Drivers of Ecosystem Stability Differ with the Intensity of Extreme Climatic Events

Yanagawa, Aki; Ueda, Rina; Yoshikawa, Sayaka; Iseri, Yoshihiko; Kanae, Shinjiro

doi:10.5194/egusphere-2026-1007

Preprints

https://doi.org/10.5194/egusphere-2026-1007

Preprints

06 Mar 2026

| 06 Mar 2026

Drivers of Ecosystem Stability Differ with the Intensity of Extreme Climatic Events

Aki Yanagawa, Rina Ueda, Sayaka Yoshikawa, Yoshihiko Iseri, and Shinjiro Kanae

Abstract. This study investigates how the dominant drivers of ecosystem stability metrics vary across gradients of hydroclimatic extremity. While previous studies have documented the impacts of droughts and heavy rainfall on ecosystem functioning and resilience inferred from stochastic fluctuations, less attention has been given to whether the relative importance of climatic, biotic, and landscape controls changes systematically under different levels of climatic stress. To address this question, we quantified vegetation resistance and event-scale recovery responses and compared the contributions of meteorological, biodiversity, and topographic factors across a global range of hydroclimatic conditions. We find that under normal to moderately dry conditions, vegetation stability metrics are primarily associated with meteorological variables, particularly temperature and precipitation, consistent with earlier global assessments. Under severe and extreme drought conditions, resistance decreases markedly across most regions, whereas recovery responses exhibit weaker and more spatially heterogeneous changes. Importantly, in sparsely vegetated ecosystems such as grasslands and open shrublands, the relative dominance of drivers shifts from climatic to biodiversity and topographic factors under intensified drought stress, indicating context-dependent regulation of ecosystem stability. Deciduous needle-leaf forests show consistently low resistance and recovery capacity across climatic regimes, suggesting elevated sensitivity to hydroclimatic variability. Overall, our findings demonstrate that ecosystem stability under climatic extremes cannot be explained solely by meteorological forcing and highlight the increasing importance of biodiversity and landscape heterogeneity in shaping stability responses under intensifying climate variability.

Received: 23 Feb 2026 – Discussion started: 06 Mar 2026

Publisher's note: Copernicus Publications remains neutral with regard to jurisdictional claims made in the text, published maps, institutional affiliations, or any other geographical representation in this paper. While Copernicus Publications makes every effort to include appropriate place names, the final responsibility lies with the authors. Views expressed in the text are those of the authors and do not necessarily reflect the views of the publisher.

Download & links

Preprint (PDF, 1688 KB)

Supplement (2721 KB)

Download & links

Aki Yanagawa, Rina Ueda, Sayaka Yoshikawa, Yoshihiko Iseri, and Shinjiro Kanae

Status: final response (author comments only)

RC1:
'Comment on egusphere-2026-1007', Anonymous Referee #1, 17 Mar 2026

The manuscript has promise and asks an important question, but the current version needs more rigorous metric justification, more careful interpretation of biodiversity proxies and feature importance before its central conclusions can be considered robust.

Citation: https://doi.org/10.5194/egusphere-2026-1007-RC1
- AC1:
  'Reply on RC1', Aki Yanagawa, 18 Mar 2026
  
  We sincerely thank the reviewer for this constructive and thoughtful comment. We appreciate the reviewer’s positive assessment that the manuscript addresses an important question and has promise. We also understand the concern that the current version does not yet provide a sufficiently rigorous justification of the resistance metric and that the interpretations of biodiversity proxies and feature importance need to be made more cautiously in order to support the central conclusions more robustly.
  In revising the manuscript, we plan to address this concern in three main ways.
  First, we will strengthen the justification of the resistance metric by clarifying its conceptual basis, its relationship to previous studies, and the rationale for adopting this metric in the context of ecosystem responses to climatic stress. We will also make the assumptions and limitations of this metric more explicit.
  Second, we will revise the interpretation of biodiversity-related variables throughout the manuscript. In particular, we will clarify that these variables are proxies rather than direct measures of biodiversity, and we will discuss more explicitly the limitations of interpreting them as representing biodiversity effects.
  Third, we will revise the interpretation of feature importance so that it is presented more cautiously. We will clarify that feature importance indicates relative predictive contribution within the modeling framework and does not by itself demonstrate causal effects. Accordingly, we will reassess the wording of the central conclusions to ensure that the interpretation remains consistent with the scope of the data and methods.
  Overall, in the revised manuscript, we intend to improve the rigor of the methodological justification and to moderate the interpretation of the results so that the conclusions are more transparent and better supported.
  
