A highly refined quantitative estimate of acute human health risk from pesticides requires water monitoring data to define the annual maximum daily concentration (AMDC) across the landscape and over time. However, water monitoring data is limited for many pesticides. We developed, evaluated and demonstrated a novel approach that integrates seasonal wave with streamflow adjustment with extended capability (SEAWAVE-QEX) modeling and machine learning modeling to predict the AMDCs in surface water at sub-regional scale across the conterminous U.S. SEAWAVE-QEX model was employed to generate AMDC for each monitoring site as the target variable using all the monitoring data that met the SEAWAVE-QEX modeling criteria from 2016 to 2020. Feature variables, including pesticide use, soil characteristics, precipitation, drainage area, and chemical properties were compiled into the dataset to aid machine learning. Then a hybrid machine learning model was developed and evaluated based on the combination of eXtreme Gradient Boosting Regressor (XGBoost) and Random Forest Regressor (RF). The hybrid model predicted sub-regional level AMDCs of four example pesticides throughout the conterminous U.S. during 2016-2019. The predictive capability was demonstrated when compared with independent monitoring datasets. Finally, the predictions were used in aggregate acute exposure estimates that include food and water exposure for the general U.S. population. The estimated acute daily exposures showed reasonable agreement with the daily maximum exposures derived from the 95<sup>th</sup> percentile values of spot urine concentrations measured in biomonitoring surveys. Overall, the fully evaluated modeling approach proved to be a valuable solution of conducting geographically oriented, highly refined acute exposure assessments for pesticides, and can be easily extended for chronic exposure durations of concern . It can also guide water monitoring designs and support spatial epidemiological studies focusing on examination of geographic variations in disease with respect to environmental exposure.

Integrated approach of regression and machine learning modeling to refine human health risk from pesticides in surface water throughout the conterminous United States

Fall 2022 abstracts, posters, and presentations will launch in mid-September.

ACS Fall 2022

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.