Non-stationary dynamics of compound climate extremes: a WRF-CMIP6-GAMLSS framework for southeastern China

Zhang, Yinchi; Xu, Wanling; Deng, Chao; Sun, Shao; Ma, Miaomiao; Wei, Jianhui; Chen, Ying; Wang, Yi; Gao, Lu; Kunstmann, Harald

doi:10.5194/nhess-26-2031-2026

Articles | Volume 26, issue 5

https://doi.org/10.5194/nhess-26-2031-2026

© Author(s) 2026. This work is distributed under
the Creative Commons Attribution 4.0 License.

https://doi.org/10.5194/nhess-26-2031-2026

© Author(s) 2026. This work is distributed under
the Creative Commons Attribution 4.0 License.

Articles | Volume 26, issue 5

Research article

|

08 May 2026

Research article |

| 08 May 2026

Non-stationary dynamics of compound climate extremes: a WRF-CMIP6-GAMLSS framework for southeastern China

Yinchi Zhang, Wanling Xu, Chao Deng, Shao Sun, Miaomiao Ma, Jianhui Wei, Ying Chen, Yi Wang, Lu Gao, and Harald Kunstmann

Download

Final revised paper (published on 08 May 2026)
Supplement to the final revised paper
Preprint (discussion started on 16 Jun 2025)
Supplement to the preprint

Interactive discussion

Status: closed

RC1:
'Comment on egusphere-2025-2438', Anonymous Referee #1, 27 Oct 2025

I think this work focuses on a very nice issue with clear novelty, as it utilizes the combined WRF-CMIP6-GAMLSS framework to investigate the non-stationary dynamics of compound climate extremes. It is applicable and important for climate change studies and its effect on compound extremes, especially considering their non-stationary characteristics. The manuscript is well-written, the results are well-discussed, and every section is appropriate with sufficient explanation.
Accordingly, I recommend this manuscript for publication; however, you may consider the minor comments as follows:
1-In the introduction, where you mention the need for fine-resolution models, you may need to add the reasons for the required high-resolution models for capturing compound extremes (e.g., local convective precipitation, spatial heterogeneity, …).
2-May you please check the following articles and discuss how you improve their work and what your innovation is compared to it?
https://link.springer.com/article/10.1007/s00382-020-05538-2
3-May you make a list of the 18 climate models you applied for your study in the supplementary file?
4-you mentioned that “The dataset used in this study covers the historical period (2005–2014)”, why did you chose only 10 years as historical data?
5-Why did you only use SSP2-4.5 and SSP5-8.5? Why not SSP1-2.6?
6-Please make clearer how the downscaling integrates with your GAMLSS framework.
7-Please make it clear what you mean by “enhanced” or “advanced” GAMLSS in the manuscript. Do you mean the GAMLSS, which considers non-stationary characteristics?
8-In the supplementary file, please provide a table showing the validation results.
9-In the results and discussion, I could not find how the results show that considering the non-stationary characteristics leads to better or more reliable results. May you compare the results with previous studies in which the time series was assumed to be stationary?
10-The English language also should be assessed more carefully.

Citation: https://doi.org/10.5194/egusphere-2025-2438-RC1
- AC1: 'Reply on RC1', yinchi zhang, 12 Nov 2025
  
  Dear reviewer, thank you for your constructive comments, which have been invaluable in helping us improve the manuscript. Please see the attached document for our point-by-point responses to each of your comments.
  
  Citation: https://doi.org/10.5194/egusphere-2025-2438-AC1
RC2:
'Comment on egusphere-2025-2438', Anonymous Referee #2, 17 Nov 2025

Please find my review in the attached file.

Citation: https://doi.org/10.5194/egusphere-2025-2438-RC2
- AC2: 'Reply on RC2', yinchi zhang, 23 Nov 2025
  
  Thank you very much for your detailed and constructive comments. They have been invaluable in helping us improve the manuscript. We sincerely apologize for any unclear expressions in the paper, and we appreciate your guidance in making these points more precise. Your feedback has significantly contributed to enhancing the quality of our work.
  
