Steinkamp et al. have applied major changes to their manuscript "Atmospheric CO2 observations and models suggest strong carbon uptake by forests in New Zealand". I am not entirely satisfied with the answers they gave to my comments but I acknowledge that given the interesting material brought by this study, some of these comments could sound like a bit too meticulous. My main remaining issue is that I feel that the text (including the new and revised parts) is still often confusing and it often lacks of structure and rigor. I also feel that the authors sometimes provide weak answers in order to avoid accounting for my comments and to avoid new analysis and discussions (in particular new analysis of the modeled vs measured concentration timeseries). Finally, a significant number of my comments (either comprised within my major comments or in my list of minor comments) were misunderstood. In conclusion I would still push for a major and general improvement of the text but the scientific issues I would still raise are minor.
Here is a list of specific problems I see in their answers and in the corresponding corrections to the manuscript:
- Regarding the test of sensitivity to the NEE diurnal cycle: the most straightforward test we could have expected was a new experiment using the real data and the reference inversion set-up, but the prior estimate of the NEE with diurnal variations. Comparing the reference inversion to this one would have provided a clear characterization of the impact of this diurnal cycle in the reference inversion results, and an evaluation of the confidence in these results. I do not say that the authors had to make this specific experiment, but their OSSE and of their analysis to make a demonstration of the weak impact of the diurnal cycle can sometimes be a bit puzzling:
* I am not 100% sure about the way they derived the diurnal cycle of the synthetic NEE: "the diurnal variation in GPP is based on hourly solar insolation": does it mean that it is exactly set-up based on the relative diurnal variations of the solar insolation ? I cannot see numbers illustrating the amplitude of this cycle (so it is difficult to check that it is "exaggerates" actual diurnal variations).
* the authors connect this OSSE to the discussion on the model-data mismatch at LAU for 15:00-16:00, but in order to assess whether uncertainties in the NEE diurnal cycle impacts these mismatch, they could had looked at the 15:00-16:00 concentrations modeled when using the flat NEE vs. the one with diurnal variations (instead of trying to read such a comparison out of the inversion results). As a consequence, they assimilate 15:00-16:00 data in the OSSE instead of both 13:00-14:00 and 15:00-16:00 as in the reference experiment. This and having several other parameters that are different in the OSSE compared to the reference inversion does not help evaluate the robustness of the reference inversion with regard to uncertainties in the NEE diurnal cycle.
* the analysis added in 7.4 to comment on the impact of uncertainties in the diurnal cycle based on Figure 14 are often confusing but yes, Figure 14 seem to indicate that this impact is small at the annual scale. The much larger impact on results at the monthly scale and for the seasonal cycle could have been more emphasized (while the claim by the author that the truth and the inversion agrees within their uncertainty is misleading). At least, for regions 7, 12, 14 and 15 it is clearly significant and often larger than the other sources of uncertainties accounted for in the estimate of the posterior uncertainty, which shows that this uncertainty should be increased to account for this additional source of error at the monthly scale.
- Regarding the need for a general improvement of the text: as I explicitly said, I only listed examples illustrating such a need in my previous review, but I did not conduct an exhaustive listing of the problems. However, the authors "sought to improve the quality of the text in places specifically pointed out by the referee" when they did not just rebut the corresponding concerns. As a consequence, they did not conduct the detailed proofreading which was needed, and which is thus still needed.
- I am still a bit puzzled regarding the topic of the model error. I am really confused by the authors' explanations regarding the precise computation of the model error in page 16-17 and how it can be consistent with what is said in page 18 regarding the 0.4ppm value. The text does not give the typical value for the model error arising from such computations. Furthermore, the authors refused to conduct some more detailed analysis of the prior and posterior model vs measured concentration timeseries (which would have been useful to discuss the theoretical value that they derive for the model error), and more generally to conduct some additional evaluation of the transport model, assuming that "a more in depth analysis of transport model bias is outside the scope of this paper".
- My comment on the weakness of the configuration of the sensitivity test has not really been accounted for even though these tests are used to provide an assessment of the robustness of the inversion results
- Some technical points:
* the mathematical notations are still problematic, e.g. Tg and T are multiplied by the same vector on page 9 or the dimensions of x and T change from page 9 to 17
* I do not understand the justification for the S term in equation (2) based on a matter of "transparency".
* I think that the authors misunderstood my point regarding their estimate of the representativeness error. I just think that strictly speaking, the 5-min scale variability of the measurements does not reflect the spatial variability of the hourly mean measurements in model box.
* Regarding the meaning of the correlations between posterior uncertainties in two flux components: the author insist in writing something wrong i.e. "strong negative correlations between two regions would indicate that the inversion has difficulties to distinguish their individual flux components with the available data". Again, if the uncertainty reduction for the flux in each region is about 99% but the remaining (posterior) insignificant uncertainties in each region have a -0.9 correlation, we would still have to say that the inversion managed to distinguish their individual flux components.
* in general: the authors easily use the term "bias" when discussing random or varying errors |