RCM could be further downscaled e.g. if RCM's resolution was 30 km it may be necessary to further downscale it to reproduce for instance climatology of precipitation fields in mountaneous terrain. There is however a lot of discussion in literature about added value of GCM downscaling: it is not always clear. Techiques for downscaling RCM and GCM are essentially the same, broadly divided onto dynamical and statistical methods.

GCM data is feed into the RCM models. Means the boundary conditions of the RCM models is fixed by the output of the GCM models. RCM is a dynamic downscaling technique to increase the spatial resolution of GCM results. The RCM results can also be downscaled to point scale. The downscaling procedure for GCM and RCM is essentially same.you can read these review paper on downscaling

As complementary, downscaling of the RCM outputs to a local scale is generally the corrections of the biases in the RCM outputs.For that you need the simulated variables or predictors(from RCM) and the observations or predictands (from local scale) and then you applied the downscaling methods in order get the scaling factors. These scaling factors will be used to correct biases in your simulated variables.

Yes RCM is finer than the GCM. In general RCM is the output of dynamic downscaling of GCM variables.

Yes theoretically RCM can be downscale further but in reality it dont happen. Further downscaling of RCM mean that more finer resolution. If you need more finer resolution you can use nested domain in dynamic downscaling and produce more finer downscaled output as per requirement but it is quite costly, time consuming and also introduce further uncertainty. 

On the other hand in statistical downscaling you can directly downscale to any point station. So issue with resolution. It always preferable to use GCM rather RCM for statistical downscaling as using RCM will provide extra uncertainty in the final downscaled product.

Dear Younis

I agree with others, You know GCMs have coarse resolution and their outputs are not exact for a fine area. But if you want to use RCM , you should remember that these models are very time consuming, make error and need to have a strong CPU with many cores, but defiantly their outputs are more precise than GCMs outputs. You know the finer model in CMIP is MIROC4h with 0.5*0.5 degree resolution. In a nutshell, the RCM outputs have exact outputs than GCMs outputs model but it depends on use model & your area of case study.

Sincerelyy yours,

Nasrin.