions in the shallow aquifer of the province, particularly pronounced toward the Po Delta, likely related to stratigraphic architecture. Redox conditions and sedimentary features emerged as the dominant controls shaping As distribution. Eventually, multiple spatial modelling approaches, both geostatistical- and ML-based, were implemented. The resulting continuous maps depict (i) the probability of exceeding the World Health Organization threshold (10 µg/L) and (ii) NBL concentrations. Comparative analyses demonstrated complementary strengths, with ML enhancing predictive accuracy and geostatistics ensuring spatial coherence. Overall, this research contributes to bridging a methodological gap between site-specific and regional groundwater studies. It leverages existing environmental datasets to enable cost-efficient investigations and NBL estimation. The multi-method spatial modelling framework provides a robust methodology with practical implications. The approach is potentially transferable to other regions, offering scientifically grounded tools to support groundwater management and policy-making under conditions of natural heterogeneity and data scarcity.