Evaluating the probability of exceeding a site-specific soil cadmium contamination threshold

A non-parametric approach for assessing the probability that heavy metal concentrations in soil exceed a location-specific environmental threshold is presented. The methodology is illustrated for an airborne Cd-contaminated area in Belgium. Non-stationary simple indicator kriging, using a soft indicator coding to account for analytical uncertainty, was used in combination with declustering weights to construct the local conditional cumulative distribution function (ccdf) of Cd. The regulatory Cd contamination threshold (CT) depends on soil organic matter and clay content, which entails that its value is not constant across the study area and also is uncertain. Therefore, soft indicator kriging was used to construct the ccdfs of organic matter and clay. Latin hypercube sampling of the ccdfs of Cd, soil organic matter and clay yielded a map of the probability that Cd concentrations exceed the site-specific CT. Cross-validation showed that the ccdfs provide accurate models of the uncertainty about these variables. At a probability level of 80% we found that the CT was exceeded at 27.3% of the interpolated locations, covering 3192 ha of the study area, illustrating the extent of the pollution. Additionally, a new methodology is proposed to sample preferentially the locations where the uncertainty about the probability of exceeding the CT, instead of the uncertainty about the pollutant itself, is at a maximum. This methodology was applied in a two-stage sampling campaign to identify locations where additional Cd samples should be collected in order to improve the classification into safe and contaminated locations.