the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 02 Jul 2026
Revealing the Mechanistic Linkage Between QBO-Modulated Stratospheric Dynamics and Southern Tibetan Plateau Precipitation by Stratospheric 10Be/7Be Isotope Characteristics
Abstract. The Tibetan Plateau, a pivotal component of the global climate system known as the "Asian Water Tower," governs freshwater availability for billions. However, the physical mechanisms linking stratospheric circulation to its precipitation variability remain poorly constrained, limiting predictive understanding. Here, this work constructs a new indicator based on the ratio of stratospheric tracer 10Be (t1/2 = 1.39 Ma) and 7Be (t1/2 = 53.29 d), to reveal the modulation mechanism of stratospheric Quasi-Biennial Oscillation (QBO) phase transitions on Tibetan Plateau precipitation processes and its possible large-scale vertical circulation associations. Analyzing synchronous wet-deposition data from Lhasa (Tibetan Plateau) and Xi'an (Loess Plateau) during the 2022–2023 QBO transition, we empirically analyzes the synchronous response relationship between isotope deposition and regional precipitation during the tropopause stable period determined by the 10Be/7Be ratio in precipitation samples. An XGBoost machine-learning model then isolates the coupled impact of the easterly QBO phase and upper-level circulation on precipitation. Our results demonstrate that during the observation period, the easterly QBO excites a meridional wind dipole, driving an anticyclonic circulation that enhances stratospheric air transport to the surface. This dynamical pathway substantially increases precipitation in the southern Tibetan Plateau by approximately 31 %. Attempting to mechanistically linking a fundamental mode of global atmospheric variability to regional water resources via stratospheric isotopic evidence, this framework advances the understanding of cross-scale interactions within the Earth system, with direct implications for evaluating climate model performance and future water security under global change.
- Preprint
(2578 KB) - Metadata XML
-
Supplement
(833 KB) - BibTeX
- EndNote
Status: open (until 24 Sep 2026)
- CC1: 'Comment on egusphere-2026-3087', Jinlong Wang, 13 Jul 2026 reply
-
CC2: 'Comment on egusphere-2026-3087', Yaoming Ma, 07 Sep 2026
reply
General comments:
This study employs an interdisciplinary approach, combining accelerator mass spectrometry measurements of 10Be and 7Be in wet deposition with multi-decadal reanalysis data and an XGBoost-SHAP machine learning framework, to assess whether stratospheric Quasi-Biennial Oscillation dynamics modulate precipitation over the Tibetan Plateau through stratosphere-troposphere exchange. The continuous observation of 10Be and 7Be in precipitation over the Plateau constitutes a valuable dataset, and the attempt to integrate machine learning with atmospheric tracers is a commendable effort. The research topic falls within the scope of the journal.
However, the current version of the manuscript suffers from fundamental weaknesses in scientific logic, cross-scale reasoning, and methodological execution. The core weakness lies in the attempt to answer a long-term regional climate question using short-term site-specific observational data, an approach that is logically untenable. Furthermore, key quantitative conclusions, notably the 31.3% QBO contribution derived from SHAP values, appear unreliable because they are affected by multicollinearity and confounding of causal direction.
The manuscript requires major revision or even rewriting. The authors need to fundamentally restructure their research framework, scale back their attribution claims, and correct core methodological errors.
Major Concerns:
- The role of the 10Be/7Be ratio is not clearly defined
The manuscript presents inconsistent descriptions of the 10Be/7Be ratio across different sections. The abstract and introduction describe it as a newly constructed indicator for revealing the modulation mechanism of the QBO on precipitation. In practice, however, the ratio is primarily used to identify stratosphere-troposphere exchange events and to filter data accordingly by excluding high-ratio points. In the subsequent mechanistic discussion, the isotope ratio is mentioned only qualitatively, lacking systematic quantitative analysis and playing no direct role in the calculation or validation of the 31.3% contribution. This leaves the reader uncertain whether the isotope ratio is a core analytical tool or merely an auxiliary data filtering parameter.
Moreover, high 10Be/7Be ratios are interpreted throughout the text as indicators of stratosphere-troposphere exchange, which is precisely the key process through which the QBO is hypothesized to influence precipitation. Excluding these points before analyzing the relationship between flux and rainfall therefore deviates from the research objective, since the excluded signal is exactly the phenomenon of interest. Similarly, the 31.3% contribution mentioned in the conclusions is derived independently from reanalysis data and the machine learning model, with no direct quantitative link to the 10Be/7Be ratio. Yet the abstract and conclusions frame this contribution in a way that could lead readers to believe it was directly revealed by the tracer data.
