Soil contamination sampling intensity: determining accuracy and confidence using a Monte Carlo simulation

Abstract

Soil pollution is an extensive global problem, and effective management depends on accurate characterization and mapping of the extent of soil contamination. The objectives of this study were to determine the accuracy of different sampling intensities and the optimal number of samples required to minimize remediation project costs. To determine the accuracy associated with different sampling intensities, a Monte Carlo simulation was conducted. A simulated contaminant plume was created based on inverse distance weighting, kriging, or multivariate adaptive regression splines. Different sampling intensities, with grid spacing ranging from approximately 10% to 50% of the site extent, were used to generate a plume map using random forest model with a Euclidean distance matrix as predictors. The relative error was then determined as part of a Monte Carlo simulation that ran 10 000 simulations for each grid intensity for a total of 90 000 simulations. The optimal number of samples was determined based on economic factors, and the error functions generated with the Monte Carlo simulations. Average error ranged from 57% for 25 data points to 5% for 2800 data points. The 90th percentile error ranged from 100% to 0.3% for the sample data point range. Based on these results, the optimal number of samples, depending on pricing, ranged from 31 samples for a 10 m3 contaminant plume to 3475 samples for a 10 000 m3 soil contaminant plume.