  Citation: https://doi.org/10.5194/egusphere-2026-1007-AC1
  - AC2: 'Reply on AC1', Aki Yanagawa, 20 Mar 2026
    
    Regarding “strengthen the justification of the resistance metric,” revisions are highlighted in yellow; regarding “interpretation of biodiversity-related variables,” revisions are highlighted in green; and regarding “interpretation of feature importance,” revisions are highlighted in light blue.
    The first file is the originally submitted manuscript with the revised portions color-highlighted so that the changes can be easily identified.
    
    Citation: https://doi.org/10.5194/egusphere-2026-1007-AC2
    
    AC3: 'Reply on AC2', Aki Yanagawa, 20 Mar 2026
    
    The second file is the revised manuscript with the modified sections color-highlighted.
    
    Citation: https://doi.org/10.5194/egusphere-2026-1007-AC3
    
    AC4: 'Reply on AC3', Aki Yanagawa, 20 Mar 2026
    
    The third file is the clean revised manuscript in which all revisions have been incorporated and the color highlighting has been removed from the main text.
    We would appreciate it if you could kindly review these files.
    
    Citation: https://doi.org/10.5194/egusphere-2026-1007-AC4
RC2:
'Comment on egusphere-2026-1007', Anonymous Referee #2, 21 Apr 2026
General comments
This manuscript presents a global-scale assessment of ecosystem resistance and resilience to extreme weather events using remotely sensed vegetation dynamics (NDVI), climate characterization via Standardized Precipitation–Evapotranspiration Index, and a broad suite of climatic and environmental predictors. The use of LightGBM to integrate these multiple drivers and identify controls on ecosystem vulnerability is timely and relevant, particularly in the context of increasing climate extremes.
Overall, the study is well conceived, methodologically sound, and clearly written. The integration of diverse datasets at the global scale is a notable strength, and the results have the potential to provide meaningful insights for ecosystem management and climate adaptation strategies. The authors also demonstrate awareness of key methodological limitations.
However, some aspects of the conceptual framing and interpretation of results require further clarification. In particular, the discussion of ecosystem stability is somewhat narrow, and the implications of methodological assumptions - especially those related to temporal recovery dynamics and NDVI limitations -are not sufficiently explored. Addressing these points would significantly strengthen the robustness and interpretability of the findings.

Specific comments
1. Conceptual framing of ecosystem stability
The manuscript focuses on resistance and resilience as indicators of ecosystem stability. While these are widely used and relevant metrics, ecosystem stability is a broader concept that may also include dimensions such as temporal variability, persistence, recovery time, and potential regime shifts.
The authors are encouraged to explicitly acknowledge that their analysis captures only part of the stability framework. Clarifying this point would improve conceptual rigor and avoid potential overgeneralization of the results.
2. Temporal definition of resilience
Resilience is quantified using NDVI in the year following an extreme event relative to normal conditions. This definition implicitly assumes that ecosystem recovery occurs within a one-year timescale.
This assumption may not hold across all ecosystems. In many cases - particularly in forests, semi-arid systems, or under severe disturbances - recovery can take multiple years. Restricting the analysis to a single subsequent year may therefore:
Underestimate resilience in slow-recovering ecosystems

Introduce biases when comparing regions with different ecological recovery timescales

The authors should discuss this limitation more explicitly. At minimum, a justification for the choice of a one-year recovery window should be provided, along with a discussion of how longer recovery periods might influence the results.
3. Implications of NDVI limitations
The manuscript acknowledges known limitations of NDVI, such as saturation in high-biomass regions, but does not sufficiently discuss how these limitations may affect the findings.
NDVI saturation can reduce sensitivity to vegetation changes in dense canopies, potentially leading to:
Underestimation of resistance (i.e., smaller apparent declines during extreme events), and

Attenuation or distortion of resilience signals during recovery

This may introduce systematic biases in the spatial patterns of ecosystem vulnerability, particularly when comparing high-biomass ecosystems (e.g., tropical forests) with lower-biomass systems (e.g., grasslands).
The authors should expand the discussion to address:
The expected direction of these biases

The regions or ecosystem types most affected

The potential implications for the machine learning results (e.g., feature importance, predicted spatial patterns)

Even a qualitative assessment would substantially improve the interpretation of the results.