- Temporal and spatial scales of the observational data do not match the research question
A fundamental problem with the study design is that the QBO is a slowly varying signal with a period of approximately 28 months, yet the authors' observations cover only 2022 to 2023, not even a single complete QBO cycle. To properly characterize the modulation mechanism of the QBO on precipitation, at least three complete QBO cycles, corresponding to roughly seven years, would be needed to observe the reproducibility of QBO effects and to exclude the contingency of a single event. For statistically reliable attribution analysis, each phase, whether easterly or westerly, would require at least 10 to 15 years of samples, totaling 20 to 30 years of continuous observations. This is because precipitation itself has strong interannual variability; data from a single year contain too much noise, and sufficient years are needed for averaging to extract a clean signal. Furthermore, the QBO period is not constant, ranging from 22 to 34 months, and both the amplitude and phase transition speed vary from cycle to cycle. Observing only one transition cannot represent the general behavior of the QBO. More importantly, the influence of the QBO on precipitation is not a direct causal effect but is transmitted through a long chain spanning stratospheric circulation, tropospheric circulation, the monsoon, and finally precipitation. Each step in this chain is subject to interference from other factors, especially ENSO.
With only two years of data, it is impossible to separate the QBO signal from stronger and more rapidly varying interannual signals such as ENSO and the Indian Ocean Dipole. This observation length fundamentally limits the scientific contribution of this study to a case observation report, documenting how isotope compositions in Lhasa and Xi'an behaved during a single QBO phase transition. Any conclusion about the mechanism of QBO modulation of precipitation far exceeds what two years of data can support. This is not a minor fix but a fundamental scale mismatch in the research design.
Spatially, the study uses a single site in Lhasa to represent the entire southern Tibetan Plateau and a single site in Xi'an to represent the entire Loess Plateau. Yet the southern Tibetan Plateau has extremely complex topography, with Himalayan mountains and valleys running through it, meaning that precipitation at a single point in Lhasa cannot represent the entire region. The authors need to reframe the narrative. Rather than claiming that the QBO drives summer precipitation, they should state that during a QBO phase transition, they observed isotopic signals that may be associated with enhanced stratosphere-troposphere exchange, making it clear that this is a case observation that is suggestive rather than a statistical demonstration of the QBO summer impact. All references to southern Tibetan Plateau precipitation should be revised to Lhasa site precipitation, including in the title, unless additional data from a broader network of stations across the region are available to support regional claims.
- Treating Xi'an as a control site is not justified
Section 2.1 defines Xi'an, with an elevation of 420 meters, as a control site for Lhasa, with an elevation of 3650 meters. The elevation difference of more than 3000 meters means the meteorological conditions are fundamentally different, making a control comparison invalid. Lhasa lies in the free atmosphere, strongly influenced by solar radiation and stratospheric processes, while Xi'an sits in the Guanzhong Basin, governed by local urban boundary layer dynamics and aerosol scavenging mechanisms. The starting height of the back trajectories differs by a factor of six, with 6000 meters for Lhasa versus 1000 meters for Xi'an, indicating that the air masses sampled at the two sites originate from entirely different altitudes. I strongly recommend that Xi'an not be referred to as a control site. Instead, it should be called a low-elevation reference point within the same monsoon corridor. The authors should directly discuss the influence of elevation differences on isotope deposition rather than attempting to compare two sites with completely different conditions. The more radical solution would be to remove the Xi'an site entirely, since the two sites are separated by more than 1500 kilometers, have different topography, different elevations, and different meteorological conditions, making any comparison of limited value, unless the authors deliberately compare the differences between the two sites throughout the paper, in which case the title would need to be revised to highlight this comparison.
- Logical problems arising from subjective data filtering
In Section 4.2, the authors set a threshold of 10Be/7Be > 1.8 and exclude these data points before running the regression again, which substantially improves the correlation, with R² rising from 0.33 to 0.92. This practice raises two problems. First, the threshold lacks physical justification because the authors provide no independent evidence that ratios above 1.8 indeed represent stratospheric intrusion, leading readers to reasonably suspect that this threshold was chosen to make the statistical results look better. Second, the manuscript falls into a logical conflict between data processing and causal inference. On one hand, the authors subjectively remove high isotope ratio points to improve linear correlation. On the other hand, they use these very points as the core evidence for stratospheric material downward transport and its modulation of precipitation. According to the authors' own theoretical premise, stratospheric intrusion with elevated isotope ratios should act as an additive component superimposed on wet deposition, not disrupting the positive correlation between total flux and precipitation. If anything, adding these points should strengthen the positive correlation. Yet the fact that R² jumps from below 0.33 to 0.85 to 0.92 after excluding them proves that these data points do not follow the conventional precipitation scavenging mechanism, as their dynamics are more consistent with dry deposition or non-precipitation processes. By treating these points as noise and removing them, the authors are methodologically acknowledging that these signals have no direct causal relationship with precipitation. However, they then go on to use these same precipitation-independent signals to explain precipitation variability and attribute it to QBO-modulated stratospheric forcing.
This constitutes a classic circular argument where the authors presuppose that a high ratio equals stratospheric intrusion as a filtering criterion, then use the strong correlation in the filtered residual data to argue for the existence of stratospheric influence, and finally use the unvalidated stratospheric intrusion to explain precipitation genesis. If QBO-modulated stratosphere-troposphere exchange is precisely the core mechanism the paper aims to reveal, then these high-ratio points carrying strong stratospheric characteristics are precisely the most critical carriers of the process signal. Removing them as interference not only deprives the core mechanism of data support but also completely breaks the causal chain linking stratospheric signals to surface precipitation. These logical issues ultimately reduce to a single question: what exactly does the isotope ratio indicate about precipitation?
