The authors of "Strong relation between atmospheric CO₂ growth rate and terrestrial water storage in tropical forests on interannual timescales" have provided a very thorough and comprehensive analysis of interannual variability of terrestrial water storage and its link to the atmospheric growth rate of CO₂. This research adds to our knowledge of interannual variability of land carbon fluxes and opens new pathways for discovery in this direction. The results are interesting and the research should be published with some modifications. However, I would really implore the authors to deepen their discussion, which I will elaborate below. Additionally, I urge the authors to use a broader spectrum of inverse models. Some other, smaller comments are provided after.

General comments

Throughout the discussion, the authors discuss the findings, but fail to provide some deeper context. In my opinion, the interesting findings of this manuscript should be placed in context better. This would improve the impact of this manuscript, as it allows for follow-up research. I would urge the authors to focus more on the (possible) mechanisms that drive the findings in this manuscript.

For example, the authors do not mention why the tropics account for such a large portion of the covariance. Is this because the IAV in the tropics outweighs IAV in temperate regions (not according to the inversions)? Or is this because droughts in the tropics (i.e. the ENSO cycle) covers the entire tropics (whereas droughts temperate regions are more driven by synoptic variability and thus happen over smaller scales). Additionally, the authors should mention why TWS is a better explanatory factor than e.g. VPD or temperature. In L.616, the covariance between temperature and water availability is mentioned but not discussed sufficiently. 

Finally, the positive correlation between temperate TWS and carbon growth rate should be mentioned, regardless of the small contribution to the growth rate. Can this be explained physically? 

Additionally, the analyses should be done with more than four inverse models. The GCB inverse fluxes are made publicly available (https://meta.icos-cp.eu/objects/FHbD8OTgCb7Tlvs99lUDApO0 for GCB2023 and more recently https://meta.icos-cp.eu/objects/GpFcABoKcZMVnRUlLHRInhdM for GCB2024). Some of these models indeed only cover the OCO2 period (2015 onwards), but for GCB2023 and 2024, 8 systems with sufficient data are provided. Therefore, I expect the analyses to be done with all available models.

Technical comments

L.75: A reader might find it strange that the first sentence of the plain language summary says the CO2 increases every year, but reads here that there is a decrease (which I understand is in the growth rate, so there could still be an increase). I would recommend to maybe rephrase to larger and smaller growth rates with El Nino/La Nina

L.96: "Most vegetation responds to soil moisture". Yes, but also to VPD (which is not included in TWS). These effects are difficult (if not impossible) to disentangle, but interesting to mention.

L.106: I think your definition of NBE is the same as the definition of NEE used here. It could be easier for a general reader to use only one of the two.

L.237: Linear detrending removes the fossil fuel trend, but also any trend in biogenic and ocean fluxes. This should be mentioned

L.238: Why are the CGR and TWS smoothed using different windows? To me it makes intuitive sense to use the same window

Fig 2: Here, the growth rate can become negative (so the first sentence of the introduction is not true, or is this detrended growth rate?)

Fig 3: The positive correlations in south Brazil and east China are quite interesting, and could be explained biophysically. This links to my first general comment as well.

L.305: It would be nice to add a horizontal line showing the global correlation to make it more explicit that the tropics can indeed explain all the correlation

L.432: It's interesting that NISMON has a larger range, and I recommend the authors to (shortly) discuss any potential reason for this (e.g. prior model, observations used, transport model). The inclusion of other models might help in this.

L.537: I'm not sure I understand the phrasing 'still a key factor'. Is this in contrast to other studies that pointed towards temperature? Otherwise just remove the 'still'

L.624: I disagree with the statement that atmospheric inverse models are "specifically designed to regionalise carbon fluxes". Atmospheric inversions cannot distinguish CO2 from different regions, and mainly constrain the atmospheric carbon budget. Especially observation-sparse regions like the tropics are hard to constrain by atmospheric inversions (even satellite-based inversions). This does not mean that inversions cannot be used to analyse regional impacts -- I think your use of multiple models is suitable as it covers a range of flux realisations. And although I appreciate the downtuning later in this paragraph, I think this statement needs refinement.

This manuscript presents a comprehensive assessment of the relationship between atmospheric CO₂ growth rate (CGR) and terrestrial water storage (TWS) at interannual timescales, with a particular focus on tropical forest ecosystems. By combining satellite-derived GRACE data, multiple atmospheric CO₂ inversion products, and land cover datasets, the authors demonstrate a consistent and statistically significant negative correlation (r = -0.70) between global TWS and CGR from 2002 to 2023. Notably, tropical America and tropical forests emerge as dominant contributors to this coupling, despite their relatively small spatial extent. The study also applies multiple methods to partition the spatial and functional drivers of this relationship and evaluates the robustness of the signal using both observational and model-based approaches.

Overall, this study is of high quality and reflects substantial analytical and conceptual work. The manuscript is well structured and clearly written. I have several comments and suggestions for the authors to consider:

1. The abstract reports a strong TWS–CGR correlation, but it lacks a statement explaining how this finding advances previous work (e.g., Humphrey et al., 2018). Please clarify further how this study uniquely extends or deepens our understanding.

1. Lines 28-31: The sentence “tropical forests exhibit the strongest CGR correlations” is important but could briefly explain why—e.g., due to high productivity sensitivity to water stress—so help readers understand the physiological context.

1. In Section 2.3, clarify whether all four inversion products use harmonized fossil fuel and biomass burning emissions (e.g., GFED versions). Differences in fire emissions datasets could bias regional flux attribution.

1. Table 1 summarizes inversion methods, but the main text should include 2–3 sentences interpreting key differences in transport models, meteorological fields, or prior flux assumptions and how they might affect tropical vs. extratropical estimates.

1. I don't think it's necessary to place Figure 1 in the main text. It is recommended to put it in the supplementary files. Overall, there are too many figures in the full text. It is suggested to combine them.

1. Lines 293-294: Please explain why some regions with high local correlation contribute minimally to the global signal—e.g., due to small TWS variance or maybe small spatial extent—and explicitly state how these cases are handled.

1. In Figure 4, report the number of grid cells per land cover class and provide standard deviations or interquartile ranges to contextualize variability in contribution estimates.

1. Annotate key ENSO or drought years (e.g., 2005, 2010, 2015–16) directly in Figure 5 to aid interpretation of CGR–TWS relationships and align with the narrative.

1. lines 390–395: reference a figure or appendix that visualizes cross-regional TWS anomaly compensation (e.g., a correlation matrix or spatial covariance map), supporting the claim of cancellation effects in croplands.

1. The sensitivity analysis in Figure 10 is valuable, but the ecological interpretation of why tropical forests show both high correlation and high sensitivity should be more deeply discussed—e.g., in terms of water-use efficiency or rooting depth.

1. lines 530–534: Clarify whether TWS–CGR correlations were adjusted for or confounded by co-varying climate factors (e.g., temperature, VPD). If not adjusted, include a cautionary note on potential indirect effects.

1. lines 543–548: In discussing cases where TWS is weakly correlated with CGR (e.g., tropical Africa in 2016), could consider fire activity, radiation, or phenological anomalies as alternative drivers.

1. In the conclusion, clearly articulate how your findings can inform terrestrial biosphere model development. For example, suggest that ecosystem models should incorporate regional water constraints with higher fidelity, particularly in tropical forests.

“Strong relation between atmospheric CO2 growth rate and terrestrial water storage in tropical forests on interannual timescales” by Petch at al. analyzes CO2 fluxes derived from atmospheric inversions in conjunction with space-based terrestrial water (TWS) to understand what regions drive the accumulation of CO2 in the atmosphere (CGR). Past studies had shown a strong correlation between the atmospheric CGR and global TWS, suggesting that ecosystem water availability is the dominant mediator of the strength of the terrestrial carbon sink and thus TWS. This paper expands on this analysis to show that tropical forests, followed by semi-arid ecosystems, are the drivers of this strong correlation, with tropical America contributing almost 70% of the global TWS-CGR correlation. The manuscript provides some improvement on analysis of globally integrated covariates by elucidating the geographic and ecosystem-type drivers.

A major concern with this analysis is lack of discussion of northern extratropics. Previous analysis has shown that these areas are important contributors to CO2 IAV (Guerlet et al., 2013 – who showed that inversions using in situ CO2 only did not see the flux variability that was visible from space-based data; Keppel-Aleks, et al., 2014 – who showed that taking annual averages of CO2 data, which is what is done to calculate CGR in this paper – masked variations that were attributable to northern extratropics). I point out these two papers in particular because the authors use inversions driven by surface CO2 rather than satellite CO2, and because their global correlations are derived from global averaged CO2. Given the four inverse models show a huge spread in the northern extratropical contribution to CO2 IAV, with NISMON showing these areas contribute only 5% to global CO2 IAV and CT2022 showing contributions closer to 35% of global CO2 IAV, more analysis of the impact of this discrepancy is required.

A second concern is that tropical Africa contributes almost as much CO2 IAV as tropical America, but this variability is not as well correlated with variations in TWS.  What, then, does this mean for the utility of the high correlation between CGR and TWS? The authors diagnose the regions that drive the apparent global correlation with TWS, but they do not provide much analysis or any conclusions about in what way the emergent global correlation can or should be used in carbon cycle research.

I am also somewhat skeptical how robust are the results from these four inverse models. The models do show substantially different regional fluxes, by at least a factor of two for each of the geographic regions considered. More inverse models are available for analysis, and I am curious about what a larger ensemble would reveal. Section 2.5 stated that correlations were calculated from an ensemble mean flux product. I am curious as to what the correlations with TWS variations would look like from each model separately. Would the conclusions be robust to this change in methodology?

Analysis from the OCO fluxMIP shows significant distinctions between inversions that are configured to adhere tightly to the ocean prior and those where the ocean can move. Are the differences in terrestrial fluxes within this small 4 member ensemble associated with tight ocean priors? And related to a point above, what would the analysis look like based on inversions that use space-based CO2 observations to constrain fluxes? In theory, the space-based data provides better spatial constraints than the sparse surface network.

In summary, I recommend more discussion and stronger conclusions about what it means that many regions that contribute a fair amount of CO2 variability do not have a strong correlation with TWS, and other regions that have strong TWS variability do not show much variation in CO2 flux. The paper would have more impact if it provided recommendations about what key carbon cycle inferences are being obscured by using the global relationship between CGR and TWS as a shortcut.

Guerlet, S., Basu, S., Butz, A., Krol, M., Hahne, P., Houweling, S., et al. (2013). Reduced carbon uptake during the 2010 Northern Hemisphere summer from GOSAT. Geophys. Res. Lett., 40(10), 2378–2383. https://doi.org/10.1002/grl.50402

Keppel-Aleks, G., Wolf, A. S., Mu, M., Doney, S. C., Morton, D. C., Kasibhatla, P. S., et al. (2014). Separating the influence of temperature, drought, and fire on interannual variability in atmospheric CO2. Global Biogeochem. Cycles, 28(11), 1295–1310. https://doi.org/10.1002/2014GB004890