|General comments: |
The revised manuscript is significantly improved, also including responses to all of this reviewer's comments/questions about the previously submitted (initial) draft.
The title is now more appropriate (operational AEP only), and numerous (formerly ambiguous or unclear) key aspects have been clarified.
The abstract is also better, though it is still lacking `in conclusion':
since the stated purpose is to "examine the extent to which the assumption of uncorrelated uncertainties...is accurate and appropriate for operational AEP calculations", you should clearly state how large of an effect on AEP uncertainty estimation occurs, due to this assumption.
Further, it should be made clear that the results are for low-uncertainty sites, i.e. simple flow-regimes, where you have rejected all sites/data that do not correlate well (R<0.6) with the re-analysis products. This should perhaps also be included in the abstract.
It would be useful and honest to state how many sites were rejected based on the LTC R>0.6 criteria (the conclusions might be different if higher-uncertainty/more complex sites were included; low-uncertainty behaviors tend to be more easily linearized and correlated).
The addition of Figure 2 helps also; though it should be clear that the regression technique is usable in part because the speeds have been density-corrected (and normalized).
There is still some lack of clarity about several points, which are mentioned in the specific corrections below.
line 13 (abstract): how is the Monte-Carlo approach more robust than simple sum-of-squares?
line 52–53: perhaps there should also be a citation of the general engineering (measurement) uncertainty standard, GUM, which includes accounting for correlations between uncertainty components.
Line 81: using M-O theory based on MERRA-2 heat and momentum fluxes?
line 95: reference the International Standard Atmosphere (ISO 2533:1975).
line 101-4: you should state how many sites were rejected due to poor correlation with the re-analysis products. Such site/data rejection limits the analysis and conclusions to sites with lower uncertainty, particularly regarding the long-term correction (windiness). Limiting to low-uncertainty sites can simplify various behaviors; e.g. the correlations may be stronger between uncertainty components.
Line 109, 111,112: "wind speed" needs to be preceeded by _'density-corrected'_ in order to support the linear regression.
line 120: remove `instead'; perhaps italicize `operational' to emphasize this in comparison with non-operational (pre-construction) uncertainties.
l.121-122: How many consultants? The term "industry standard" is (likely) too strong for the characteristics you list, unless you can support the variety and number of conversations, and their representativeness for the wind industry across the world. (E.g. is this from consultants in the Americas, or Europe, or Asian markets?)
line 130, 131, 134: not just monthly average wind speed, but _density-corrected_ monthly-average speeds. Be clear about what is being 'operated upon'.
line 135: not just "gross" energy production, but _estimated_ gross energy production.
l.154-5: you don't sample meter data per se, you generated/synthesized using a Gaussian distribution.
l.156: "coherent" should be "consistent" (or is it equal to that in the IEC-60688?)
lines 160-2: You are in effect using the variability between re-analysis datasets as a proxy for uncertainty; this should be stated, because it might be larger or smaller than the uncertainty in using a given re-analysis dataset. This is analogous to an _ensemble uncertainty_ measure. (perhaps include reference)
l.166-174: include reference, e.g. to the GUM (JCM100:2008).
lines 184-7: do you mean that you randomly pick a number of years between 10 and 20? Or are you randomizing, or perhaps bootstrap-sampling, in another way? Please clarify.
Fig.5: can't see windiness; why not try a logarithmic scale on y-axis?
Section 3.1 (l.216-...): The rejection of sites not well-correlated (R<0.6) with the RA datasets will affect the uncertainty in the linear regression rather significantly (increasing it), and possibly the reference-data uncertainty. As such, the value of 1.5% depends on the rejection threshold (what happens when e.g. R<0.5, or 0.8?).
l.297-8 / Fig.12 caption: not really "dependence", should be 'relationship between' or "mutual behavior of"
l.300: not "direct", but a "positive" correlation emerges
l.305-7: this is not necessarily true -- the correlations are likely to be weaker, when the re-analysis data are less correlated with the site-specific data. As I noted above, some component uncertainties will also increase.
l.319: 6% absolute, or relative to (percent of) %uncertainty?
Table 1: "measured or modeled" should occur before "long-term" in the reference-data description.
line 79: Remove "v2" and put "Version 2 of" at the beginning of the line; remove ")(" between "MERRA-2" and "Gelaro".
line 203: 10,000 times
Fig.6 caption: percent _difference_ between CoV
l.249: pluralize coefficient
l.296: dependence, not 'dependency'
l. 313: isn't it five, not six?
Line 379: The ISO reference appears to have been garbled a bit via BibTeX/reference manager.