Novel statistical modelling approaches for pesticide residues.

Abstract

Plant protection products play an important role in protecting our food supply against pests, diseases and weeds. As global food demand rises, their role in maintaining the quality and quantity of our food production is likely to increase in the absence of other control methods. To manage the risks associated with pesticide usage, EU laws regulate the placing of plant protection products on the market and the monitoring of pesticide residues in food. This involves assessing the potential risks associated with human dietary exposure by conducting dietary risk assessments which take both consumption patterns and residue levels of pesticides in and on food items into account. Residue levels will vary from one food item to the next so we need to know what the distribution of residues over food items is in order to assess how high residue levels can be.

In this thesis we introduce novel statistical approaches that can be used to obtain better estimates of the variation and uncertainty in pesticide residue levels on raw agricultural products. The rst approach uses monitoring data and pesticide usage information to model the correlation in pesticide residue levels when multiple pesticides have been used. Next we introduce an approach that can be used to describe the variation in log-residue levels in units, assuming that multiple data sets share a common shape. The nal model describes both within-eld and between-field variation of residue levels. These new approaches, which provide promising alternatives to existing methods, can be implemented in existing dietary risk assessment software and will expand the suite of models available to risk assessors when assessing dietary exposure to pesticides.