The article presents a novel method for investigating the driving factors of ecosystem evaporative stress—an area that has already been extensively studied—using multi-source remote sensing data. This method incorporates the self-enhancement mechanism of drought events to achieve more accurate sensitivity estimations. Overall, the paper is well written, with a clearly presented theme, meticulous attention to detail, rigorous conclusions, and notably indicative results.

Main comments

The manuscript is overly verbose in some sections. For example, the detailed description of data processing in Section 2 ("Materials and Methods") is unnecessarily elaborate; well-known formulas should be omitted or simplified. I argue that, under climate change, global ecosystems will face increasingly severe water scarcity, with soil moisture deficit being a key factor affecting land–atmosphere feedback. While this observation supports the authors' findings, the portrayal of the ESI as representing the "balance" of the water–vegetation–atmosphere system is overemphasized. Although the balance metaphor is vivid, it is not clearly defined; the ESI should be presented modestly as an indicator of ecosystem water deficit.

The manuscript validates the proposed method by comparing results from traditional approaches and different temporal resolutions. It is a solid study. My primary concern is that the MDLM model relies on linear assumptions to quantify sensitivity, despite the results showing time-varying sensitivity in the real world. It remains unclear how these findings might inform improvements in Earth System Models, which typically involve complex, nonlinear interactions. Therefore, the authors should discuss whether alternative nonlinear sensitivity models have been considered or acknowledge the limitations of the linear approach. I suggest that, when highlighting the significance of the study, the phrase "providing a reference for improving Earth system model parameters" be replaced with a more objective description emphasizing the "revelation of time-varying effects."

Minor comments

Line 55: It is recommended to clarify that the ETₚ calculation should more accurately be described as reference crop evapotranspiration (ET₀), although the two are often used interchangeably.

Lines 72, 76, 78, 100: The punctuation is not standardized; for example, a space should be inserted after parentheses in citations. A thorough check throughout the manuscript is advised.

Line 89: Replace “variations” with “differences” in the phrase “Variations in regional…” to more accurately convey the intended meaning.

Line 116: The term “high-frequency analysis” is unclear; please either provide clarification or remove the term.

Lines 172–185:（1）Is the formula accounting for the CO₂ water-saving effect widely accepted? Are there studies that validate its accuracy?（2）Are subsequent analyses based entirely on the CO₂-adjusted ESI? This comparison is only made in Section 3.1; clarification in other sections is needed.

Line 245: The citation “work of Yanlan Liu, (2019)” requires correction in its formatting.

Line 250: Some border lines in Figure 3 impair readability; please check and remove them as needed.

Line 317: The phrase “consistent with Liebig's law of the minimum” should be accompanied by an appropriate reference.

Line 345:Why soil moisture in the 100–289 cm depth range shows higher relative importance compared to other depths in Figure 6？

Lines 360–365: Attributing the soil moisture sensitivity peak in 2016 solely to extreme climatic conditions may be one factor, but I believe it is unlikely to be the primary reason. Many drought indicators exhibit a unimodal pattern peaking around 2010, suggesting that complex interactions among various ecosystem variables may be involved.

Line 366: The reported standard deviation in “…with an estimated 4.43% (±6.88%) increase…” is notably large, even exceeding the mean. Please explain this occurrence.

Line 410: In Figure 8g, why does southern China (e.g., the lower Yangtze region) appear magenta, and why does evaporative stress increase despite an increase in LAI? A detailed explanation is required.

General Comment on Figures: The text in the figures is not sufficiently clear. It is strongly recommended to increase the resolution of all figures to improve readability.

The authors examine the monotonic trend of evapotranspiration stress (measured by ESI). They investigate the sensitivity of ESI and its changes to the hydrological, atmospheric, and vegetation variables across China. They employ explainable machine learning methods to discern key variables that determine the ESI. They then establish another novel regression model with time varying parameters to investigate the sensitivity of the controlling factors. The authors find a predominant role of soil moisture to ESI changes. The feedback loop of the water-vegetation-atmosphere the authors discussed is interesting. The authors did extensive data-driven analyses, and their findings could inform the land use policies. I am generally supportive of publishing the paper, once below comments about the wording, methods, mapping and structure are clarified or addressed.

1)    Title Clarity: The title may be misleading. While the authors emphasize the increasing sensitivity of ESI to soil moisture, it is important to note that both the sensitivity (\theta svm) and the actual soil moisture (\delta svm) are changing and contributing to ESI changes (Figures 5-6). It is unclear whether the monotonic trend (\delta change) or the variability changes (\theta change) are more dominant in driving ESI changes. Therefore, the title “Increasing sensitivity … predominantly intensifies …” might be misleading.

2)    Temporal Resolution Consistency: The connection between Section 3.1 and other sections appears weak. Figures 4a and 4c depict the ESI trend, but it is unclear whether this represents the annual mean or another temporal scale. The regression models (RF, XGBoost, MDLM) are built at monthly and sub-monthly scales. The authors need to reconcile these inconsistencies by clarifying the temporal resolutions of the ESI and sensitivity trends.

3)    Model Comparison: The authors claim that the MDLM model is novel and superior to the MLR method based on inconsistent results (Lines 393-394) and higher R² values (Line 490). However, an in-depth comparison of the sensitivity and its changing patterns derived from both methods is lacking. Additionally, the authors should specify the names of the “two distinct methodologies” mentioned in Line 394.

4)    Materials and Methods Structure: The Materials and Methods section is overwhelming. The subsection 2.1.1 “Study Area” could be moved to the Introduction section as background information. Given the extensive list of datasets and variables used, with various spatiotemporal scales, sources, and pre-processing procedures, I suggest the authors compile this information into tables for better readability.

5)    Map Issues: The gridline spacing in Figures 2b and 2c should be consistent. I recommend using a narrower legend and specifying kilometers as the unit for clarity.

6)    Model Performance Metrics: The performance metrics (e.g., R², RMSE) for the RF regressor and XGBoost model are not provided. These metrics are crucial for evaluating model performance. Additionally, the rationale for choosing RF over XGBoost should be explained.

7)    Multicollinearity and SHAP Values: Despite efforts to address multicollinearity, the SHAP values for soil moisture appear contradictory to precipitation (Figure 5). Given that these variables are generally positively correlated, the authors should explain this discrepancy.

8)    Group Differences: Figure 5 groups samples into cropland, forest, and grassland, but differences among these groups are not adequately discussed. For instance, grassland shows different behavior compared to cropland and forest for svm0-7. The authors should explore the possible physical mechanisms behind these differences.

9)    Soil Moisture Sensitivity Changes: The reasoning behind the changes in soil moisture sensitivity is not convincing. The authors attribute these changes to anomalies in the year 2016 (Lines 362-365). It is unclear how a single-year anomaly can explain long-term changes.

10)     Key Findings Summary: The key points of the paper are not clearly summarized. I recommend the authors provide a concise summary of their key findings at the end of the paper to enhance clarity.