Abstract:
Complex deterministic models are being used within the context of pesticide
registration to assess the potential for crop protection products to impact on the
environment. Although calibration is in many ways at the heart of pesticide fate
modelling, it has received little attention in the past. Sensitivity analyses were
carried out for the four main leaching models used for pesticide registration in
Europe (PELMO, PRZM, PESTLA and MACRO) using four different leaching
scenarios and two approaches to sensitivity assessment (one-at-a-time and Monte
Carlo sensitivity analyses). Also, an inverse modelling approach was used to
estimate values for sorption and degradation parameters from leaching data for seven
lysimeters using the PESTRAS model.
The overall conclusions of the PhD can be summarised as follows:
1. Sensitivity analyses for the four leaching models mainly used for pesticide
registration in Europe demonstrated that predictions for pesticide loss are most
sensitive to parameters related to sorption and degradation. In a small number of
scenarios, hydrological parameters were found to also have a large influence on
predictions for pesticide loss.
2. Sensitivity analysis proved to be an effective approach not only for ranking
parameters according to their influence on model predictions, but also for
investigating model behaviour in a more general context. However, the research
questioned the robustness of the Monte Carlo approach to sensitivity analysis as
issues of replicability were uncovered.
3. Inverse modelling exercises demonstrated that non-uniqueness is likely to be
widespread in the calibration of pesticide leaching models. Correlation between
parameters within the modelling, such as that between sorption and degradation
parameters when predicting pesticide leaching, may prevent the robust derivation
of values through an inverse modelling approach. Depending on the calibration
system considered, these parameters may act as fitting variables and integrate
inaccuracies, uncertainties and limitations associated with experimental data,
modelling and calibration.
4. A special implementation of error surface analysis termed lattice modelling was
proposed in the PhD as an efficient technique to i) assess the likely extent of nonuniqueness issues in the calibration of pesticide leaching models; and, ii) replace
traditional parameter estimation procedures where non-uniqueness is expected.
Care should be exercised when assessing the results obtained by both modelling and
inverse modelling studies. Suggestions to improve the reliability in the calibration of
pesticide leaching models have been proposed.