Environmental and landscape effects on cross‐pollination rates observed at long distance among French oilseed rape Brassica napus commercial fields

Abstract

Summary Evaluation of our ability to predict cross‐pollination rates (CPR) at long distance using dispersal functions estimated in experimental set‐ups differing by either their scale or their environmental conditions is crucial to establish appropriate management rules following the release of genetically modified (GM) crops. From 1998 to 2004 we measured oilseed rape CPR in commercial fields for 44 donor‐recipient couples separated by 220 up to 2000 m, from up to three different pollen donor cultivars in two French oilseed rape production areas. We followed the same sampling and screening designs and tested the effect of region, year, size, cultivar and distance on the observed CPR. We then compared observed CPR to predictions from six empirical pollen dispersal models based on dispersal kernels that were fitted previously at the local and landscape scales. These predictions allowed us to test the possibility to extrapolate and up‐scale dispersal kernels. The observed CPR varied from 0% to 0·092%. They were higher in Champagne‐Ardennes, where they depended negatively on distance, than in Bourgogne where they did not depend on distance. CPR also differed among years, being nil for the last 3 successive years, partly because of different environmental conditions and detection issues. CPR depended further on the source cultivars due to differences in pollen production. Dispersal kernels fitted at the local scale lead to systematic and huge underestimation of CPR, when observed. The power‐law kernel fitted at the landscape scale under‐estimated CPR by two orders of magnitude above 400 m but followed the rate of decrease of the observed pollination with distance. Synthesis and applications. Caution should be taken when extrapolating and up‐scaling dispersal kernels and models because predictions may differ greatly from observations. Models should rely at least upon dispersal kernels estimated from landscape‐scale data obtained in different regions. Models should also integrate several phenomena at the agro‐ecosystem scale including the dispersal by insects and become more mechanistic to account for variations observed among years, environments and distance to borders.