The manuscript investigates the spatial and temporal variations of ecosystem productivity sensitivity to the atmospheric CO2 growth rate (ACGR) across East Asia, utilizing the Global Carbon Budget and a subset of TRENDY v13 dynamic global vegetation models. While evaluating the regional divergence of carbon cycle feedback is an important pursuit, the manuscript currently suffers from critical methodological flaws. Specifically, concerns regarding the statistical definition of "sensitivity," unaddressed confounding climatic variables, and structural inconsistencies in the narrative must be thoroughly addressed. Extensive revisions are required to ensure the robustness of the proposed physical mechanisms.

Major comments

1. The entire study hinges on defining sensitivity as the slope of a simple 20-year moving window linear regression between detrended GPP/NEP and detrended global ACGR. This approach is fundamentally problematic. Interannual variations in global ACGR are strongly driven by large-scale climate modes (e.g., ENSO) and tropical temperature anomalies, which concurrently impact regional climate parameters (such as temperature, precipitation, and vapor pressure deficit) in East Asia. Performing a bivariate regression without controlling for these dominating climatic covariates inevitably captures confounding climate-driven correlations rather than a true physiological or ecological sensitivity to CO2. The authors must adopt a multivariate statistical framework or factorial simulations to isolate the CO2 effect from background interannual climate variability.
2. The manuscript does not clearly provide the exact regression formula, the units of the sensitivity, the detailed detrending procedure used in the main Methods, or how significance was assessed. At minimum, the authors should report the full equations and units explicitly.
3. The manuscript's focus shifts abruptly between NEP, GPP, and Ra. The results initially focus on NEP to establish the baseline sensitivity , but Section 3.2 suddenly transitions to analyzing GPP, justifying this by stating GPP is more suitable for assessing changes in ecosystem productivity. This rationale is post-hoc and fragments the logical flow. The manuscript should clearly establish its primary target metric in the introduction and methods, and maintain a consistent analytical thread throughout.
4. In Section 3.3, the text abruptly diverges into an analysis of precipitation and cloud cover trends to explain variations in Photosynthetically Active Radiation (PAR). While PAR undeniably influences GPP, this sudden meteorological attribution distracts from the core focus of the paper. If the authors wish to retain this section, they must integrate it more cohesively, demonstrating mathematically how cloud cover-induced PAR anomalies specifically modulate the calculated CO2 sensitivity. Additionally, section 4 (Discussion) is conspicuously brief and fails to adequately synthesize or contextualize the study's findings within the broader scientific literature. Currently, it merely presents an additional result regarding land cover distribution (MODIS IGBP classification) to tentatively explain regional differences. A robust discussion should delve deeply into the results, rather than included a new finding.

Minor comments

Lines 54-59. The definition of ACGR needs clarification. The text describes ACGR as the annual change in atmospheric CO2 concentration, but the unit is mass-based unit rather than a concentration-based unit.

Lines 88-94. The exact regression framework is under-specified. Please provide the explicit regression equation used to define "sensitivity", the units of the regression slope, and how significance was assessed for each moving window.

Lines 105-106 and 245-267. PAR is analyzed using "actual PAR values instead of ACGR sensitivity", whereas the other axes in Figs. 4-6 are sensitivity metrics. This mixes variables of different statistical meaning.

Lines 110-112. The sentence over-interprets the sign of the fitted slope. A negative NEP sensitivity does not, by itself, demonstrate a weakening of ecosystem carbon uptake ability.

Lines 112-117. The model-selection criterion needs stronger justification. Retaining models based on sign consistency or significance only in the last 20-year window appears rather ad hoc, especially because the subsequent analyses cover the full 1959-2023 period.

Lines 121-127. The regional partition requires a stronger justification. The text itself acknowledges that the GPP sensitivity pattern does not perfectly align with the East Asian monsoon boundary, yet the main comparison is built on that mask. A sensitivity test with an alternative regional definition would strengthen the analysis.

Lines 140-146. Expressions such as "downward trend in negative NEP sensitivity" and "reduction in negative GPP sensitivity" are sign-ambiguous and hard to interpret. Please rewrite these passages using clearer language, for example by stating explicitly whether the values are becoming more negative or moving towards zero.

Lines 156-158. The opening sentence of Section 3.2 is grammatically incomplete and conceptually abrupt.

Lines 170-174. There are both language and presentation issues here.