  Citation: https://doi.org/10.5194/egusphere-2025-2438-AC2

Peer review completion

AR – Author's response | RR – Referee report | ED – Editor decision | EF – Editorial file upload

ED: Reconsider after major revisions (further review by editor and referees) (24 Nov 2025) by Marleen de Ruiter

AR by yinchi zhang on behalf of the Authors (02 Dec 2025) Author's response Author's tracked changes Manuscript

ED: Referee Nomination & Report Request started (02 Dec 2025) by Marleen de Ruiter

RR by Anonymous Referee #1 (09 Dec 2025)

RR by Anonymous Referee #3 (17 Jan 2026)

Suggestions for revision or reasons for rejection

The manuscript presents an innovative framework integrating high-resolution dynamical downscaling (WRF) with non-stationary statistical modeling (GAMLSS) to assess the future evolution of compound climate extremes (CCEs) in the Minjiang River Basin (MRB). The manuscript is promising and methodologically interesting, but key methodological definitions and statistical interpretations must be clarified, and validation/uncertainty treatment should be strengthened to support claims about non-stationary recurrence risk.
Major comments:
1. The authors state that changes in CCEs are primarily governed by temperature-driven physical processes, while precipitation remains stable. However, a stable mean precipitation does not preclude changes in precipitation intensity or timing (e.g., shorter, more intense rainy seasons). The authors should discuss whether the shift in "Hot-Wet" (HW) extremes is influenced by changes in precipitation distribution or atmospheric moisture capacity (Clausius-Clapeyron scaling).
2. The manuscript states that wet events are based on exceedance of the 90th percentile, while dry events are defined as seven consecutive days without rainfall. This mixes a daily-threshold event (wet) with a multi-day process definition (dry). It is currently unclear how HD/CD are counted in annual totals (e.g., whether each day within a 7-day dry spell is counted, or the spell is counted as one event; how overlap with hot/cold days is handled).
3. For a monsoon-dominated coastal basin like the MRB, how do shifts in the Western Pacific Subtropical High (WPSH) or typhoons contribute to the detected non-stationarity?. Incorporating a brief discussion on these external drivers would strengthen the "Discussion" section.
4. The study runs a 10-year historical simulation (2005–2014) using CMIP6bc and ERA5 forcing and uses 30 stations for validation. CMIP6bc is described as an 18-model ensemble mean and the paper argues that 10 years is “sufficient” given computational cost. For a paper focusing on extremes and return periods, readers will expect more targeted evaluation (especially for precipitation extremes and compound-event-relevant indices).
5. The study currently uses "time" as the sole covariate to represent non-stationarity. While common, time is a proxy for physical change. The authors should comment on the potential benefits of using physically-based covariates, such as Global Mean Surface Temperature (GMST) or regional circulation indices, to improve the causal link between forcing and response.
6. The GAMLSS framework relies on selecting the best distribution using AIC. For discrete count data (number of days of CCEs), it is crucial to specify which distributions (e.g., Poisson, Negative Binomial) were tested and how the model handled "zero-inflation," as some grids may have many years with zero CCE events.
7. The authors use a 3 km convection-permitting resolution to capture local convective precipitation and orographic effects. It would be highly beneficial to explicitly state the "added value" compared to the original 1.25° CMIP6 data. Are there specific sub-basins or high-elevation areas where the low-resolution GCMs completely fail to capture CCEs that the 3 km WRF model successfully identifies?
8. While the CMIP6bc dataset is bias-corrected, the WRF model can introduce secondary biases in orographic precipitation. How were these secondary biases validated against the 30 monitoring stations mentioned in Section 2.2?
9. The finding that the 100-year return period frequency of CCEs increases significantly (3.12 days per decade) has profound implications for regional infrastructure. The authors should briefly discuss how these findings might impact reservoir operations or urban flood management in downstream cities like Fuzhou.
10. You already note systematic WRF biases in orographic precipitation. Consider briefly indicating whether any bias correction/post-processing is applied before CCE detection; if none, state this clearly and discuss implications.
11. Ensure the statistical significance levels (e.g., 90% and 99% confidence) are clearly visible in the spatial plots for the transition from stationary to non-stationary behavior.
12. Ensure that recent studies regarding "compound hot-dry" events in China are consistently cited to contextualize the MRB findings.
13. “Generalized dditive Models…” should be “Generalized Additive Models…”.

Hide

ED: Reconsider after major revisions (further review by editor and referees) (19 Jan 2026) by Marleen de Ruiter

AR by yinchi zhang on behalf of the Authors (28 Jan 2026) Author's response Author's tracked changes Manuscript

ED: Referee Nomination & Report Request started (05 Feb 2026) by Marleen de Ruiter

RR by Anonymous Referee #3 (21 Feb 2026)

ED: Publish as is (23 Feb 2026) by Marleen de Ruiter

ED: Publish as is (29 Apr 2026) by Bruce D. Malamud (Executive editor)

AR by yinchi zhang on behalf of the Authors (30 Apr 2026)

Download

Article (13868 KB)
Full-text XML

Short summary

Most studies of compound extremes assume stationary climate conditions. Here, we combine high-resolution regional climate modeling with non-stationary statistical methods to assess future changes in compound extremes over southeastern China. Our results demonstrate that non-stationary models more accurately capture shifts in the evolution of these events, suggesting that conventional approaches may systematically underestimate future risks.