Additional sampling using in-situ portable X-ray fluorescence (PXRF) for rapid and high-precision investigation of soil heavy metals at a regional scale

Traditional soil heavy metal (HM) investigation usually costs a lot of human and material resources. In-situ portable X-ray fluorescence spectrometry (PXRF) is a cheap and rapid HM analysis method, but its analysis accuracy is usually affected by spatially non-stationary field environment factors. In this study, residual sequential Gaussian co-simulation (RCoSGS) was first proposed to incorporate both continuous and categorical auxiliary variables for spatial simulation of soil Cu. Next, additional in-situ PXRF sampling sites (n = 300) were allocated in the subareas with high, medium, and low conditional variances in the proportions of 50%, 33.33%, and 16.67%, respectively. Then, robust geographically weighted regression (RGWR) was established to correct the spatially non-stationary effects of field environmental factors on in-situ PXRF and further compared with the traditionally-used multiple linear regression (MLR) and basic GWR in correction accuracy. Finally, RCoSGS with the RGWR-corrected in-situ PXRF as part of hard data (RCoSGS-PXRF) was established and further compared with the model with one or multiple auxiliary variables in the spatial simulation accuracy. Results showed that (i) RCoSGS effectively incorporated both SOM and land-use types and obtained higher spatial simulation accuracy (RI = 37.52%) than residual sequential Gaussian simulation with land-use types (RI = 19.44%) and sequential Gaussian co-simulation with SOM (RI = 20.92%); (ii) RGWR significantly weakened the spatially non-stationary effects of field environmental factors on in-situ PXRF, and RGWR (RI = 58.96%) and GWR (RI = 39.61%) obtained higher correction accuracy than MLR; (iii) the RGWR-corrected in-situ PXRF (RI = 66.57%) brought higher spatial simulation accuracy than both land-use types and SOM (RI = 37.52%); (iv) RCoSGS-PXRF obtained the highest spatial simulation accuracies (RI = 83.74%). Therefore, the proposed method is cost-effective for the rapid and high-precision investigation of soil HMs at a regional scale.