the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 02 Feb 2026
Causal drivers of alpine flood variability from 1300 to 2020 revealed by climate time series analysis
Abstract. Understanding flood variability in the European Alps is critical for risk management, yet causal mechanisms linking climate drivers to floods remain unclear. Previous studies have largely relied on correlations, limiting causal attribution. Here we apply a constraint-based causal inference framework using the PC-stable algorithm, combined with bootstrap stability analysis and Granger causality validation, to multi-centennial climate and flood proxy records from the Hasli-Aare catchment (1300–2020 CE). Our results reveal that total solar irradiance modulates summer atmospheric circulation, notably the summer North Atlantic Oscillation, which causally influences alpine flood frequency. These relationships are strongest during the preindustrial period and weaken under modern anthropogenic forcing, indicating a shift in dominant flood drivers. Our study demonstrates the utility of causal inference methods in paleoclimate research and offers a framework for investigating changes in the drivers of hydrological extremes, important for climate attribution and risk assessment in mountain environments.
- Preprint
(1014 KB) - Metadata XML
-
Supplement
(1019 KB) - BibTeX
- EndNote
Status: open (until 21 Apr 2026)
- RC1: 'Comment on egusphere-2026-303', Anonymous Referee #1, 02 Mar 2026 reply
Viewed
| HTML | XML | Total | Supplement | BibTeX | EndNote | |
|---|---|---|---|---|---|---|
| 108 | 75 | 14 | 197 | 19 | 9 | 24 |
- HTML: 108
- PDF: 75
- XML: 14
- Total: 197
- Supplement: 19
- BibTeX: 9
- EndNote: 24
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
The authors investigate the potential causal linkage between climate drivers and flood frequency in the Hasli-Aare catchment from 1300 to 2020 CE. Based on different sources of climate and flood proxy records, the PC-stable algorithm combined with bootstrap stability analysis and Granger causality validation is used to detect the climate-flood relationships. The results show that total solar irradiance modulates the summer atmospheric circulation, which may causally influence the flood frequency. Under modern anthropogenic forcing, these relationships are weaker than those during the preindustrial period.
In general, the structure of the manuscript is clear and the proposed causal inference framework sounds technically promising. However, there are some concerns, as well as some more specific comments that should be addressed by the authors (please see the relevant section below), before it can be considered for publication.
Major concerns:
For the Introduction section, it clearly outlines the analytical framework adopted in the manuscript and provides sufficient background to understand the methodological steps that follow. However, in its current form, it reads more as a retrospective summary of the authors’ previous work than as a critical positioning of the present study within the broader literature. A substantial proportion of the cited references derive from earlier publications by the same research group in the same catchment, which limits the integration of this study into the wider scientific context. Regarding the stated research gap of moving beyond correlations to identify causal mechanisms linking climate drivers and floods, this objective is relevant but not sufficiently substantiated. The Introduction does not clearly explain why correlation-based approaches are inadequate in this specific context, nor does it discuss how other studies have addressed similar questions and where key challenges persist. I encourage the authors to incorporate a broader range of external literature to demonstrate what has been achieved in this field, where traditional correlative methods face limitations, and why formal causal identification provides added value in disentangling the complexities of Alpine flood attribution.
All datasets used in this study are currently described in the Supplementary Material, presumably to maintain a streamlined structure in the Methods section. While this intention is understandable, given that the analysis relies on multiple paleo-reconstructed datasets and model simulations for several key variables, I suggest moving Sections S1.1 and S1.2 into Section 2 of the main text. This would improve transparency and allow readers to more easily connect the variables introduced in the Methods with the results presented later, particularly in Section 4. As it stands, it is difficult to directly link specific outcomes to the corresponding time series without repeatedly consulting the supplementary material.
In addition, Flood_F1, which represents flood frequency and could be one of the core variables in the analysis, is only briefly described with reference to previous publications. Given its central role, more detailed information should be provided in the main text, including its temporal resolution, exact coverage period, construction method, and associated uncertainties. Similar issues apply to several other time series used in the study. This is particularly important because the title specifies 1300–2020 as the study period, whereas the end years listed in Table S2 for several series appear to fall between 1800 and 2005 and the others remain unclear. The manuscript does not clearly explain how these discrepancies in temporal coverage are handled, nor how differences in time span and resolution across datasets are reconciled within the causal analysis.
Finally, the inferred causal relationships are derived from multiple reconstructed time series based on different paleoclimate proxies and model simulations, each of which carries its own methodological and reconstruction uncertainties. These uncertainties arise from proxy calibration, dating errors, spatial representativeness, and model assumptions, and they may propagate into the causal discovery framework. At present, these datasets appear to be treated implicitly as ground truth, with limited discussion of how their uncertainties might influence the robustness of the inferred causal links. At least, the manuscript should discuss this issue more explicitly. I am also curious whether the identified causal linkages persist when referring to the recent observational period, where instrumental data are available. Testing the proposed relationships using modern observations could provide an important consistency check and help evaluate whether the inferred mechanisms are robust.
Specific comments:
Line 65, Fig. 1: It is a bit hard to find the relevant information in this figure and I suggest separating the two panels into two figures.
Line 77, Fig. 2: It is also hard to see the context in this figure and I suggest restructuring the layout to place the input data in the left panel and the part of the causal framework in the right panel. Furthermore, there is a lack of consistency between the figure and the Supplementary Material, as several variables in TIER 3 and TIER 5 are missing corresponding descriptions in Tables S1 and S2. Lastly, it should be “Spring Precipitation” in TIER4.
Line 90 and the other lines: The references to tables and figures in the main text should be checked because their numbering does not match that in the supplementary materials.
Line 168: The results provided in Table 1 are the same as in Supplementary Material 2.
Lines 171 and 186: The choice of 0.58 as highly stable edges for the unconstrained DAG and 0.41 for the constrained DAG requires clearer statistical justification. With 100 bootstrap replicates, an edge strength of 0.58 implies absence in 42% of resamples, which may not constitute strong stability given the inherent dating uncertainties of ±20 years. The relaxation to 0.41 further raises concerns regarding the empirical support for the constrained model.
Fig. 3, Fig. S6 and Fig. S9: I would appreciate clarification regarding the apparent inconsistency between the causal network results and the Granger causality analysis. Specifically, the DAGs show either no link or an inconsistent link between the flood proxy (Flood_F1) and Alpine precipitation (Pamj_Alps), whereas the Granger causality test indicates a statistically significant relationship (p < 0.001).
Line 205: For section 4.5, what is the bootstrap stability threshold for the DAG? While differences in DAG topology between pre-industrial and industrial periods demonstrate structural evolution in the inferred network, further caution is needed before interpreting these changes to anthropogenic forcing since no explicit anthropogenic variables were included in the network.
Lines 277-282 and Lines 290-294: These parts appear to contain repetitive content.
Lines 359 and 427: There are some symbol errors.