You are asking some good questions here, but they do not all have definitive answers.
The effect of changes in air pollutant emissions is generally larger than the effect of climate change, especially over the near term. When comparing 2023 or 2024 to 2030, most practitioners, at least at EPA, would keep the meteorology unchanged in the future year and model the effect of the future emissions scenario.
Different researchers have employed different approaches for downscaling global climate model simulations with WRF. Our group has chosen to use long-term continuous simulations with nudging (Bowden et al., J Clim 2012; Otte et al., J Clim 2012; Spero et al., JAMC 2015). Others have not used nudging but have instead used frequent reinitializations. There is by now a fairly large and growing body of research on regional climate downscaling. Several review articles have been published, and there are model intercomparison activities (e.g., CORDEX, NARCCAP).
I use CMAQ driven by WRF after it has been processed by MCIP. In other words, the meteorology is “offline” or “uncoupled” from the simulated air pollution. That would not be the case with WRF-Chem. I have simulated continuous decadal (actually 11-year) periods, subset from the multidecadal WRF simulations our group has conducted. These are at relatively coarse (36-km) spatial scale. Others have chosen to sample specific days for modeling, e.g., Mahmud and Kleeman, ACP 2012).
How to account for interannual variability (e.g., Fiore et al., JGR 2022) while keeping the modeling computationally tractable is a challenge.
It is theoretically possible to run CMAQ for long periods continuously. In practice, most researchers at EPA (including myself) execute the model for 24 hours at a time, for convenience in scripting and input data preparation. In these runs, the model state at the end of one day is the initial condition for the next, so we consider these to be continuous simulations.