In terms of uncoupled AGCM studies, I've seen people use periods within a decade (say, 5 years) as spin-up time, but I also see some people take tens of year as spin-up in some recent studies. So I'm thinking if it's related to the model resolution in different simulations. So does GCM spin-up time affected by the resolution they use? For example if the same model is run in different resolution, would the spin-up time be any different? Thanks!
Without spin-up time, our initial conditions would be blank or generic.
I guess the simple answer is "NO" spin-up time is not really affected by the resolution, unless the model resolution changes are so great that it takes much longer for diffusion / advection between cells to flush out. Though that wouldn't explain a 5 year change in spin up time unless you made drastic changes to a model with very large time steps. Increased model resolution could theoretically increase the necessary spin-up time by a small amount, if the changes in resolution were very large and the time steps used were coarse. However, in a reputable research setting the spin up time is chosen based on how long (in model time) it takes for things to get to an acceptable initial condition, and changes in resolution are usually not so extreme.
Complexities in the model can increase desirable spin-up time if they have long time-horizons before an acceptable initial condition is achieved. The spin-up time you choose for your model will influence the degree to which the initial conditions actually represent a realistic distribution of numerics across all parameters, assuming the model is adequately developed.
I would say that the biggest factor influencing spin-up time is whether the model includes a dynamically-coupled ocean component. A regional or even global atmosphere-only model, with a realistic initial state and forced by realistic boundary conditions, will reach equilibrium pretty quickly; certainly a decade or less. An ocean model may not reach equilibrium even after a thousand years, and in fact this can be seen in many of the CMIP5 model control runs.
It's probably also worth pointing out that while model drift is often attributed to incomplete spin-up in coupled models, often drift is more due to some kind of spurious energy 'leakage' in the model's energy balance, which the ocean component acts to restore. In this case, the model will never reach an equilibrium, 'drift free' state.