|I had two major issues with the original manuscript, and both have been largely addressed. The first was the lack of a convergence study of the MSIMIP3d Std experiment, which is now present. The second was a tendency to insist that intercomparison exercise needed, rather than benefitted from, more than one Stokes code. That has been toned down, but I still have some minor disagreements.|
The convergence study suggest an error estimate in GL position of ~300 m at dx_min = 50m. But the difference between ‘grown’ and ‘shrunk’ GLs at dx_min = 50 is ~5 km . That, together with the neither linear nor quadratic form of the Richardson estimates tells us that the FELIX-S is not in the asymptotic regime (nor is Elmer/Ice), either because a finer resolution is needed at the GL, or the region of fine resolution needs to be bigger. However, I don’t believe that anyone else has achieved better results than this. Perhaps the lesson is ‘estimate the error in as many ways as possible and take the largest number’
In the abstract, we have ‘at a particular resolution, the span of grounding lines positions provides one estimate for model error … More importantly we show that the grounding line positions appear to converge to within the estimated truncation error of Elmer/Ice’. I think this needs to be rephrased, because you always need the convergence study (different experiments result in different truncation errors), and so the model difference should never be needed as an error estimate - but might provide a useful complementary estimate where the convergence study cannot be carried to the asymptotic regime.
I noticed that the other two reviewers seemed to think this was more of a GMD paper than a TC paper. I don’t fully agree - the paper could be equally at home in GMD, but we do want glaciologists. rather than model developers, to understand the limitations of the models they choose. So I see no reason why this discussion cannot be in TC, especially since parts of it (e.g the Gagliardini 2016 paper) already are. There was also a question of novelty. There I think that, since we need results to be independently reproduced before we believe them, there must be some avenue open to publish reproduced results, especially the first reproduced results.
P6, L30 : ‘very good agreement’ -> close agreement, or something that does not involve ‘very’
P7, L3: ‘the horizontal velocity’, u -> ‘ the x-component of the horizontal velocity’, u
P7, L24 ‘...superior to the other’ : and in any case should vanish as dx -> 0
P7, L26 : ‘very’ …
P7, L32: (50 m in the vicinity of the GL and 2500m , respectively) what about the x-resolution far from the GL?
P7,L34 ‘in equlibrium … based on’ -> ‘close to equilibrium… according to’
P8 L3 ‘more retreated’ -> ‘further upstream‘
P8 L35 solution error -> estimates of solution error
P9: L1 ‘clearly convergent’ -> seemingly convergent
P11, L5 ‘gradients in ice sheet geometry’ -> differences in ice sheet geometry between nodes