Technical corrections
Consider adding a short introductory paragraph at the beginning of each main section to guide the reader before presenting subsections.

Briefly clarify how the 5° × 5° grid aggregation may influence the interpretation of heterogeneous landscapes.

If not already included, add a short statement on the robustness of the machine learning model (e.g., sensitivity to hyperparameters or data partitioning).

A brief mention of alternative vegetation indices (e.g., EVI) could complement the discussion of NDVI limitations.
Citation: https://doi.org/10.5194/egusphere-2026-1007-RC2
- AC5:
  'Reply on RC2', Aki Yanagawa, 24 Apr 2026
  Response to Reviewer Comments
  Manuscript: egusphere-2026-1007
  We sincerely thank the reviewer for the careful reading of our manuscript and for the constructive and insightful comments. We appreciate the positive evaluation of the overall study design and the recognition of the global integration of multiple datasets. In the revised manuscript, we have addressed all points raised below. To improve clarity, the revised manuscript now more explicitly defines the conceptual scope of stability examined in this study, clarifies the rationale and limitations of the one-year recovery window, expands the discussion of NDVI-related limitations and their implications, corrects and clarifies the spatial harmonization description, adds section-opening guide paragraphs, and includes a more explicit statement on machine-learning robustness and future sensitivity analyses. In the marked manuscript, revised passages are shown in bold.
  General comment
  “This manuscript presents a global-scale assessment of ecosystem resistance and resilience to extreme weather events using remotely sensed vegetation dynamics (NDVI), climate characterization via Standardized Precipitation–Evapotranspiration Index, and a broad suite of climatic and environmental predictors. The use of LightGBM to integrate these multiple drivers and identify controls on ecosystem vulnerability is timely and relevant, particularly in the context of increasing climate extremes.
  
  Overall, the study is well conceived, methodologically sound, and clearly written. The integration of diverse datasets at the global scale is a notable strength, and the results have the potential to provide meaningful insights for ecosystem management and climate adaptation strategies. The authors also demonstrate awareness of key methodological limitations.”
  Response: We sincerely thank the reviewer for this positive assessment of the manuscript. We appreciate the recognition of the study’s conceptual relevance, the global integration of diverse datasets, and the value of the LightGBM framework for understanding ecosystem vulnerability under increasing climatic extremes. We have retained the core analytical framework of the study while revising the manuscript to improve conceptual clarity and to more fully address the limitations noted by the reviewer.
  General comment
  “However, some aspects of the conceptual framing and interpretation of results require further clarification. In particular, the discussion of ecosystem stability is somewhat narrow, and the implications of methodological assumptions - especially those related to temporal recovery dynamics and NDVI limitations - are not sufficiently explored. Addressing these points would significantly strengthen the robustness and interpretability of the findings.”
  Response: We agree with this important point and have revised the manuscript accordingly. Specifically, we now state more explicitly in the Abstract, Introduction, Discussion, and Conclusion that the present analysis addresses only part of the broader ecosystem stability framework, namely event-conditioned resistance and one-year recovery responses estimated from NDVI relative to normal climatic conditions.
  These terms were revised as follows:
  ecosystem stability -> event-conditioned ecosystem stability
  
  recovery -> one-year recovery or short-term recovery
  
  We also expanded the Discussion to address the implications of the one-year recovery assumption, the likely direction of bias caused by NDVI saturation in high-biomass systems, the ecosystem types most affected by this limitation, and the possible implications for feature importance and predicted spatial patterns.
  These terms were revised as follows:
  The fourth point discussed in Section 4.3 (Limitations)
  
  The fourth paragraph of the Conclusion
  
  These changes were made to strengthen both the conceptual framing and the interpretability of the results.
  Specific comment 1. Conceptual framing of ecosystem stability
  “The manuscript focuses on resistance and resilience as indicators of ecosystem stability. While these are widely used and relevant metrics, ecosystem stability is a broader concept that may also include dimensions such as temporal variability, persistence, recovery time, and potential regime shifts.
  
