Geogenic arsenic (As) contamination poses a major environmental concern, yet mapping its Natural Background Levels (NBLs) remains challenging in data-scarce, human-impacted regions within heterogeneous sedimentary aquifers. NBLs are essential for distinguishing natural from anthropogenic contributions and for establishing threshold values in environmental policy and groundwater management. This study applies a Random Forest (RF) algorithm to predict arsenic NBL exceedance of the WHO guideline (10 μg/L) in shallow groundwater of the Ferrara province (Po Plain, Italy). Three model configurations were tested using: (i) in-situ hydrochemical variables (hydrochemical model), (ii) spatially continuous variables (geospatial model), and (iii) a combination of both (mixed model). The mixed model achieved the highest performance (AUC = 0.88), followed by the hydrochemical model (AUC = 0.86). The geospatial model performed slightly lower (AUC = 0.83) but enabled the generation of a continuous 50 m-resolution probability map, identifying 17% of the area as potentially exceeding 10 μg/L. Iron, redox potential and dissolved oxygen were the dominant hydrochemical predictors, while organic carbon, clay content, and surficial geology were the dominant spatial variables, reflecting geochemical and sedimentary controls on As mobilization through reductive dissolution processes. Elevated exceedance probabilities are identified in the northeastern sector, proximal to the Po River Delta, and in peat-rich central areas. This study demonstrates that spatially continuous variables can effectively complement hydrochemical data for continuous NBL mapping, capturing meso-scale heterogeneity that single, groundwater body-scale values conventionally adopted in European NBL frameworks are unable to resolve, thus offering a useful tool for contamination assessment and groundwater management.
Landi, L., Podgorski, J., Berg, M., Rotiroti, M., Zanotti, C., Amorosi, A., et al. (2027). Machine learning prediction of arsenic natural background level heterogeneity in shallow groundwater: Integrating in-situ hydrochemistry and spatial proxies at the meso-scale. JOURNAL OF CONTAMINANT HYDROLOGY, 284 [10.1016/j.jconhyd.2026.105154].
Machine learning prediction of arsenic natural background level heterogeneity in shallow groundwater: Integrating in-situ hydrochemistry and spatial proxies at the meso-scale
Rotiroti, Marco;Zanotti, Chiara;
2027
Abstract
Geogenic arsenic (As) contamination poses a major environmental concern, yet mapping its Natural Background Levels (NBLs) remains challenging in data-scarce, human-impacted regions within heterogeneous sedimentary aquifers. NBLs are essential for distinguishing natural from anthropogenic contributions and for establishing threshold values in environmental policy and groundwater management. This study applies a Random Forest (RF) algorithm to predict arsenic NBL exceedance of the WHO guideline (10 μg/L) in shallow groundwater of the Ferrara province (Po Plain, Italy). Three model configurations were tested using: (i) in-situ hydrochemical variables (hydrochemical model), (ii) spatially continuous variables (geospatial model), and (iii) a combination of both (mixed model). The mixed model achieved the highest performance (AUC = 0.88), followed by the hydrochemical model (AUC = 0.86). The geospatial model performed slightly lower (AUC = 0.83) but enabled the generation of a continuous 50 m-resolution probability map, identifying 17% of the area as potentially exceeding 10 μg/L. Iron, redox potential and dissolved oxygen were the dominant hydrochemical predictors, while organic carbon, clay content, and surficial geology were the dominant spatial variables, reflecting geochemical and sedimentary controls on As mobilization through reductive dissolution processes. Elevated exceedance probabilities are identified in the northeastern sector, proximal to the Po River Delta, and in peat-rich central areas. This study demonstrates that spatially continuous variables can effectively complement hydrochemical data for continuous NBL mapping, capturing meso-scale heterogeneity that single, groundwater body-scale values conventionally adopted in European NBL frameworks are unable to resolve, thus offering a useful tool for contamination assessment and groundwater management.
I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
