Review of MWR 2112 (BS) Hines
“Development and Testing of Polar WRF: Part I: Ice Sheet Meteorology”
Reviewer A
Recommendation: Accept with major revisions as outlined below.
General Comments:
The paper presents a study examining the performance of the Weather Research and
Forecasting (WRF) mesoscale model over ice sheets in Greenland during December 2002
as the first step towards development of the so-called “Polar WRF”.
As opposed to another recent paper I reviewed that dealt with high latitude applications
of the WRF model, this paper was generally disappointing, in that arguments and
conclusions strike me as a bit weak and leave me with many questions, which are
addressed below. Some additional work is needed to better substantiate the conclusions
reached.
Also, many regional climate models have been applied to this area at much greater
horizontal resolution than 40km, suggesting important needs for parameterization
development to properly simulate processes at a hierarchy of scales. It seems a stretch to
suggest that tests at 40km are a good indication of the results that regional climate
modelers will achieve at the much higher resolutions they will want to run WRF at to
examine some of the key unresolved questions that have been brought to the fore by
SHEBA, FIRE-ACE and other recent field efforts.
Specific Comments:
1. Page 3, section 1a. One wonders if the section heading a) is really needed since
there is no b section.
2. Page 3, section 1a, paragraph 1: Second sentence is awkward grammatically,
specifically the clause starting with “maintained”. Suggest a minor rewrite of the
sentence to correct this.
3. Page 3, section 1a, bottom of page: Here this seems to directly refer to the AMPS
forecasts…why not just state that Polar MM5 is utilized as the basis for AMPS?
4. Pages 4-7: The initial impression I had, upon reading the Polar MM5 section of
the Introduction , was that the Polar MM5 modifications had been translated to
Polar WRF. This is not at all clear from most of the Polar WRF discussion (save
the land surface discussion) which reads mostly like an overview description of
standard WRF options than anything else. Were modifications made to any of the
other packages?
5. Pages 5-6, paragraph that is split over these two pages. Simply say: “The
Advanced Research WRF (ARW) is a modular non-hydrostatic model…” and
omit the first full sentence on page 6.
6. Page 6, first full paragraph. 28 vertical levels, with a top at 10 hPa and
supposedly “high resolution” in the PBL (though more than the lowest two levels
is not explicitly specified), suggests rather coarse resolution somewhere in the
vertical column and leads one to wonder if some important processes in the mid
and upper troposphere are properly represented. I am not sure of the need to
cover so much of the lower stratosphere (~ 200 – 10 hPa) , which is generally
only represented by satellite-retrieved information in most operational analyses.
Some justification of the need for such a high model top would be welcome.
7. Page 7, bottom and all occurrences thereafter. “NOAH” is an acronym and
should be represented as such. Also, why was the 5-layer thermal diffusion land
surface model not tested given the period of interest is December and subsoil
water is likely frozen, making the water phase changes represented in the other
land models less significant?
8. Page 8, ending of section 2: The treatment of snowpack and heat transfer at the
snow surface is at best interesting, at worst it may violate inherent assumptions in
the NOAH model. What is the rationale for this approach, which comes across as
rather ad hoc? At minimum a discussion of exactly what parameters and/or
formulations within NOAH are impacted and how.. is appropriate.
9. Page 9: The crux of my resolution related comment above is here. While I
understand a desire to have comparative tests with Bromwich et al (2001), I think
tests at higher resolutions more likely to be utilized by regional climate modelers
in an age of increased computer power at least relevant to attempt and could
provide much useful information on the utility of this particular WRF
configuration. Also see next comment.
10. Page 9: What is the resolution of the Ekholm (1996) digital elevation dataset?
The author notes that 40 km resolution is not really sufficient to resolve the
steepest margins of the ice sheet…how much spatial area is involved here and
could it be a significant impact on the results. Again, at least one higher
resolution experiment would clarify many of these issues.
