The manuscript “The utility of simulated ocean chlorophyll observations: a case study with the Chlorophyll Observation Simulator Package (version 1) in CESMv2.2” by Clow et al. describe a satellite based ocean chlorophyll observation simulator driven by the NCAR coupled carbon-climate Earth system model and its implication for the interpretation of satellite chlorophyll observations under the confounding influence of clouds, limitations of passive solar reflectance observations and other practicalities of polar orbiting satellite observations.  The analysis is an important assessment of uncertainty and bias in these observationally based estimates and I have only minor points where I thought more explanation was necessary.

Specific points:

62 – what about the uncertainty associated with comparison of satellite optical depth and fully vertically resolved models?

Figure 1 – Cloud cover scale looks like it should only go down to 40%... how much of the oceans is the 20-40% range?

70 – The mechanistic explanation is fairly simple and should be explained here – Areas where the ocean is cooler than the air (like upwelling regions) tend to cool the air, raise the humidity, and form clouds.  This is in addition to all ocean areas tending to raise the humidity.

187 – “equilibrium of the deep ocean can take hundreds of years”  Actually, equilibrium of the deep ocean can take thousands of years.

234 – “As expected”… It is not clear to me which of the model simulator versus real world differences make the strong difference in Fig. 4 “expected”… is it the return period of once a day in the model versus once every two days in the observations (line 178 “low latitude gaps are then filled during an orbit on the subsequent day”? More explicit attribution is needed here given the large number of differences described above… is the following sentence “The ISCCP configuration of ChlOSP samples more frequently than real-world sensors…” intended as the explanation?  The connection to the previous discussion points is not clear.

241 – “white caps, coccolithophores, and aerosols are already simulated in some capacity in CESM and could be added to ChlOSP with minor modifications” if this would have improved the model with only “minor modifications” why was it not done in the present study?

Table 1 – My interpretation of the longer e-folding time scales in the simulator than the observations is that the underlying model is missing important local scale forms of variance such as eddies, fronts, jets, etc. such that there is little correlation between the values separated two days apart.  It is not clear the mechanism to which the authors are attributing this difference.  In the simulator which is sampling daily, there is a resolved signal of autocorrelation.  In the observations, there is not which contradicts the model behavior of signals persisting for 4 days.  It is hard to know how to interpret the value of the model as an autocorrelation simulator in this case.

273 – “We have also shown that the missing data leads to a more realistic modeled representation of the chlorophyll variance.” I do not see this assertion being supported anywhere as autocorrelation and variance are different things… has the variance in both the underlying model and simulator been assessed?

Figure 9 – I am not sure the meaning or value of this figure.  The title refers to “chlorophyll temporal correlation, but I think the authors intend “chlorophyll bias correlation”… Are positive values where cloud cover leads to a positive bias in chlorophyll, and negative where it leads to a negative bias or is it truly the temporal correlation of high clouds when chlorophyll is temporally increasing as the title suggests?  If the latter, than what is the significance?  More clarity is needed.

319-320 – “If satellite sensors could see through clouds (as in the 320 Clear-Sky configuration), the global chlorophyll mean would be overestimated by 14 to 22 %.” Is this due to the diurnal sampling bias, the seasonal sampling bias, or something else?  The following sentences discuss mechanisms and may be connected.  Answering this may simply involve switching the order of the sentences such that the mechanisms are attributed directly to the result… e.g. “… As a result, if satellite sensors could see through clouds (as in the 320 Clear-Sky configuration), the global chlorophyll mean would be overestimated by 14 to 22 %.”

Review of “The utility of simulated ocean chlorophyll observations: a case study with the Chlorophyll Observation Simulator Package (version 1) in CESMv2.2” by G.L. Clow, N.S. Lovenduski, M.N. Levy, K. Lindsay, and J.E. Kay

The Authors present a tool for generating synthetic observations of surface ocean chlorophyll that can be incorporated into fully coupled Earth system models.  Overall, I am highly in favor of such work and believe the Authors did a satisfactory job in showing the utility of their tool.  For example, while ocean color satellite data records are widely used and highly valuable, they suffer from sampling biases that are not often considered.  The ChlOSP presented here offers an opportunity for understanding and improving upon interpretation of these biases. The manuscript is also very well written and the figures are clear.

