The cultivation of Genetically Modified (GM) crops may have substantial impacts on populations of non-target organisms (NTOs) in agroecosystems. These impacts should be assessed at larger spatial scales than the cultivated field, and, as landscape-scale experiments are difficult, if not impossible, modelling approaches are needed to address landscape risk management. We present an original stochastic and spatially explicit modelling framework for assessing the risk at the landscape level. We use techniques from spatial statistics for simulating simplified landscapes made up of (aggregated or non-aggregated) GM fields, neutral fields and NTO's habitat areas. The dispersal of toxic pollen grains is obtained by convolving the emission of GM plants and validated dispersal kernel functions while the locations of exposed individuals are drawn from a point process. By taking into account the adherence of the ambient pollen on plants, the loss of pollen due to climatic events, and, an experimentally-validated mortality-dose function we predict risk maps and provide a distribution giving how the risk varies within exposed individuals in the landscape. Then, we consider the impact of the Bt maize on Inachis io in worst-case scenarii where exposed individuals are located in the vicinity of GM fields and pollen shedding overlaps with larval emergence. We perform a Global Sensitivity Analysis (GSA) to explore numerically how our input parameters influence the risk. Our results confirm the important effects of pollen emission and loss. Most interestingly they highlight that the optimal spatial distribution of GM fields that mitigates the risk depends on our knowledge of the habitats of NTOs, and finally, moderate the influence of the dispersal kernel function.
Keywords: Bt maize; Environmental risk assessment; Inachis io; Landscape risk assessment; Pollen dispersal.
Copyright © 2017 Elsevier B.V. All rights reserved.