Application of mixed effects models for characterizing contaminated sites

Area of Science:

• Environmental Science
• Geostatistics
• Statistical Modeling

Background:

• Site characterization often involves nested data within boreholes, leading to dependencies ignored by classical models.
• Left-censored data below detection limits further complicate accurate statistical analysis in environmental studies.

Purpose of the Study:

• To present a mixed effects model for analyzing contaminated site data, addressing within-borehole dependencies and left-censored observations.
• To demonstrate the application of this methodology in a brownfield site characterization in Montreal, Canada.

Main Methods:

• Utilizing a mixed effects model to simultaneously account for within-borehole correlation and left-censored concentration data.
• Analyzing concentration data from a real-world brownfield site to validate the model's performance.

Main Results:

• The mixed effects model accurately captures within-borehole data dependency and handles censored values effectively.
• Underestimating correlation by ignoring censored data leads to biased estimates.
• The model provides insights into optimal sampling strategies, including borehole and sample size determination.

Conclusions:

• Mixed effects models offer a superior approach to classical methods for contaminated site characterization, especially with nested data and censoring.
• Accurate analysis of contamination extent and compliance decisions are improved by accounting for data dependencies and censored values.