A Markov Chain Monte Carlo technique for parameter estimation and inference in pesticide fate and transport modeling

Abstract

A Bayesian method involving Markov Chain Monte Carlo (MCMC) technique was implemented into a pesticide fate and transport model to estimate the best input parameter ranges while considering uncertainties included in both the observed pesticide concentrations and in the model.The methodology used for integrating the MCMC technique into a pollutant fate and transport models was detailed. The uncertainties encompassed in the dissolution rate and in the adsorption coefficient of the herbicide mefenacet were greatly reduced by the MCMC simulations. In addition, an optimal set of input parameters extracted from the MCMC simulations accurately reproduced mefenacet concentrations in paddy water and paddy soil as compared to the original published dataset. Consequently, by simultaneously optimizing multiple parameters of environmental models and conducting uncertainty analysis, MCMC technique exhibits powerful capability for improving the reliability and accuracy of computer models.The main strengths of the MCMC methodology are: (1) the consideration of uncertainties from both input parameters and observations and (2) the prior distributions of the input parameters which can be reformulate when additional knowledge is available.