Modelling cropping system effects on crop pest dynamics: How to compromise between process analysis and decision aid

Managing crop pests (weeds, insects, pathogens, etc.) to limit both crop production loss and environmental impacts is a major challenge of agriculture. Because of the large number of factors and the complexity of interactions, models are invaluable tools to synthesize our knowledge on pests and to quantify the effects of cropping systems on pest dynamics. These models must be able to rank candidate cropping systems as a function of pest frequency and severity, and to account for variability in effects to estimate the risk of success or failure of a particular system. Two contrasting approaches are possible. Mechanistic models describe variability with process-based, usually deterministic sub-models quantifying interactions between cropping system components and environmental conditions. Empirical models directly relate observations to input variables, using few parameters, and usually quantify variability with probabilistic (stochastic) functions. The present paper critically evaluates these a priori contradictory approaches, i.e. deterministic vs. stochastic and mechanistic vs. empirical representations of cropping system effects in pest dynamics models, relative to model objectives and scales, pest species, scientific disciplines and knowledge level. We do not attempt to be exhaustive but analyse a small number of contrasting models to identify their advantages, disadvantages and complementarities. The paper concludes that models using a mechanistic representation of the cropping system × environment interactions are best for quantifying effects and accounting for their variability, combined with a subsequent transformation with in silico experiments into empirical models of the major cropping system components.