Application of mixed effects models for characterizing contaminated sites

Abstract

In a typical data collection process for the purpose of characterizing contaminated sites, boreholes are usually drilled in different locations based on a sampling plan; and consequently, multiple samples are collected from each borehole. As a result, it is quite plausible that a certain degree of dependency or similarity exists among observations nested within a borehole. However, when classical regression models are employed, such dependencies are often ignored, resulting in biased estimates. In site characterization studies, further complication arises due to the presence of left-censored observations, those falling below the detection limit of measuring instruments. To overcome the above issues, this paper employs a mixed effects model that allows accounting for the within-borehole data dependency while accommodating left-censored concentrations. The benefits of the adopted methodology are explored by analyzing concentration data obtained from characterization study of a brownfield site located in Montreal, Canada. This paper illustrates that the estimated within-borehole correlation can be used to determine the optimal number of boreholes as well as the sample size to be collected from each borehole. Such correlation is underestimated when censored values are not accommodated in the model but substituted with a constant prior to data analysis. In addition, the adopted methodology provides an accurate insight into the vertical extent of contamination that can result in different compliance decisions when compared with classical approach.