Climate change may dramatically affect the distribution and abundance of organisms. With the world's population size expected to increase significantly during the next 100 years, we need to know how climate change might impact our food production systems. In particular, we need estimates of how future climate might alter the distribution of agricultural pests. We used the climate projections from two general circulation models (GCMs) of global climate, the Canadian Centre for Climate Modelling and Analysis GCM (CGCM2) and the Hadley Centre model (HadCM3), for the A2 and B2 scenarios from the Special Report on Emissions Scenarios in conjunction with a previously published bioclimatic envelope model (BEM) to predict the potential changes in distribution and abundance of the swede midge, Contarinia nasturtii, in North America. The BEM in conjunction with either GCM predicted that C. nasturtii would spread from its current initial invasion in southern Ontario and northwestern New York State into the Canadian prairies, northern Canada, and midwestern United States, but the magnitude of risk depended strongly on the GCM and the scenario used. When the CGCM2 projections were used, the BEM predicted an extensive shift in the location of the midges' climatic envelope through most of Ontario, Quebec, and the maritime and prairie provinces by the 2080s. In the United States, C. nasturtii was predicted to spread to all the Great Lake states, into midwestern states as far south as Colorado, and west into Washington State. When the HadCM3 was applied, southern Ontario, Saskatchewan, and Washington State were not as favourable for C. nasturtii by the 2080s. Indeed, when used with the HadCM3 climate projections, the BEM predicted the virtual disappearance of 'very favourable' regions for C. nasturtii. The CGCM2 projections generally caused the BEM to predict a small increase in the mean number of midge generations throughout the course of the century, whereas, the HadCM3 projections resulted in roughly the same mean number of generations but decreased variance. Predictions of the likely potential of C. nasturtii spatial spread are thus strongly dependent on the source of climate projections. This study illustrates the importance of using multiple GCMs in combination with multiple scenarios when studying the potential for spatial spread of an organism in response to climate change.

1 The pea leafminer Liriomyza huidobrensis (Blanchard) (Diptera: Agromyzidae) is an invasive species in North America and a serious economic pest on a wide variety of crops. We developed a bioclimatic envelope model (BEM) for this species and examined the envelope's potential location in North America under various future climates. 2 We compared the future bioclimatic envelopes for L. huidobrensis using either simple scenarios comprising uniform changes in temperature/precipitation or climate projections from general circulation models (GCMs). Our simple scenarios were: (i) an increase of 0.1 • C per degree in latitude with a 20% increase in summer precipitation and a 20% decrease in winter precipitation and (ii) an overall increase of 3• C everywhere, also with the same changes in precipitation. For GCMmodelled climate change, we used the Canadian Centre for Climate Modelling and Analysis GCM (CGCM2) and the Hadley Centre climate model (HadCM3), each in combination with two scenarios from the Special Report on Emissions Scenarios (A2 and B2). 3 The BEM results using the simple scenarios were more similar to each other than to the results obtained using GCM projections. The results were also qualitatively different (i.e. spatially different and divergent) depending on which GCM-scenario combination was used. 4 This modelling exercise illustrates that: (i) results using first approximation simple climate change scenarios can give predictions very different from those that use GCM-modelled climate projections (comprising a result that has worrying implications for empirical impact research) and that (ii) different GCM-models using the same scenario can give very different results (implying strong model dependency in projected biological impacts).