A meta-analysis of published semivariograms to determine sample size requirements for assessment of heavy metal concentrations at contaminated sites

Abstract

Soil contamination poses substantial risks to human and ecosystem health, justifying the need for accurate delineation and remediation of contaminated sites. The number of soil samples collected at a site during assessment is limited by cost and time available for assessment, increasing the potential for misclassification due to insufficient samples. Using distributions of heavy metals sourced from semivariograms provided in published studies, the first stage of this study sought to determine how many samples were required for the confidence interval around the mean to be above or below the Australian guideline value for each specific metal and study. Estimated sample size for assessing mean contamination across a site ranged from two to four samples however, some distributions possessed a higher amount of variation and therefore required more samples. The second stage of the investigation explored sample size requirements for mapping contaminated sites. Unconditional Gaussian simulations created from published semivariograms were sampled using 15 different sample sizes, and the samples used to obtain predictions of the simulated distributions. For each sample, observed (simulated) and predicted (kriged) metal concentrations were classed as being below or exceeding the guideline values and compared through quantification of the number of misclassifications that occurred. When mapping a site of 5 km2 or less, uncertainty and misclassification decreased with increasing sample size, stabilising at around 200 samples however, the lowest uncertainty occurred at around 500 samples. The study acknowledges this may be unrealistic and economically inefficient, so in addition to these findings it is worth exploring improvement in other areas of investigation, such as in the detection and mapping stages.