Lines 190-191. There is a clear inconsistency between the text and Fig. 4f. The manuscript reports r = 0.66, but Fig. 4f shows a negative relationship and labels it as r = -0.66. Please correct either the text or the figure.

Lines 239-242. The causal language is too strong. A coefficient of determination from a bivariate relationship is not sufficient to conclude that PAR is "the primary factor" weakening GPP sensitivity.

Lines 250-264. Trend slopes for precipitation, cloud cover, and PAR are reported without units.

Lines 270-281. The land-cover discussion has a temporal mismatch. The manuscript uses mean MODIS land cover from 2001-2023 to interpret sensitivity behaviour over 1959-2023. This should be acknowledged more explicitly as a limitation.

Lines 295-306. The conclusion again goes beyond what the presented metric can directly support.

The study examines the sensitivity of GPP and NEP parameters to fluctuations in the rate of increase in global CO₂ concentrations in the East Asian region, aiming to improve the accuracy of predictive dynamic models of atmospheric CO₂ concentrations that account for carbon cycle feedbacks. While the authors acknowledge the significant influence of regional factors on carbon balance, these factors were not integrated into the methodology used to assess the sensitivity of ecosystem productivity. The methodology section requires significant expansion. Currently, there is insufficient justification for the chosen approach and insufficient detail on the specific data processing steps.

Major comments:

1. From the paper, it is unclear which detrending methods were applied to each parameter explored. It is important due to the different robustness of the existing methods and, therefore, for the accuracy of the final results.

2. Using a 20-year sliding window can introduce significant distortions in the final result due to potential cyclic fluctuations or extreme events. Therefore, it is necessary to either select a shorter window - with the length justified in advance based on additional analysis - or use other methods to calculate interannual differences.

3. The behaviour of the data presented in Fig. 4 (specifically b, c, f, and, in particular, e) suggests that the used linear approximations (likely based on linear correlation analysis) are incorrect. An exploratory data analysis is required before linear correlation can be applied. Furthermore, transitions between samples in the year-ordered sequence demonstrate different directions across various time periods. This factor also calls into question the applicability of linear regression analysis.

4. The correlation analysis between pairs of sensitivities used in Section 3.2 raises doubts about its advisability from a mathematical standpoint, since ultimately the dACGR terms drop out of the ratio, leaving only the relation of the incrementAsians of the studied parameters, for example, [dGPP/dACGR] <-> [dSoilMoisture/dACGR] => dGPP <-> dSoilMoisture (Fig. 4.a,b,e,f), i.e., it will not affect the correlation.

5. The analysis presented in this study does not account for factors related to ecosystem evolution, assuming that the ecosystem of the selected area is stable, even though ecosystems can both develop and degrade over such a long period, which in turn leads to changes in productivity characteristics.

Minor comments:

1. Table 1 lacks relevant study parameters. Consider moving it to the appendix and updating it with a detailed description of DVGM's characteristics pertinent to the study.

2. Lines 83-85: It is stated that the spatial resolution of the Land Cover data has been changed to 0.5° × 0.5°, whereas the rest of the data uses 1.0° × 1.0°. The use of a different resolution requires an explanation of why this was done and how it affects the estimation methods. It is also noted that when changing the resolution, the most frequent land cover type is used; however, no estimates or background information are provided for this concept.

3. Lines 88-89: For their research, the authors adopted the methodology of Li et al. (2024). It is valuable to include this work in the outlook, explain the disadvantages of the existing approach, and clarify the relevance and sense of the proposed improvements.

4. Fig. 1: Correct dAGR to dACGR.

5. Lines 110-112: It is stated that the NEP sensitivity is examined for each DGVM model over the period 1959-2023; however, the results presented (Fig. 2a) show only a single value per model rather than a time series. There is no explanation of how the presented values were obtained.

6. Lines 133-134: It is stated that spatial analysis uses average values of GPP sensitivity to increases in CO₂ concentration. Given the dynamics of the characteristics under study over the selected period, using an average value for the assessment is questionable. A justification should be provided, or the possibility of using secondary parameters should be analysed.

7. Fig. 4-6: All captions include "***p<0.01; **p<0.05; *p<0.1; not significant p>0.1" part, but each figure does not include data relevant to all the probabilities.

8. Line 192: the sign of the correlation coefficient in the text does not correspond to the one in Fig. 4f.