High-resolution heavy metal contamination mapping (HRMMs) helps to accurately identify areas requiring risk management or remediation. Traditional HRMMs based on grid model soil sampling, chemical analysis and geostatistical interpolation methods to draw pollution distribution maps were costly, slow, and not suitable for highly heterogeneous pollution sites. This study proposes a new method to improve PXRF (Portable X-ray fluorescence analysis) data through multiple nonlinear regression, use the improved PXRF data for collaborative spatial interpolation, draw HRMMs maps and predict the distribution of heavy metal pollution. In order to support the establishment and validation of the model, an Mn and Zn contaminated site in northwest China was selected for research. The results show that: (1) The introduction of PXRF data as an auxiliary variable of Co-Kriging interpolation can effectively improve the interpolation accuracy, while the corrected PXRF data can further improve the spatial characterization accuracy. The average errors of the corrected PXRF Co-Kriging interpolation for heavy metals Mn and Zn were 4.5% and 78.2% lower than those of the original PXRF Co-Kriging interpolation. (2) The change in the point density of the primary variable will change the accuracy of the PXRF Co-Kriging interpolation after correction. Taking Zn as an example, when the main variable point density was 4 holes/(104
), the corrected PXRF Co-Kriging interpolation accuracy was significantly reduced. (3) Increasing the point density of auxiliary variables can significantly improve the accuracy of Co-Kriging interpolation. When the point density of the auxiliary variable was increased to 7 holes/(104
), the mean error and root mean square error of PXRF Co-Kriging interpolation after Zn correction were reduced by 92.4% and 34.7% respectively. The research shows that the correction of PXRF data can effectively improve the accuracy of pollutant Co-Kriging interpolation. At the same time, Co-Kriging interpolation needs to meet a certain amount of requirements for the point density of primary variables, and the higher the point density of auxiliary variables, the higher the accuracy of Co-Kriging interpolation.