Dear Authors,

Your manuscript is now accepted for publication subject to final minor revisions, in response to the thoughtful and specific comments of Referee 2 (repeated below). Once you have addressed these comments, I can make a final decision on publication.

Best Regards,
Bob Marsh


Report of Anonymous Referee #2:

The authors have substantially improved the manuscript and have
addressed most of the points raised in my first review. This is a nice
paper. There are however a few minor points which needs to be fixed
before the paper is published.

- The references dealing with wetting and drying and adaptivity need
to be clarified in the introduction (line 15) and in section
5.3. The paper of Harig et al, 2008 is not dynamically adaptive (it
uses a variable-resolution unstructured mesh which is fixed in
time), it does not really deal with wetting and drying but
represents complex coastlines using the unstructured mesh. The paper
by Popinet & Rickard, 2007 does not deal with wetting and drying
either but is dynamically adaptive and represents coastlines using a
"cut cell" approach. The paper by Popinet, 2011, mentioned in my
first review but not included in the revised version, deals with
wetting and drying, is dynamically adaptive and includes adaptive
tsunami simulations (including the 2004 tsunami). Other references
with similar properties include the AMRClaw papers of George,
Leveque, Berger et al.

- The additional data provided in the revised version regarding
computational efficiency are still not sufficient to estimate the
computational efficiency of the model. From the sentence (line 633):

"In terms of actual cpu time, it takes on average 9.1s of wall clock
time for 1s of physical time for the 475 m on 256 cores."

I can try to estimate the computational speed but only if I know
what the timestep is? and how many grid points are used? I can try
to estimate this from the sentence line 696 and from figure 13. The
average compression ratio is something like 600 which would give an
average number of points of about 5e6. The timestep at 475 m, based
on the gravity wave speed should be around one second (or less)
which gives an absolute computational speed of roughly 5e6 / 9.1 /
256 = 2146 points x timesteps / sec / core. This seems very low, so
my estimate must be wrong somewhere (maybe the timestep is only 0.1
sec etc...). Of course the authors have all the data necessary to
give an accurate number for this speed and should do so, as
indicated in my first review. Without such data it is very difficult
to be fully convinced by the claims of efficiency made by the
authors (as indicated by my crude estimate above).

On a related note, I find the sentence (line 634)

"Since the code has 94 % strong parallel scaling efficiency for (at) this number
of cores operational forecasting should be possible using a few thousand cores."

to somewhat contradict the sentence in the conclusion (line 734)

"The grid adaptation will only be done in the horizontal plane and so
the three-dimensional model should actually have better parallel
performance than the model on the sphere since the computational load
will be better balanced."

"better parallel performance" than 94%? Really?

What would seem more reasonable to me is that the code does not
achieve a 94% parallel efficiency in the 2D case (has this really been
measured for the tsunami or wind-driven ocean circulation cases?) and
that it will indeed scale better in the 3D case.

Dear authors,
I am satisfied that you have completed the minor revisions and your paper is now accepted for publication - well done!
Best Regards,
Bob Marsh