The authors' responses indicate that the concerns raised by the reviewers can be addressed rather well. I consider these revisions minor, and I will review the revised manuscript myself.

The authors have done an excellent job of responding to the reviewers. I have a few additional comments that would be worth addressing, mostly in the form of additional discussion in the manuscript:

1) Because the authors were perturbing parameters, they didn't really need information about how variables change in time. That could have been a separate analysis, which the authors do to some degree. I think this separation could be clearer.

2) I think there are a few places where the authors overstep. Higher or lower correlation does not mean more or less information, nor can it tell you which variables are more preferable to use. The actual relationship between the variables is also important.

3) The authors do not discuss interaction terms, i.e., what happens if multiple parameters are changed at once. We know the climate system contains nonlinearities, so these interaction terms should not all be zero.

4) If possible, it would be interesting to discuss some "minimum data requirements" for carrying out this method. For example, the authors were using two time periods that were fairly far apart. What if they were only 20 years apart? Or 30? Would the results be as clear?