My biggest issue is with my lack of understanding of the chlorophyll retrieval itself (how it is calculated and/or how the weights are used) and of its performance and validation. Some comments below touch on this.  I also left wondering about how the results would look if shorter time scales were considered (from daily to monthly to annually).  For the most part, decadal averages are presented.  This paints a great average picture, but shorter time scales are important too, particularly regarding some of the applications of ChlOSP presented by the Authors.  It would be appreciated if they add a quick case study on shorter time scales.

Overall, I’d advocate for publishing this manuscript after minor revision, with attention in particular to the chlorophyll retrievals.

Specific comments:

Line 29: Since Hu uses a line-height approach, you might mention this in addition to just the blue-to-green ratio.

Line 139: The satellite “sees” a vertically optically weighted chlorophyll signal, not a sum of chlorophyll layers. Consider commenting on the difference between the remotely sensed signal and the sum that you use.  Note also that in some waters, the satellite will not see 10 m deep in some (blue) wavelengths used in the retrieval algorithm.

Lines 135 and 141: “missing” and “viewable” refers to present and absence of clouds, etc.?  Suggest clarifying in both places.

Line 144:  Suggest changing to “prevent satellite detection using passive instruments”.

Line 150:  Change to “Level 3 ocean color retrievals have a high spatial resolution”.

Line 155:  I don’t understand as written why you’d multiply the surface chl by the weights.  I get that you’re deriving a fraction of viewable area in each 1-deg model cell that represents how much of that cell MODIS would see (0.5 indicating 50% of the 1-deg cell would be visible to MODIS).  Right?  But, applying that weight to the chl value would artificially lower the chl value, no?  As an example, if you’re 1-deg cell has a chl value of 1.0 mg/m3 and that cell in reality is completely homogeneous, then even if half the cell was not viewable by MODIS, the satellite pixels would still have values of 1.0 mg/m3.  But, multiplying by 0.5 would lower the value to 0.5 mg/m3.  I have no doubt that I’m missing something, so please elaborate in this section.

Line 156: What does “desired frequency” mean?

Line 203:  Sorry if I missed it, but I think this is the first time you mention “time-average”.  Please elaborate.

Figure 3:  Suggested adding a 3^rd row that presents a difference map.

Line 230:  Please identify from where these data were acquired.

Line 242:  Mentioning sun glint is important, thank you.  Consider elaborating on how omitting sun glint in this study might influence your results.  Given that sun glint is regional, I would expect meaningful impacts at low latitudes.  In practice for the future, you’d have to stratify your interpretation of results by sensor – SeaWiFS, e.g., tilts and saw far less sun glint than MODIS, which has complicated their intercomparison.

Section 2.4:  I appreciate the careful evaluations presented, but I was surprised that direct-ish comparisons of satellite-derived and ChlOSP-derived chlorophyll were not presented. More specifically, I don’t see much in the way of a performance assessment of the ChlOSP chlorophyll retrievals. Is there a reason for this?  Fig 11 touches on this, but in a highly averaged (20+ years) way that doesn’t really validate the ChlOSP chlorophyll retrieval.  Could some short-term temporal subset of model output be resampled in a way using distributions of valid MODIS retrievals to create map(s) that enable a semi-direct comparison of chl in common bins? I realize that it won’t be perfect (pre-industrial average vs. modern day MODIS), but at least seeing maps and their differences would provide the reader some knowledge of the scale of differences between the two chl retrievals.  Overall, I guess what I’m asking is, is there more than can be done to convince a reader that the ChlOSP chlorophyll retrievals are validated? 

Line 278:  Not obvious to me how Figure 6 shows anything “temporal”.  Would “long-term mean” be more appropriate wording?

Table 2:  Might be informative to expand this table to include specific basins or regions?

Line 318:  Re: “the regions that show” – maybe a map of sample sizes would be useful?