  The authors are encouraged to explicitly acknowledge that their analysis captures only part of the stability framework. Clarifying this point would improve conceptual rigor and avoid potential overgeneralization of the results.”
  Response: We appreciate this suggestion and agree that the original wording could imply a broader treatment of ecosystem stability than is actually provided by our metrics. In response, we revised the manuscript to state explicitly that ecosystem stability is a broader concept that may include temporal variability, persistence, recovery time, and regime shifts, whereas the present study focuses on two operational dimensions: event-year resistance and one-year recovery responses derived from NDVI. This clarification has been incorporated into the Introduction and reiterated in the Discussion and Conclusion. We believe that this revision substantially improves conceptual rigor and reduces the risk of overgeneralization.
  These terms were revised as follows:
  The first paragraph of the Introduction (However, ecosystem…)
  
  The last sentence of Section 4.2, “Feature Importance of Resistance and Resilience Across Climatic Gradients”
  
  The third paragraph of the Conclusion
  
  Specific comment 2. Temporal definition of resilience
  “Resilience is quantified using NDVI in the year following an extreme event relative to normal conditions. This definition implicitly assumes that ecosystem recovery occurs within a one-year timescale.
  
  This assumption may not hold across all ecosystems. In many cases - particularly in forests, semi-arid systems, or under severe disturbances - recovery can take multiple years. Restricting the analysis to a single subsequent year may therefore:
  
  Underestimate resilience in slow-recovering ecosystems
  
  Introduce biases when comparing regions with different ecological recovery timescales
  
  The authors should discuss this limitation more explicitly. At minimum, a justification for the choice of a one-year recovery window should be provided, along with a discussion of how longer recovery periods might influence the results.”
  Response: We thank the reviewer for highlighting this important limitation. We have clarified both the rationale for and the limitations of the one-year recovery window. In the revised Introduction and Methods, we explain that the first post-event year was adopted because annual peak NDVI is globally consistent across land-cover types and climatic categories and represents the most useful period for defining a recovery metric following an event. At the same time, we emphasize that this metric should be interpreted as a one-year recovery response rather than as complete ecological recovery. In the revised Discussion, we explicitly acknowledge that this approach may underestimate resilience in slowly recovering systems, particularly forests and some semi-arid ecosystems, and may introduce bias when comparing regions with different recovery timescales. We also note that future studies should evaluate multi-year recovery trajectories where sufficiently long and consistent vegetation records are available.
  The sixth paragraph of the Introduction (We adopted a one year…)
  
  The introductory sentences describing the resistance and resilience equations Section 2.4, “Calculation of Resistance and Resilience”
  
  The fifth point discussed in Section 4.3 (Limitations)
  
  The last sentence of Discussion and Conclusion
  
  Specific comment 3. Implications of NDVI limitations
  “The manuscript acknowledges known limitations of NDVI, such as saturation in high-biomass regions, but does not sufficiently discuss how these limitations may affect the findings.
  
  NDVI saturation can reduce sensitivity to vegetation changes in dense canopies, potentially leading to:
  
  Underestimation of resistance (i.e., smaller apparent declines during extreme events), and
  
  Attenuation or distortion of resilience signals during recovery
  
  This may introduce systematic biases in the spatial patterns of ecosystem vulnerability, particularly when comparing high-biomass ecosystems (e.g., tropical forests) with lower-biomass systems (e.g., grasslands).
  
  The authors should expand the discussion to address:
  
  The expected direction of these biases
  
  The regions or ecosystem types most affected
  
  The potential implications for the machine learning results (e.g., feature importance, predicted spatial patterns)
  
  Even a qualitative assessment would substantially improve the interpretation of the results.”
  Response: We fully agree and have substantially expanded the Discussion on this point. The revised manuscript now explains more clearly that NDVI saturation in dense canopies can compress the apparent magnitude of vegetation declines during extreme events, which may make resistance appear higher than it actually is in high-biomass systems. We also note that the same limitation can attenuate post-event differences during recovery, thereby weakening or distorting resilience signals. We explicitly identify tropical forests and other dense-canopy, high-biomass ecosystems as the most likely to be affected. In addition, we now discuss the implications of this limitation for the machine-learning results, noting that muted responses or lower apparent importance of some predictors in high-biomass regions may partly reflect measurement constraints of the vegetation index rather than true ecological buffering. Finally, we added a brief statement that future work should compare NDVI-based results with alternative vegetation indices such as EVI, especially in dense forests.
  The fourth point discussed in Section 4.3 (Limitations)
  
