Just to be clear, that is not so much an issue if you are not changing the timestep or resolution much. If you just want to continuously do short-term forecasts from data coming in in real time, but sticking with the typical time/length scales, that won't encounter the same issues, and I've learned that there are some people who do something similar for CAM. But it's hard to say without knowing more about what you want to accomplish.If you are running something closer to a typical WACCM case (2 deg, 1/2 hour physics time step), the main challenge will be continually creating new initial conditions from data coming in. Such data will typically not have all of the variables necessary to initialize the model (especially for WACCM). Furthermore if you simply interpolate the dynamical variables (wind speed, temperature) onto a given grid, this will typically be very noisy, and may even cause dynamical instability that crashes the model. So you have to work out how to get reasonably self-consistent initial conditions with a minimum of "spin-up" time. You may want to consider using specified chemistry instead of the full model, since this is cheaper and may be simpler to prepare input data for.
