Development and validation of a simulation method, PeCHREM, for evaluating spatio-temporal concentration changes of paddy herbicides in rivers

Abstract

In pesticide risk management in Japan, predicted environmental concentrations are estimated by a tiered approach, and the Ministry of the Environment also performs field surveys to confirm the maximum concentrations of pesticides with risk concerns. To contribute to more efficient and effective field surveys, we developed the Pesticide Chemicals High Resolution Estimation Method (PeCHREM) for estimating spatially and temporally variable emissions of various paddy herbicides from paddy fields to the environment. We used PeCHREM and the G-CIEMS multimedia environmental fate model to predict day-to-day environmental concentration changes of 25 herbicides throughout Japan. To validate the PeCHREM/G-CIEMS model, we also conducted a field survey, in which river waters were sampled at least once every two weeks at seven sites in six prefectures from April to July 2009. In 20 of 139 sampling site-herbicide combinations in which herbicides were detected in at least three samples, all observed concentrations differed from the corresponding prediction by less than one order of magnitude. We also compared peak concentrations and the dates on which the concentrations reached peak values (peak dates) between predictions and observations. The peak concentration differences between predictions and observations were less than one order of magnitude in 66% of the 166 sampling site-herbicide combinations in which herbicide was detected in river water. The observed and predicted peak dates differed by less than two weeks in 79% of these 166 combinations. These results confirm that the PeCHREM/G-CIEMS model can improve the efficiency and effectiveness of surveys by predicting the peak concentrations and peak dates of various herbicides. Environmental signicancePesticides pose potential risks to aquatic ecosystems. Multiple pesticides are oen used simultaneously in agricultural elds over a relatively short period, and such use leads to spatially and temporally skewed distributions of pesticide concentrations in rivers. Methods capable of evaluating dynamic changes in the spatial distributions of multiple pesticides in the environment are needed for pesticide risk management. We developed a simulation method for evaluating daily concentration changes at a resolution of several kilometers, and then, focusing on paddy herbicides, we used a multimedia environmental fate model to calculate concentrations of 25 herbicides in rivers throughout Japan. Finally, we validated the results with monitoring data. This work will contribute to pesticide risk assessment efforts by making it possible to predict both peak pesticide concentrations and the dates on which the concentrations reach their peak values.

Validation of AIST‐SHANEL Model Based on Spatiotemporally Extensive Monitoring Data of Linear Alkylbenzene Sulfonate in Japan: Toward a Better Strategy on Deriving Predicted Environmental Concentrations

“…Although several quantitative evaluations of exposure models have been conducted, no uniform approach exists. Using a set of metrics similar to that in the present study, Imaizumi et al (2018) compared spatiotemporal data for 25 herbicides at 7 sites versus a few to 10s of sites using Pesticide Chemicals High Resolution Estimation Method (PeCHREM) or G-CIEMS, and reported good agreement (RMSLE: 0.57-9.03). In the present study, although we focused on only 1 chemical (LAS), AIST-SHANEL was capable of estimating chemical concentrations over a wider range of river water bodies at a high resolution with similar accuracy to PeCHREM/G-CIEMS, even when comparing more than 10 000 points of large-scale environmental monitoring data (RMSLE: 0.90).…”

“…The RMSLE and 10× factor values were calculated to enable relative comparisons of the estimation accuracy between models, where smaller values, as a result of changes in model conditions, were indicative of higher estimation accuracies. The RMSLE was calculated as described previously (Imaizumi et al ): where P is the predicted concentration (µg/L), O is the observed concentration (µg/L), and n is the number of pairs of predicted and observed concentrations.…”

“…The RMSLE and 10× factor values were calculated to enable relative comparisons of the estimation accuracy between models, where smaller values, as a result of changes in model conditions, were indicative of higher estimation accuracies. The RMSLE was calculated as described previously (Imaizumi et al 2018):…”