  Technical correction
  “Consider adding a short introductory paragraph at the beginning of each main section to guide the reader before presenting subsections.”
  Response: Thank you for this helpful suggestion. We have added short introductory guide paragraphs at the beginnings of the main sections, specifically Materials and Methods, Results, and Discussion, to improve readability and to orient the reader before the subsections are presented.
  Technical correction
  “Briefly clarify how the 5° × 5° grid aggregation may influence the interpretation of heterogeneous landscapes.”
  Response: We have revised this part for clarity. First, we corrected the spatial harmonization description to indicate that the datasets were harmonized to a 5-min × 5-min grid, which is the actual analytical resolution used in this study. We then added clarifying text in the Methods and Discussion stating that, even at this common grid resolution, sub-grid heterogeneity in land cover, topography, management, and species composition is inevitably averaged. We therefore note more explicitly that the results should be interpreted as broad-scale patterns rather than as fully resolving within-cell ecological heterogeneity.
  The second sentence of Section 2.1, “Overview of the Analytical Workflow and Spatiotemporal Scales”
  
  We have added the following sentence in the Limitation, “Finally, although all datasets were harmonized to a common 5-min grid, each cell may still contain heterogeneous land cover, terrain, and management conditions. Consequently, the results should be interpreted as broad-scale patterns rather than as fully resolving within-cell ecological heterogeneity.”
  
  Technical correction
  “If not already included, add a short statement on the robustness of the machine learning model (e.g., sensitivity to hyperparameters or data partitioning).”
  Response: We appreciate this suggestion and have added a clarifying statement in the Methods
  The second paragraph of the 2.7 “Modeling Approach”
  
  Technical correction
  “A brief mention of alternative vegetation indices (e.g., EVI) could complement the discussion of NDVI limitations.”
  Response: We agree and have added a brief statement in the revised Discussion noting that future studies should compare NDVI-based results with alternative vegetation indices such as EVI, particularly in dense-canopy and high-biomass ecosystems where NDVI saturation is more likely to affect the interpretation of resistance and resilience patterns.
  The fourth point discussed in Section 4.3 (Limitations)
  
  Citation: https://doi.org/10.5194/egusphere-2026-1007-AC5

Aki Yanagawa, Rina Ueda, Sayaka Yoshikawa, Yoshihiko Iseri, and Shinjiro Kanae

Supplement

https://doi.org/10.5194/egusphere-2026-1007-supplement

Data sets

Drivers of Resistance and Resilience under Different Intensities of Extreme Climatic Events Aki Yanagawa et al. https://doi.org/10.5281/zenodo.18428493

Aki Yanagawa, Rina Ueda, Sayaka Yoshikawa, Yoshihiko Iseri, and Shinjiro Kanae

Viewed

Total article views: 1,084 (including HTML, PDF, and XML)

HTML	PDF	XML	Total	Supplement	BibTeX	EndNote
746	204	134	1,084	193	47	82

HTML: 746
PDF: 204
XML: 134
Total: 1,084
Supplement: 193
BibTeX: 47
EndNote: 82

Views and downloads (calculated since 06 Mar 2026)

Month	HTML	PDF	XML	Total
Mar 2026	656	147	121	924
Apr 2026	83	48	12	143
May 2026	7	9	1	17

Cumulative views and downloads (calculated since 06 Mar 2026)

Month	HTML	PDF	XML	Total
Mar 2026	656	147	121	924
Apr 2026	83	48	12	143
May 2026	7	9	1	17

Viewed (geographical distribution)

Total article views: 1,075 (including HTML, PDF, and XML) Thereof 1,075 with geography defined and 0 with unknown origin.

Country	#	Views	%

Latest update: 10 May 2026

Short summary

This study explores how the main drivers of ecosystem stability change under different levels of extreme weather. Using global vegetation observations and climate data, we compared the roles of climate, biodiversity, and landscape features in vulnerability. We found that stability is mainly controlled by temperature in normal years, but during extreme drought biodiversity become more important. The results suggest that conserving biodiversity can reduce vulnerability to future climate extremes.


Total:	0
HTML:	0
PDF:	0
XML:	0