Citation:

Sayre, R., J. Arnot, P. Fantke, K. Isaacs, M. Serre, AND J. Wambaugh. Concentration estimates for thousands of organic chemicals in surface water using Bayesian consensus modeling. Society of Environmental Toxicology and Chemistry (SETAC) Meeting, Virtual, OR, November 14 - 18, 2021.

Impact/Purpose:

Submission to the Society of Environmental Toxicology and Chemistry (SETAC) Meeting November 2021

Description:

Regulatory bodies need efficient tools to examine risk decision scenarios for surface water systems. Because monitoring data are often not available for all scenarios of interest, multimedia fugacity models based on physicochemical properties can help fill data gaps. When analyzing scenarios across the thousands of chemicals that could enter the environment in the United States, the most reliable models are predictive of empirically observed values, and the most relevant models can make predictions for many chemicals. This project presents a metamodel describing relationships between nation-scale surface water model predictions and a set of representative surface water concentrations for over 900 organic chemicals from hundreds of monitoring sites across the United States from 2008 to 2018. The models predict chemical mass distribution in the environment per unit emission, and therefore are dependent upon environmental loadings, which are here sourced from databases and models. A Bayesian multivariate linear regression evaluates correlations between thousands of combinations of model and loading weights and observed distributions of water quality data to create consensus probabilistic estimates of concentrations. The metamodel is set up to initially give each model a regression weight centered on zero; the weight range increases when observed data confirm the model predictions. Predictions on a validation set not included in the training data have lower uncertainty than estimates made using simpler methods, with the added benefit of being able to partially explain observed concentrations from widely available information. Once the uncertainty has been characterized, the metamodel can be used to calculate probable concentration ranges for hundreds of non-monitored chemicals within its domain of applicability. Finally, we test the metamodel’s utility by evaluating a simple risk scenario: the ratio of predicted no-effect concentrations in a sensitive freshwater species to the predicted concentrations. The views expressed in this abstract are those of the authors and do not necessarily represent the views or policies of the U.S. EPA.

Record Details: