|2nd Review of “Feasibility of improving a priori regional climate model estimates of Greenland ice sheet surface mass loss through assimilation of measured ice surface temperatures” by Navari et al.|
The manuscript has improved, like the introduction, and quite a few errors or omissions have been corrected. Still, at some points the wording must be improved, but after that the manuscript is suitable for publication.
My comments are all related to the role of refreezing on the results. In brief: as no subsurface parameters in CROCUS are varied, the OSSE test thus implicitly assumes that refreezing is perfectly modeled. As refreezing is one of the poorest captured processes on the SMB of the GrIS, this is an important issue to be mentioned in the manuscript and abstract. However, except at the transition from the ablation zone to the percolation zone, refreezing has a limited effect, it either refreezes all (percolation zone) or relatively little (lower ablation zone) of the meltwater production. Therefore, no fundamental revision of the manuscript is required.
P2 L21 or L24, add something like: Since subsurface model properties were kept constant, meltwater refreezing is treated as perfectly modeled which positively affects the performance for estimating runoff and surface mass loss.
P3 8-12: “While” expresses a contradiction while for me the results of Van Angelen, Fettweis and Enderlin point to the same ongoing change: runoff will determine future mass loss and is its current importance is already growing.
P7 L22: “melt runoff from the snowpack”: use “meltwater” instead an of “melt”, furthermore, the ice sheet surface is not always a snowpack, in case of runoff there is most of the time only ice at the surface.
P7 L22: Add a comment that refreezing is implicitly included in this equation, in the sense that it is not a term in equation (1) but affects the SMB because it reduces runoff.
P8 L9-16: I requested this new part in my previous review, but the formulations included now are so general that they give little help on how the model works – and they are incorrect too. Have a look on your data how in general the fluxes are directed and remove this part. Instead:
P8 L23: Add a few lines on how the turbulent module works. I had a short look in Brun et al (1989), so I now know that CROCUS seems to use some kind of bulk method but the tuning parameters a and b (Brun et al, left column of p 334) are not given for CROCUS. As suggestion: if CROCUS were using the same method as explained in Van den Broeke et al (2008, Int J Clim), section 3.1, I would summarize the turbulent routine for this paper as
“CROCUS derives the turbulent sensible and latent heat fluxes using the bulk method (Van den Broeke and others, 2008), which applies Monin-Obukhov similarity theory to estimate fluxes from the near-surface wind speed and the temperature and humidity differences between the surface and the temperature at X m, prescribed by MAR. For non-neutral conditions, the stability functions proposed by Holtslag and De Bruijn (1988) and Dyer (1974) are applied to adapt the fluxes for stable and unstable atmospheric conditions, respectively.”
P19 L12-13: Add a comment that snowmelt is not necessarily zero but that refreezing inhibits runoff.
P19 L20: Add a comment that refreezing mitigates runoff and that refreezing is a poorly constrained process and not optimized by the OSSE described here; it is assumed to be modeled correctly.
P23 L24-P24 L2: The main problem is not that the truth is an outlier, but that the SML depends on refreezing and I don’t see how the SEB, which is optimized by OSSE, influences refreezing. Therefore, it should be added at this point that the effect of refreezing on SML is not optimized. This has no effect in the upper part of the percolation zone and dry snow zone because refreezing absorbs all melt water. It is also a relatively small problem in the lower ablation zone because SML fluxes are magnitude larger than the refreezing capacity. However, at the edge between the ablation zone and the percolation zone, this adds a significant non-assessed uncertainty. Here, refreezing reaches is largest contribution to the SMB. As refreezing not yet perfectly understood, hard to validate and differently modeled by various RCMs (Vernon 2013) despite the comparable overall SMB numbers and trends, it is by no way granted that it is correctly modeled by CROCUS. An idea how good OSSE removes errors in refreezing can only be assessed by using data from RCMs or adding additional constraints.
P24 L19: in this context of constraining refreezing, 10m snow temperatures might help because they provide a proxy of refreezing.