Precision Illusion and Reliable Decision-making: A Comparative Study of Stability between Descriptive Statistics and Geostatistical Interpolation in Soil Pollution Investigation Data

Accurate assessment of soil contamination forms the scientific basis for risk management and remediation strategies at industrial sites. Spatial interpolation techniques, particularly kriging, are widely used today as they effectively visualize contamination distributions. These methods exhibit limited stability owing to their strong dependence on sampling strategies. Based on two independent sampling campaigns conducted on the same site (with 229 and 116 sampling points, respectively), this study systematically compared the stability differences between statistical analysis and spatial characterization results for arsenic contamination. The results demonstrate that key statistical parameters, particularly the 95% upper confidence limit (95% UCL) of the mean, the median, and the 95th percentile, exhibited high stability with relative deviations generally below 15%. The 95% UCL of the mean demonstrated remarkable stability, with a relative deviation of less than 5%, confirming its strong suitability for risk assessment applications. In contrast, the results obtained through Ordinary Kriging interpolation displayed poor stability. The delineation of contamination boundaries, identification of hotspot distributions, and quantification of risk-exceeding zones displayed marked inconsistencies compared to the established control threshold. This instability was particularly pronounced in contexts characterized by weak spatial autocorrelation, as indicated by nugget-to-sill ratios close to 1, where the interpolation results reflected the spatial distribution of sampling points rather than the true contamination pattern. The study warns that relying solely on spatial interpolation risks flawed remediation decisions, causing either excessive or inadequate remediation efforts. The study advocated for an integrated assessment framework that prioritizes rigorous statistical parameters as its foundation, complemented by spatial trend visualization and systematic uncertainty quantification. These findings can strengthen contamination assessment reliability and decision-making rigor, guiding next-generation site investigation technologies.