| 12 Aug 2026
Impact of meteorological conditions on rockfalls: a case study of the Saint Eynard cliff
Abstract. Rockfall events are influenced by various climatic factors. Identifying these factors is crucial for anticipating rockfall events and managing the risks to infrastructure and populations. Previous studies have particularly highlighted rainfall and freezing conditions as key triggers, based on rockfall catalogs derived from visual observations or diachronic comparisons, with temporal precision ranging from a few to several days.
The purpose of this study is to use a catalog derived from seismic detection, giving a much greater temporal accuracy, to explore and better understand the meteorological conditions that contribute to rockfall at Mont Saint-Eynard, a limestone cliff in the French Alps. This study introduces a probabilistic approach that combines different temporal horizons and cumulative models capturing delayed effects to analyze the influence of rainfall and freezing conditions on rockfall triggering. By integrating multiple statistical metrics relative to a specific condition including empirical probabilities, lift measures, confidence intervals and Chi-squared tests, we identified not only the likelihood but also the strength and reliability of triggering patterns. We identified the significant role of short term triggers like recent intense rainfall on high magnitude rockfall, and delayed effect of freezing depth in rockfall activity. This knowledge could be used to implement a risk management strategy.
General comments
This study uses an inventory of rockfalls derived from seismic detection, which provides a better temporal accuracy than other detection methods. Several approaches are used to study the influence of meteorological factors (rain and temperature) on rockfalls. A) Comparison of the global distributions of the meteorological factors (daily and hourly) with their distributions when rockfalls occur. B) Cross-correlation analysis to explore how meteorological factors influence rockfall activity over time. C) Empirical conditional probability analysis. The paper is well written and comprehensible.
However, the choice of the authors to estimate conditional probabilities for forecast horizons of several days rather than for individual days at d1, d1+1, d1+2… makes it difficult the interpretation of the results. The decrease of the rockfall probability with the elapsed time since the rainfall is not sufficiently highlighted. It should be more underlined that the probability shown in different tables is not a daily probability but the probability that a rockfall occurs within several consecutive days following the rainfall period (see my suggestions for line 389).
Displaying some global statistical results could help the lecturer: total number of days considered (11 years x 365 days = 4015 days? Or are there gaps during the monitoring period?), number of days with at least one rockfall, daily probability of having a rockfall (“baseline” probability, to compare with Table 15), number of rainy days during the eleven years of observation, daily probability of having a rainy day, number of freezing days, daily probability of having a freezing day.
Cite references according to NHESS rules: In terms of in-text citations, the order can be based on relevance, as well as chronological or alphabetical listing, depending on the author's preference. In-text citations can be displayed as "[…] Smith (2009) […]", or "[…] (Smith, 2009) […]". If the author's name is part of the sentence structure only the year is put in parentheses ("As we can see in the work of Smith (2009) the precipitation has increased"). If the author's name is not part of the sentence, name and year are put in parentheses ("Precipitation increase was observed (Smith, 2009)"). If you refer to multiple references at the same position all references are put in parentheses separated by semicolons ("Precipitation increase was observed (Smith, 2009; Mueller et al., 2010)").
Figures: Each legend should be self-contained, offering all the necessary information for readers to understand the figure without referring back to the text. I suggest to complete some figures.
Discussion
The rockfall probability in a given cliff depends mainly of the size of the study area and of the rock type. It is, therefore, not relevant to compare the probabilities in very different study areas. However, it would be relevant to compare how the baseline probability is increased by the meteorological factors, using for example the ratio P(RF/M)/P(RF). Such comparisons could be made using the results of Delonca et al. (2014) already given in the introduction of the present article. As the Saint Eynard cliff was already monitored by D’amato et al. (2016), a comparison should be made. Note that D’Amato et al. (2016) observed rockfalls that are globally smaller than in the present study.
The location of the seismic stations should be given.
Suggested modifications
Lines 44-48. The description of D’Amato et al. work is incomplete. I suggest to suppress lines 44-48 and add this description line 67: “D’Amato et al. (2016) overcame this problem by combining LiDAR and a near-continuous photographic survey with a temporal resolution of 10 minutes (with a good visibility). Their results show that rockfall frequency at the Saint-Eynard limestone cliff can increase by up to a factor of seven during freeze-thaw cycles and by a factor of 26 when the average rainfall intensity (since the onset of rainfall) exceeds 5 mm/h.”
Fig. 1a. Please complete the legend (meaning of the yellow texts)
Lines 87-101 describe results and not data. They should be suppressed from section 2.
Fig. 6. In summer, the mean is higher than the percentiles? Is there an error? Please clarify!
Lines 184. It must be mentioned what meteorological station was used and its elevation, compared with the elevation of the cliff.
Line 193. This part should not be in the section “Study site and data”, but in the analysis section.
Lines 284 and 297. I think it would be sufficient and clearer to write:
M = 1 ⇐⇒ ∀j ∈ {−kR, . . . ,−1}, ej = m
R = 0 ⇐⇒ ∀j ∈ {1, . . . ,kP }, ej = r
Line 380. The sentence “The number of rockfalls that occur after different lengths of continuous rain events (RH) is presented in Figure 14” is INCONSISTENT with the legend of Figure 14 “Heatmap of |{o ∈ O | P = 1∩R = 1}|”, which is not the number of rockfalls, but the number of days (or periods) where it rained (P = 1) during the entire RH period preceding d1 and that at least one rockfall (R = 1) occurred during the FH period starting at d1 (line 368). This point must be clarified.
Line 389. For a better understanding, I suggest to add, after “to reach its maximum probability of 1”: “If there was no influence of rainfall, the number of rainfalls in a line of Figure 12 should be proportional to the number of days of the FH. Considering for example the line RH=1, if the number of rockfalls during 14 days is 1227, the number in 1 day should be about 88 compared with 276. This shaws that the shorter the delay after a rainfall, the higher the rockfall probability. Considering the influence of the retrospective horizon, the conditional probability generally increases with RH, suggesting that the longer the rainfall period, the higher the rockfall probability. But this increase is always less than 50%.”
Other suggested modifications are in the pdf of the manuscript.