|Thanks to the authors for their efforts in revising the paper. The updated manuscript is significantly improved, and I can now see a path to publication. However, I have a few clarifications & queries arising from the responses:|
I apologise for my slightly misleading comments about Boutle & Morcrette 2010. The discussion in there that I was thinking of was:
"Cv and the condensate amount (qc ) are obtained from
averages of Cv and qc on the sublevels, and are used
when calculating the microphysical transfers that lead
to the formation of precipitation (Wilson and Ballard,
i.e. when the UM is run with a diagnostic cloud scheme (similar to ECHAM-HAM), the sub-level interpolation method is used to create new (increased) values of condensed water and volume cloud fraction, which are communicated to all parts of the model, not just the radiation. It was this experiment that I was suggesting with SC-SUND, i.e. in that case, a very similar thing is happening - you are diagnosing new (increased) condensed water values, which could be passed to all parts of the model, not just the microphysics. You would still be using the volume assumption for cloud fraction. I agree that in your application of SC-VOL and SC-MAX, where you only calculate the area cloud fraction, and do not alter the condensed water, it is only appropriate to pass this to the radiation (as with the prognostic implementation discussed in Boutle & Morcrette 2010 which you mention).
However, your answer to the next point makes me think that this probably wouldn't actually be that helpful, because it appears that the amount of additional cloud being created by SC-SUND is negligible. Do you understand why such a negligible amount of extra condensate is created with SC-SUND? The increase in fraction (up to 30%) seems quite significant, and presumably given the constrained link between fraction and condensate in the Sundqvist scheme, the increase in condensate is commensurate to what would otherwise be obtained in clouds of this fraction?
Which leads me to a further question/clarification. My understanding is that SC-SUND is only used to create new clouds where previously there were none? What if it were also used to increase the water content of pre-existing clouds? The inversion sharpening code could be used in exactly the same way to re-diagnose the water content of previously existing clouds. If, what you have discovered so far, is that clouds with fraction of 0-30% are radiatively unimportant due to small water contents, increases to the water content of previously existing clouds should stand a better chance of being radiatively important. Is it the case that the increase in water content with the Sundqvist scheme is quite sensitive to the cloud fraction, i.e. increasing fraction from 0-20% gives a smaller increase in water content than increasing from 40-60% would (for example)?
I would still push the authors to investigate some combination of the above experiments, as I feel it could be useful in understanding what is going on. Certainly some further discussion on the behaviour of the Sundqvist scheme and why it does not produce much condensed water at low cloud fractions would be useful.
I also think that you could strengthen some of the discussion about the frequency of occurrence of stratocumulus clouds in the model (and perhaps include this in the abstract). I think this strengthens your work. The methods you propose are only really useful for targeting errors in "amount when present" (with the exception of SC-SUND). But it appears the main bias in the model is in "frequency of occurrence", and therefore it is possibly unsurprising that the proposed changes have limited benefit. It could be argued they would be much more successful in models with good frequency of occurrence and poor amount when present.