Dear authors,

Thank you for the detailed response and the revised version.
One of the reviewers (#2) provided a detailed review on your revised manuscript.
I think the concerns are severe and I like to ask you to respond to the comment, which you can find below.

Best regards,
Volker Grewe

REVIEW (by reviewer #2):

Lewis and colleagues have provided responses to all of the points raised in my first review. However, their responses are not satisfactory. My main concern remains, related to the potential geophysical inconsistency which the authors are introducing between global AOGCM emulations and regional RCM simulations driven by AOGCM boundary conditions. The authors mention in their response that they „have no AOGCM realizations“. My understanding from a careful read of the manuscript is thus that they are using 19 AOGCM tuning files (apparently being used without knowledge of the underlying AOGCM data), together with AOGCM boundary conditions for their RCM derived from one AOGCM (HadAM3p). Their response suggests that they consider using the global mean response from 19 different AOGCMs (represented by their tuned MAGICC emulations) to scale the regional response is a physically acceptable approach. In my earlier review I requested evidence for the robustness of this assumption. The authors indicate that running a full 190 member ensemble with 19 AOGCMs and 10 carbon-cycle models is computationally impossible. I agree. However, there is a much simpler method which allows to illustrate the internal consistency between the AOGCM response assumed in MAGICC and the boundary conditions used for their RCM, based entirely on data already available in the public domain.

The internal and physical consistency of joining global AOGCM emulations with regional patterns of a single RCM-AOGCM pair can be shown by simply computing the boundary conditions for the RCM domain that would be implied by each of the CMIP3 AOGCMs that is emulated. Data for this is available at the CMIP repository. This exercise can be carried out for a small set of salient parameters, either at a given warming level, or at a given concentration level/time slice (for example if a 2xCO2 experiment is used). A statistical analysis of these 19 possible RCM boundary conditions and comparison of these statistics with the actual boundary conditions used for the RCM will show whether there are significant differences between them. “Significant” can here either mean statistical significance, but it can also involve expert judgment about how much these variations would actually matter.

If there are no significant differences between the fields, the authors have provided the necessary evidence that data is combined in a physically plausible way. However, if there are significant statistical differences between the RCM boundary conditions derived from the 19 CMIP3 RCMs and the HadAM3p boundary conditions, the authors either need to clearly argue that these differences would not affect the EOFs used at the regional scale by referring to the earlier literature, or accept that their method implies an unphysical combination of global-mean AOGCM responses emulated by MAGICC and regional responses simulated by their RCM. In the latter case, this caveat should be made appropriately and its implications for the applicability and robustness of the method should be discussed up front.