RESUMO
BACKGROUND: As an option for higher-tier leaching assessment of pesticides in Europe according to FOCUS, pesticide properties can be estimated from lysimeter studies by inversely fitting parameter values (substance half-life DT50 and sorption coefficient to organic matter kom ). The aim of the study was to identify adequate methods for inverse modelling. RESULTS: Model parameters for the PEARL (Pesticide Emission Assessment at Regional and Local scales) model were estimated with different inverse optimisation algorithms - the Levenberg-Marquardt (LM), PD_MS2 (PEST Driver Multiple Starting Points 2) and SCEM (Shuffled Complex Evolution Metropolis) algorithms. Optimisation of crop factors and hydraulic properties was found to be an ill-posed problem, and all algorithms failed to identify reliable global minima for the hydrological parameters. All algorithms performed equally well in estimating pesticide sorption and degradation parameters. SCEM was in most cases the only algorithm that reliably calculated uncertainties. CONCLUSION: The most reliable approach for finding the best parameter set in the stepwise approach of optimising evapotranspiration, soil hydrology and pesticide parameters was to run only SCEM or a combined approach with more than one algorithm using the best fit of each step for further processing. PD_MS2 was well suited to a quick parameter search. The linear parameter uncertainty intervals estimated by LM and PD_MS2 were usually larger than by the non-linear method used by SCEM. With the suggested methods, parameter optimisation, together with reliable estimation of uncertainties, is possible also for relatively complex systems.
