Thank you for your response. Clearly you've already been thinking about the issues I wrote about.
I'll elaborate a bit about why I suggest flexibility in dealing with field studies that span few years. We would like to think of a random year factor as providing us with a basis for inference to years in general. But in the absence of a time machine, we cannot obtain a random sample of years from the statistical population to which we want to make inference--so, in practice, we don't really have a basis for temporal inference, on top of which we typically have very few years of data. Field experiments definitely need multiple years because we have scant control over environmental conditions and we know that environmental conditions matter. Philosophically I think of multiple years as exercises in "repeatability" rather than "replication", which segues into metareplication sensu Johnson https://dl.sciencesocieties.org/publications/cs/abstracts/46/6/2486 .
This paper http://onlinelibrary.wiley.com/doi/10.2307/1941729/epdf provides a practical discussion of whether year is fixed or random, and this paper touches on that topic and its inferential implications https://www.jstor.org/stable/3546293
In your initial post, you were seeking input on GLIMMIX code. If you are still interested, post the code that you have so far and the community can go from there.
Good luck, and have fun!
... View more