High–accuracy spatial mapping of soil heavy metal(loid)s (HMs) is essential for targeted environmental management. However, intensive human disturbances complicate the spatial variation structures of soil HMs and raise the investigation costs. To tackle these challenges, we propose a three-stage framework for spatially predicting soil HMs. First, a robust geographically weighted regression model integrating land-use types (RGWR-LU) was developed to spatially calibrate the in-situ field portable X-ray fluorescence (FPXRF) multi-HM dataset in an intensively human-impacted area of China. Second, a robust spatial receptor model integrating the RGWR–LU–calibrated in–situ FPXRF multi-HM dataset (RAPCS/RGWR–FPXRF) was established to derive the dominant source–contribution covariate and the refined land‑use effects free from interferences of other sources. Third, a robust residual cokriging framework integrating the calibrated FPXRF data, the dominant source–contribution covariate, and the refined land–use effects (RRCoK–FPXRF·SC·LU) was proposed for spatially predicting soil HMs. Key results demonstrate that: (1) RGWR–LU achieved higher spatial calibration accuracy for in–situ FPXRF multi-HM dataset ( RI > 72.08%) than RGWR, GWR, and the traditional ordinary least squares regression; (2) RAPCS/RGWR–FPXRF provided higher source–apportionment accuracy than the basic robust spatial receptor model (RAPCS/RGWR) and the traditional receptor model (APCS/MLR), yielding more effective source–contribution covariate and refined land–use effects for spatial prediction; and (3) RRCoK–FPXRF·SC·LU exhibited the highest spatial prediction accuracy among the seven evaluated models (e.g., for Cu, RMSE decreased from 7.46 mg kg⁻ 1 for ordinary kriging to 2.03 mg kg⁻ 1 for RRCoK‑FPXRF·SC·LU, with RI = 72.79%). This study presents a cost-effective, high-accuracy methodology for spatially predicting soil HMs in intensively human-impacted areas.
Qu et al. (Wed,) studied this question.