No abstract

Simulations of pesticide fate in soils are often based on persistence models developed nearly 30 years ago. These models predict dissipation in the field on a daily basis by correcting laboratory degradation half‐lives for actual soil temperature and moisture content. They have been extensively applied, but to date no attempt has been made to evaluate existing studies in a consistent, quantitative way. This paper reviews 178 studies comparing pesticide soil residues measured in the field with those simulated by persistence models. The simulated percentage of initial pesticide concentration at the time of 50% measured loss was taken as a common criterion for model performance. The models showed an overall tendency to overestimate persistence. Simulated values ranged from 12 to 96% of initial pesticide concentrations with a median of 60%. Simulated soil residues overestimated the target value (50% of initial) by more than a factor of 1.25 in 44% of the cases. An underestimation by more than a factor of 1.25 was found in only 17% of the experiments. Discrepancies between simulated and observed data are attributed to difficulties in characterizing pesticide behavior under outdoor conditions using laboratory studies. These arise because of differences in soil conditions between the laboratory and the field and the spatial and temporal variability of degradation. Other possible causes include losses in the field by processes other than degradation, deviations of degradation from first‐order kinetics, discrepancies between simulated and actual soil temperature and moisture content, and the lack of soil‐specific degradation parameters. Implications for modeling of pesticide behavior within regulatory risk assessments are discussed.

No abstract

This study used field lysimeter leachate and pesticide concentration data within an inverse modeling framework to estimate pesticide degradation and sorption parameters. Experimental data comprising four pesticide applications during 3 yr were used to compare a local parameter estimation algorithm (Levenberg–Marquardt, LM) with a global algorithm (Shuffled Complex Evolution Metropolis, SCEM). Good model fits (only marginally better model fits using SCEM) with respect to both the observed leachate volumes and corresponding pesticide concentrations were obtained using both algorithms. Parameter optima found with LM and SCEM were very similar, thus suggesting that LM correctly located the global optimum for our experimental data. Equally as important as the optimal parameter values, however, are the estimated parameter uncertainties. This study revealed that LM (using a Jacobian‐based approach) provided too large parameter uncertainties. A logarithmic transformation of the parameter tended to decrease the uncertainty in most cases. The overestimation of parameter uncertainty by LM suggests that model sensitivity close to the optimal parameter set was relatively small and underestimated the sensitivity to large parameter changes. A multiobjective Pareto analysis was subsequently compared with a sequential single‐objective approach to reveal the capability of the multiobjective approach to verify model structure and model concept. Our results indicate that a multiobjective SCEM approach is recommended when the objective is to estimate pesticide degradation and sorption parameters and their uncertainty.

No abstract