Sensitivity analyses for a model simulating demography and genotype evolutions with time: Application to GeneSys modelling gene flow between rape seed varieties and volunteers

Abstract Sensitivity of models to input variables and parameters should be analysed before evaluating and using them in order to identify (a) the minimum input data required for satisfactory simulations, (b) the major cultivation techniques to adapt when designing new cropping systems and (c) the main parameters to be estimated with special care. Conventional methods of sensitivity analysis being inadequate, a new method based on Monte Carlo simulations was developed in the present work. The analysed model was G ene S ys which quantifies the effects of cropping system on demography and genotype of cropped and volunteer rape plants. The studied output variables were the proportion of harvested rape seeds polluted by unwanted genes and the density and herbicide sensitivity of rape volunteers in winter cereals. Sensitivity of the model to the initial seed bank was first tested by simulating rape evolution from initial seed banks resulting from various seed density, genotype, age and distribution. Then, the effect of cropping systems was studied by simulating rape evolution for 100 000 randomly chosen combinations of crop succession and cultivation techniques. The effect of the model parameters was finally analysed by simulating rape evolution in six contrasted cropping systems and 100 000 randomly chosen parameter values. At all steps, linear models were used to test the significance of seed bank, cropping system variables or parameters. If the number of variables was too large for directly applying a linear model, a segmentation tree was used first to eliminate negligible variables. The study showed that the seed bank influenced simulated output during 5–6 years, which is therefore the minimum duration for simulations with an unknown initial seed bank. A survey list of cropping system variables to determine was established for comparing simulated and observed output. Major variables such as crop succession must be known for at least 3 years preceding the analysed crop, minor variables such as the date of set-aside cutting only for the preceding year. The major model parameters were those determining seed movements and survival in soil and seedling emergence. Seedling survival parameters relative to inter- and intra-specific competition were significant only when analysing volunteer density.