Investigating Rice-Farming Pesticide Concentrations in River Water Using a Basin-Scale Runoff Model with Uncertain Inputs

Abstract

In predicting time-series concentrations of pesticides in river water using diffuse pollution hydrological models, farming schedules (including pesticide application) and pesticide sorption/decomposition rates greatly affect prediction accuracy. For large, basin-scale catchments, precise acquisition of these data is not possible and substantial estimation uncertainty inevitably exists. This article presents the development of a basin-scale diffuse pollution hydrological model, a Monte Carlo method for creating input data, and its effectiveness in predicting the concentrations of paddy-farming pesticides in river water from a large catchment (1882 km2). The Monte Carlo method created input data for numerous compartments of a paddy field in the basin model: the pesticide products, amounts and dates of pesticide application, rice varieties, rice seedling transplanting dates, time variation of water depth in rice paddies, and parameter values for pesticide decomposition and sorption. The model was calibrated with hydrological data only, without reference to observed pesticide concentration data. Results showed that the uncertainty bounds estimated for model outputs with Monte Carlo inputs encompassed observed data and that the model predictions were better with Monte Carlo inputs than with deterministic input. The Monte Carlo method provides a surrogate approach for obtaining precise data on individual farming schedules (including pesticide application dates), degradation rates, and sorption coefficients in each soil.