11. Page 9: “Some surface energy balance terms are available…” Which ones?
Shouldn’t verification be as consistent as possible across the very limited station
base available (more on that later)?
12. Page 9, bottom: Many in the land surface modeling community feel that, for
longer term climate/regional climate simulations, longer spin-up times (on the
order of weeks) may be needed to truly reach an initial state that will be free of
model shocks. This point is still debated to varying degrees, so some justification
(maybe a sentence that is more definitive…ie, saying that Bromwich et al (2001)
tested various spin up times and found 24 hours to be sufficient) is needed.
13. Page 10, primary paragraph. OK, this paragraph floored me. Unless verification
is very carefully conducted with the disparate grid resolutions, we could well
have an apples versus oranges situation w/r/t verification. It is not appropriate to
assume equivalence between model quantities on grids of different resolutions as
a unique indicator of relative performance. If anything, all that could be
represented is the difference in grid resolution. The situation is further muddied
by the need to time-average results and then therefore the observations. Many
sins can be covered by such averaging processes, and it lowers my confidence of
the significance of the results obtained.
14. Page 11, second paragraph. “…the optional WRF physics” makes it sound as if
WRF can be run with no physics (which in some sense it can but that is not the
point here).. Better to say “the WRF physics options”.
15. Page 12: Some discussion of how the interpolated values for Swiss Camp are
derived is in order, given the sharp terrain slope, which, as was pointed out
earlier, may not be well-represented at the 40km resolution. Such interpolation
usually needs to be conducted very carefully as to not introduce bias, since
standard relationships for reduction to 2- or 10-m values often can produce overor undershoot of values. Numerous papers related to similar issues with the
Tibetan Plateau are in the literature.
16. Page 12 and hereafter. While averaging of the error statistics to produce a general
picture of performance is often done, it can often hide a multitude of sins,
especially for such a small number of stations. Are these results statistically
significant? Can statistically significant verification even be done with such a
small verifying dataset? Significance testing, for individual stations over the
month-long period as well as the aggregate, is needed for there to be real
confidence in the validation.
17. Page 13. I question whether the results for one point (Swiss Camp) are really
sufficient basis to eliminate certain configurations. And the question also arises
as to whether the simulation of the katabatic flow regime, important as it may be ,
and as represented by the wind speed simulations at a single point, is the most
important criteria for disregarding the other simulations. Given that the large
domain covers a much larger area than this point realistically represents from a
meteorological point of view, this may be throwing out the baby with the bath
water.
18. Page 13, discussion of Figure 6. Of course absolute correlations will likely be
high, as the computation implicitly includes the impact of the entire column as
opposed to the small variations. Anomaly correlations would provide a truer
sense of the performance of model in simulating pressure field variations.
19. Page 13. Similarity of correlations does not rigorously equate to independence of
the temporal structure of the errors. Independence should be more rigorously
tested if claimed.
20. Page 13. 2nd sentence from the bottom of the page. This is a curious point. Since
the MM5 also comes equipped with the RRTM longwave radiation package,
would simulations using Polar MM5-RRTM produce similar results to the WRF
configuration? If so, does that lessen the uniqueness of the improvements to the
simulation via the use of WRF (and thus partially negating the rationale for the
development of Polar WRF). A comparison run here would be very illuminating.
21. Page 14. I understand what the author is trying to do here, but the fact that the
Reanalysis vertical profiles perform worst near the surface leads one to suspect
that any comparison of PBL vertical structure between WRF and NCEP-2 is at
best tenuous. And how much does the resolution of the NCEP-2 reanalysis play
into this assessment? The sentence at the end (which continues onto page 15) adds
nothing to the discussion at hand and should be omitted.
22. Page 15, first full paragraph. Grammatical error, 2nd sentence: “…the western
Greenland slopes Swiss Camp….”
