|The manuscript is much improved in this revised version. The authors are thanked for their careful attention to my first set of comments. I am satisfied with almost all of their revisions.|
However, I still feel that the treatment of soil respiratory fractionation in the model is insufficient to support the primary conclusion of the study, that rainfall d18O variation is the main driver of the observed d18Oatm variation during Heinrich Stadials. There may be no easy solution to this dilemma, but the authors should at least temper their conclusion to acknowledge that they have not successfully treated the soil respiration part of the problem, and therefore have added some (unavoidable) uncertainty to their conclusion. But indeed there might be a solution, as discussed below.
The essential problem is that they use a fixed value of discrimination, 10.1‰, for tropical soil. However, it is well known that when soils dry out, they become well- aerated and therefore have much stronger discrimination (Angert et al.). This drying out of soils is exactly what happened during Heinrich Stadials. So the very heart of the problem that the authors set out to address, namely the cause of the enrichment in d18Oatm during Heinrich Stadials, is not really treated in their present version of the model with its fixed value of discrimination.
The changing water content of tropical soils during Heinrich Stadials is well documented by various lines of evidence. For one, the mixing ratio and isotopic composition of atmospheric nitrous oxide, a gas that is largely produced in waterlogged tropical soils, shows that tropical soils generally became less waterlogged during Heinrich Stadials, as inferred from the lower atmospheric nitrous oxide concentration and its isotopic composition during Heinrich Stadial 1 (Schilt et al., Nature 2014). Note that because the atmosphere is well-mixed, the nitrous oxide concentration not only records the northern hemisphere soil wetness but the effective global value, indeed as does d18Oatm.
Perhaps the authors could run a sensitivity experiment to evaluate the uncertainty in their conclusion that arises from neglect of soil respiratory fractionation change. This could be done by scaling the fractionation factor between values of 10‰ and 20‰ based on the amount of rainfall that the model produces in the tropical regions, with 20‰ corresponding to a well-aerated soil. The soil aeration could be quantitatively related to the model’s rainfall amount and land surface slope by using existing parameterizations employed in models that predict nitrous oxide production (e.g. the model used in Schilt et al., LPX-Bern, which is a state-of-the-art bottom-up dynamic global vegetation and land surface process model*).
*Stocker, B.D. et al. Multiple greenhouse-gas feedbacks from the land biosphere under future climate change scenarios. Nature Clim. Change 3, 666–672 (2013).
This sensitivity experiment with variable fractionation factor would greatly increase the usefulness of the manuscript by showing the magnitude of the uncertainty in the final conclusion that is introduced by using a fixed fractionation factor for tropical soils.
The authors could take this one step further and check their tropical soil wetness parameterization by seeing if their model can reproduce the observed nitrous oxide concentrations known from ice core records. This could be done, for example, by including in the model the nitrous oxide production parameterization from the LPX-Bern model. [Ultimately, it would be desirable to develop a nitrous oxide-d18Oatm relationship from observations of soil gases, but this is beyond the scope of the present manuscript.]
In summary, the value that the authors find for the change in respiratory fractionation during HS (-0.03) is almost certainly incorrect and the sign is wrong. There must be a positive value for this change due to lower soil moisture, as attested to by the lower values of atmospheric nitrous oxide and other paleohumidity indicators such as lake levels and cave records from the low-latitudes when weighted by the greater land area in the northern vs southern tropics. The authors can correct this problem by allowing their 10‰ fixed fractionation factor to vary as a function of soil moisture.
On page 25 it is stated:
Summarizing, in high latitudes, cold temperatures lead to a weak photorespiration but strong soil isotope fractionation. In low latitudes, high temperatures and variable C4 fraction lead to a weak soil isotope fractionation and a highly variable photorespiration. This compensatory mechanism between photorespiration and soil respiration fractionations explains thus the minor role of respiration in δ18Oterr anomaly during HS
This last sentence is probably incorrect, in the sense that low latitude photosynthetic oxygen production is much greater than high latitude oxygen production. Because total respiration of oxygen is tied closely to total production of oxygen, the same dominance of the low latitudes must be true for respiratory oxygen consumption. Therefore, this compensation between high and low latitudes cannot be very effective, since the low latitudes dominate. Instead, the minor role of respiration that the authors find is probably a model limitation. The wording should be corrected accordingly.
In Fig, 1, the Siple Dome record (shown in red) includes some data points at 15 ka that we know to be artifacts due to firn disruption and consequent thermal fractionation that occurred at this time (see Supplementary Materials in Severinghaus et al., 2009). We downweighted these points in our curve fitting exercise (see Supplement, Fig. S8). It would be helpful if you could do something similar to your Fig. 1, to avoid misleading the reader. You could leave these artifacts in the plot, for example, but only as points rather than drawing lines between them. The same goes for the points at 35 ka. Alternatively, you could show our curve fit rather than the original data points. This is available on the online data repository.