Use of observations below detection limit for model calibration.

Area of Science:

• Environmental Science
• Water Quality Analysis
• Geochemistry

Background:

• Nondetect values in water samples, representing concentrations below analytical detection limits, pose challenges in environmental modeling.
• Traditional methods of deleting or substituting these values can introduce significant bias in parameter estimation and model predictions.
• Accurate representation of low chemical concentrations is crucial for understanding environmental processes and risks.

Purpose of the Study:

• To develop and present a novel method for incorporating nondetect values into model calibration.
• To provide a more realistic and less biased approach to calibrating environmental models with censored data.
• To demonstrate the advantages of the proposed method, particularly for transport models dealing with low concentration data.

Main Methods:

• Proposing a censored-residual approach for handling nondetect observations during model calibration.
• Calculating residuals as the difference between the detection limit and simulated value when the simulated value exceeds the detection limit.
• Assigning a residual of zero when the simulated value is below the detection limit.

Main Results:

• The censored-residual approach offers a more realistic inclusion of nondetect values compared to deletion or substitution.
• This method reduces bias in estimated parameter values and model predictions.
• The approach is particularly beneficial for calibrating transport models using datasets with many low concentration measurements.

Conclusions:

• The censored-residual approach is a statistically sound and practical method for incorporating chemical nondetect values in model calibration.
• This technique enhances the accuracy of environmental models, especially those sensitive to low concentration data.
• Adoption of this method can lead to more reliable assessments of water quality and contaminant transport.