<p>Chronic dietary exposure to arsenic (As) represents a major public health concern, particularly in regions reliant on contaminated groundwater for irrigation. This study assessed the spatial variability of As across the aquifer–soil–plant continuum in arsenic-affected blocks of Nadia district, West Bengal, India. A total of 510 soil samples, 102 irrigation water samples, and 51 rice plant samples were collected and analyzed. Spatial autocorrelation using Moran’s I indicated a largely random distribution of As in both soil (–0.19 to 0.0007) and irrigation water (–0.27 to 0.08). Ordinary kriging (OK) was applied with cross-validation metrics—mean prediction error (MPE), root mean square error (RMSE), mean standardized error (MSE), and root mean square standardized error (RMSSE)—to generate spatial prediction models. The best-fit semi-variogram models were Gaussian for Chakdah and Haringhata, and Circular for Ranaghat-II and Santipur, with RMSE values ranging from 2.38 × 10^–4 to 1.16 × 10⁻¹ and prediction accuracies of 88–92%. The nugget-to-sill ratio for soil and groundwater As ranged from 0–4.8% and 0–9.16%, respectively, indicating strong spatial dependency likely influenced by pedogenic properties. These interpolated maps were used to estimate soil-available As (Av_As) and irrigation water As (As_irrigation). Health risk indicators, including average daily intake (ADI), hazard quotient (HQ), and target cancer risk (TCR), were calculated based on rice consumption. HQ values ranged from 2.87 to 4.49 (&gt; safe limit of 1), and TCR values ranged from 1.29 × 10⁻³ to 2.02 × 10⁻³, far exceeding the acceptable threshold (10⁻⁴), indicating serious carcinogenic risk in the study area.</p>