23. Page 15. Are the 10-m wind speeds for the Polar MM5 and Polar WRF both
computed directly as output fields, or are they obtained by interpolation, and if the
latter, is the interpolation algorithm the same? Otherwise the interpolation
algorithm could be introducing relative bias into the comparison.
24. Page 15, bottom paragraph. “Polar WRF simulates a larger speed by about 2 m/s
in the upper troposphere at Swiss Camp”. I see a larger speed all the way down to
850 hPa, which is hardly the upper troposphere.
25. Page 16, section 5. It seems very odd to focus so much attention on winter
verification at Swiss Camp while the summer verification is for a different station
(Summit). What is the rationale for this? Are we just being shown the “good”
results, carefully selected to give the best possible impression of WRF?
26. Page 17. Top paragraph discussing radiation. This is interesting, but what is the
physical process that is poorly simulated in Polar MM5…are the cloud bases too
high (thus colder clouds and less downwelling IR)….is cloud coverage (# hrs/day)
too little? Poor track forecasts of synoptic systems (leading to one or the other of
the previous possibilities) It is hard to determine where a parameterization might
be failing if the details are not examined. Did you utilize available satellite
imagery to determine if the cloud coverage had spatial/temporal errors?
27. Page 17. How is the diurnal cycle smoothed? It could play a role in the results
and interpretation, particularly the point at the bottom of the page talking about
the damping of the diurnal cycle in the WRF runs.
28. Page 17, bottom of 1st full paragraph. Why not show what the differences in the
cloud fields (even averaged ones) are? That way you can make conclusions
instead of merely speculating.
29. Page 18, 4th line. “We shall use this opportunity…” sounds a bit like the author is
giving a lecture. It is overly grandiose for a written article.
30. Page 18, last line, 1st full paragraph. Awkward. Better: “…simulations will
employ the Goddard shortwave scheme in response to the results shown in Figure
13”.
31. Page 18, last paragraph through page 19. “…initial subsurface
temperature….monthly average skin temperature from an earlier WRF run”. How
much earlier? What is the justification? This is an approach that is more suited to
operations rather than a rigorous model performance research study. What is the
exact formula used? All of this certainly seems a lot like “tuning” and given it is
only applied based upon the results at a single point, is not likely to be generally
valid. Is radiation allowed to penetrate the snowpack as indeed happens to
some extent in the real world? NOAH would likely not be designed to include
this under the assumption that radiation strikes a ground surface and does not
penetrate to any depth.
32. Page 20. Given the importance of the cloud mixing ratios to this discussion,
inclusion of ice, graupel and rain may still be significant from a radiative point of
view. If need be, they should be plotted on a separate graph.
33. Page 20: “the boundary layer cloud…not adversely impacting the incident
shortwave radiation” Direct, diffuse or total? This seems like an odd result.
34. Page 20: It may be an exaggeration to say that Greenland has a pristine
atmosphere….I believe…though I may be wrong…in saying that Arctic haze has
occasionally spread as far as Greenland, not to mention the occasional, though
very infrequent influx of volcanic ash from nearby Iceland. More importantly in
this same paragraph, why do you want to encourage moisture loss from the
atmosphere? You do not state that this is necessarily a deficiency in any of the
simulations. What is the physical basis for this?
35. Page 21. OK, now Swiss Camp comes back into the validation…why?
36. Page 22, top. Based upon Figure 16, you can only claim that Polar MM5 better
simulates the summer surface layer , not necessarily the entire boundary layer,
which is of a greater depth and not necessarily adequately evaluated solely by
looking the surface level.
37. Page 22, bottom-Page 23. This is really a stretch. The values for December 2000
and 2001 are only representative if the mean synoptic/mesoscale patterns
(including cloud cover, etc) come somewhat close to the patterns for December
2002. You have provided no indication, via figure or discussion that this is the
case. Ditto for the use of January 2006 observations. Unless you can demonstrate
equivalence, this entire discussion is invalid.