I would suggest that you obtain an environmental envelope for it where it´s native (bioclimate, elevation, vegetation type, etc.). I.e. fom a trainning set of records of distribution of the insect and correspondent observed environmental features. Then make a Species Distribution Model (using Maxent, GLM, GAM etc.) on the correspondent environmental features in Europe. Hope it helps. Cheers. JC.

It depends what your goal is - if predicting the eventual distribution is the goal then I agree it would be good to take into account not only temperature but hosts if this information is available. These models are getting better resolution all the time but still need validating.

Depends on your objectives, the quality and quantity of data that you have available to parameterize, verify, and validate the model, the data you have available on host distribution. the degree of biophysical resolution desired for the model, the combined expertise of the scientists who will be constructing and/or using the model, and the amount of time you are willing to spend on such a project. If you propose to determine the likelihood of spread under different field- and regional-scales and for a range of management approaches then I recommend you use a model structure that allows simultaneous movement to and from different habitats, with each habitat having attributes of host stage, abundance, relative attractiveness, daily/hourly temperature and wind speed, and access for management intervention. If you wish to model the population as age and space dependent cohorts, then I recommend you handle insect development as a distributed maturation process with ˚C, intrinsic developmental variability, and host quality as drivers. If you choose to model individual fly movement, then you would need to use an individual-based model structure., with spatial/temporal movement handled probabilistically as an age dependent dispersal rate modified by ˚C and wind vectors each time step. C# works well for individual-based modeling and can handle at least several hundreds of thousands of flies at a time. Given the nature of dispersal, you can greatly speed up the model using parallel and/or distributed processing. You could also use high-speed super computers. 

My experience with incorporating dispersal into both cohort-based and individual-based type of models is that the transition from a random dispersal to wind aided dispersal  is relatively easy to do. However, the incorporation of host preference is much more difficult due to lack of quality spatial/temporal data for model parameterization, verification, and validation.

Yes - and how do you deal with human assisted dispersal ? In Barclay et al. 2011 on fruit fly eradication we thought it was too hard to include - but in Suckling et al. 2014 on LBAM colonization of California we could see the human influence on top of natural movement.

Yes, that's a tough one. I have a few ideas that maybe we can discuss some day. 

As mentioned above, a lot depends on research questions and the scale of interest. At regional scale, human assisted translocations might be the key, at the local-site scale (few hectares - several square miles), the behavioural features of individual insects may matter more. Simulating behaviour, movements, development etc. of a cohort of individuals (say several thousands or more) in a mosaic spatiotemporally complex environment can be done in several ways, depending on the type of questions one wants to ask, and even relatively simple models (such as Lux 2014) can give reasonable approximations and provide interesting answers . The bottleneck is our understanding of behaviour and ecology, and I am not sure we already know enough about D. suzukii to get (validate) good models, although it might be just the right time to think about it, and take into account while collecting bio-data.