- Clear temporal mismatch between two years of observations and twenty years of model data
The isotope observations are limited to the 2022 to 2023 summer seasons, while the XGBoost machine learning attribution model is based on long-term ERA5 reanalysis data from 2001 to 2020. The 2022 to 2023 observations are not even within the training or validation period of the machine learning model, which ends in 2020. This means that the core new observational data, which the authors spent considerable effort collecting and analyzing, were logically not used to train, calibrate, or test the machine learning model that produced the 31.3% contribution figure.
Furthermore, the paper's stated motivation is to use new isotope observations to reveal stratospheric dynamic mechanisms, yet the actual structure is that the first half discusses isotopes for 2022 to 2023, and the second half abruptly shifts to two decades of large-scale meteorological reanalysis and machine learning from 2001 to 2020. The authors do not establish a clear mathematical or physical link to explain which specific features or events in the 2001 to 2020 model correspond to the high-ratio anomalies measured in Lhasa during 2022 to 2023. As a result, the isotope observations and the machine learning attribution analysis operate independently of each other. This disjointed structure makes the logical chain less coherent, rendering the isotope observations a relatively isolated component that lacks substantive connection to the subsequent machine learning attribution. The spatiotemporal context on which the machine learning model is based has no direct overlap with the isotope observation data, meaning the resulting 31.3% contribution is entirely computed from long-term reanalysis data and lacks direct cross-validation or linkage with the isotope observations that the paper emphasizes as its innovative contribution.
- Methodological problems in the machine learning approach
The authors acknowledge in Section 2.5 that correlations among variables affect SHAP analysis results, yet in Section 4.4 they present a precise 31.3% contribution, creating an internal inconsistency. More seriously, the model uses concurrent near-surface relative humidity and specific humidity as predictors of precipitation. Because precipitation and humidity co-occur, using humidity measured at the same time as precipitation to predict precipitation reverses the causal direction.
Additionally, translating SHAP values into percentages lacks a mathematical basis. SHAP values represent the marginal contribution of features to model output, with units matching those of precipitation itself. Dividing a feature SHAP value by total precipitation and claiming that this feature contributed 31.3% has no mathematical justification, especially since SHAP values can be positive or negative, and feature interactions mean that individual SHAP values do not strictly sum to the prediction value. I strongly recommend that the authors conduct variance inflation factor tests and remove variables with VIF greater than 10, replace concurrent humidity with humidity from prior time steps such as previous day values, and use SHAP only for relative importance ranking while refraining from presenting an absolute percentage like 31.3%.
- Back trajectory starting heights do not reach the stratosphere
The Lhasa back trajectories start at 6000 meters above sea level. Since the Lhasa site itself is at 3650 meters, the effective starting height is approximately 9650 meters or about 300 hPa, which is still in the upper troposphere. Because the QBO signal resides near 30 hPa at approximately 24 kilometers, a vertical gap of about 14 kilometers remains unbridged. This altitude is far from reaching the stratosphere, meaning the trajectories track upper tropospheric local circulation rather than stratospheric air masses. The authors should either raise the trajectory starting heights to reach stratospheric levels or explicitly acknowledge this limitation in the text.
- Structural disorganization of paragraphs
The manuscript suffers from structural disorganization, with multiple paragraphs belonging in the Results section placed in the Discussion, and material appropriate for the Introduction appearing in later sections. For example, the regression analysis of isotope fluxes against precipitation amounts is presented in the Discussion rather than in Results, while the extended discussion of QBO background characteristics in the Results section would be more appropriately placed in the Introduction. A systematic reorganization of the text is needed to establish a logical progression from empirical observations to numerical modeling and synthesis.
Specific comments:
Line 26: analyzes –> analyze
Line 33: linking –> link
Line 49: 370 billion yuan –> 370 billion CNY
Line 412: V200 –> V100
Line 452: Hugh Coe, 2024 –> Coe, 2024
Citation: https://doi.org/10.5194/egusphere-2026-3087-CC2
Viewed
| HTML | XML | Total | Supplement | BibTeX | EndNote | |
|---|---|---|---|---|---|---|
| 76 | 43 | 11 | 130 | 18 | 7 | 13 |
- HTML: 76
- PDF: 43
- XML: 11
- Total: 130
- Supplement: 18
- BibTeX: 7
- EndNote: 13
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
Liu et al. present daily observations of cosmogenic 10Be and 7Be in precipitation from Lhasa and Xi'an during the 2022–2023 rainy seasons. They propose using the ¹⁰Be/⁷Be ratio as a tracer for stratosphere-troposphere exchange processes modulated by the Quasi-Biennial Oscillation (QBO), thereby revealing the impact of these processes on summer precipitation over the southern Tibetan Plateau. The topic is interesting and combines isotope observations, reanalysis, trajectory analysis, and machine learning. The observational dataset itself is also valuable, particularly the measurements from the Tibetan Plateau. particularly the measurements from the Tibetan Plateau. However, several issues need to be addressed before publication. I suggest making minor revisions based on the following concerns:
Others: