1950&ndash;2100 climate trends in avalanche activity in Haute-Maurienne, French Alps

Doussot, François; Viallon-Galinier, Léo; Eckert, Nicolas; Hagenmuller, Pascal

doi:10.5194/egusphere-2026-336

Preprints

https://doi.org/10.5194/egusphere-2026-336

Preprints

03 Feb 2026

| 03 Feb 2026

1950–2100 climate trends in avalanche activity in Haute-Maurienne, French Alps

François Doussot, Léo Viallon-Galinier, Nicolas Eckert, and Pascal Hagenmuller

Abstract. Avalanche activity in alpine regions is sensitive to climate change. However, without consistent historical data, it is challenging to estimate past trends in avalanche activity and assess future avalanche scenarios from climate projections. To tackle this challenge, we use avalanche observations and simulated snowpack conditions to train a machine learning gradient-boosting regression model, which predicts the number of avalanches per day. We focus on a small alpine domain with high-quality data: the Haute-Maurienne valley in the French Alps, where avalanche paths span elevations from approximately 1800 to 2700 m a.s.l. First, we demonstrate that accounting for the uncertainties in avalanche occurrence dates and using only the most recent period (2006–2023) with homogeneous observations during the training step is essential for achieving consistent results. We then use this machine learning model to reconstruct the past avalanche activity (1958–2023) from reanalysed meteorological and snow data, and to project future avalanche activity (1950–2100) from a downscaled ensemble of snow-climate simulations. We evaluate climatic trends in avalanche activity using three indicators: the number of avalanches per winter season, the number of avalanches per month, and the annual maximum number of avalanches in one week, which quantifies the largest avalanche cycles. Based on reanalysed snow-climate simulations, the model estimates that avalanche activity decreased in the past: the mean number of avalanches per year declined by approximately 9 % per decade between 1958 and 2023, with a stronger decrease in spring avalanche activity, and the 30-year return level associated with large avalanche cycles decreased at a slower rate of around 4 % per decade. In the future, avalanche activity is also expected to decrease. For the emission scenarios RCP4.5 and RCP8.5, the annual number of avalanches is expected to decrease by around 5 % and 9 % per decade, respectively, mainly due to a reduction in spring avalanche activity. Large avalanche cycles, quantified by the 30-year return level, are also expected to decrease in intensity but at slower rates: around 2 % per decade for RCP4.5 and 5 % per decade for RCP8.5. This study quantifies the impact of climate change on avalanche activity in an exemplary alpine valley. It demonstrates that combining statistical learning with climate simulations can help produce reference scenarios for mitigation strategies in high mountain environments.

Received: 21 Jan 2026 – Discussion started: 03 Feb 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

François Doussot, Léo Viallon-Galinier, Nicolas Eckert, and Pascal Hagenmuller

Status: closed

RC1:
'Comment on egusphere-2026-336', Simon Horton, 21 Apr 2026
GENERAL COMMENTS
This manuscript analyzes trends in the number of avalanches in a French alpine valley from 1950 to 2100 using historical data and future climate scenarios. The authors combine weather datasets, a snow cover model, and a machine learning approach to predict the daily number of avalanches by aspect sector, calibrated against a long-term observational record.
The study examines changes in total seasonal avalanche counts, monthly distributions, and the most active week each winter. A key strength is moving beyond the binary “avalanche day” metric commonly used in previous climate studies to predicting daily avalanche counts, providing a more informative measure of changes in hazard severity with clear relevance for planning and risk management.
Overall, the manuscript is clearly written and structured, uses high-quality data and methods, and presents and interprets results that are relevant to the natural hazards community. It is well suited for NHESS. I recommend publication after addressing the comments below.

SPECIFIC COMMENTS
Predictors across elevation bands: The rationale for selecting predictors from multiple SAFRAN elevation bands is not fully clear. As most predictors relate to avalanche release, conditions at start-zone elevations should be more relevant than those near runout elevations. Taking certain variables taken from a single elevation (e.g., wind direction, persistent weak layers) and others from multiple elevations accentuates an inconsistency. Snowpack conditions at lower elevations are likely secondary influences (e.g., sufficient snow depth to cover ground roughness, snow available for entrainment, or weak layers for step-downs). Mixing these predictors with those from start zone elevations may weaken model performance and, as noted in the discussion, adds complexity due to elevation-dependent climate change effects. A more consistent approach that either directly targets start-zone elevations or more clearly justifying the inclusion of lower elevations would improve interpretability.

Results by aspect: It would be interesting to report model performance by aspect (e.g., mean absolute error for each aspect) to better understand the skill of the model chain. Since the model predicts avalanches by aspect, this also raises the question of whether past and future trends differ by aspect (for example is there a stronger decrease on south-facing slopes than north-facing slopes?). While a full additional analysis may be beyond scope of this paper, briefly reporting performance by aspect and discussing potential aspect-dependent trends would add value.

Underpredicting biggest cycles: This is an interesting result that leads to questions about whether it is a limitation of the physical models, the selected predictors, or the extreme value statistics models. This could deserve a bit more direct attention in the discussion.

Length and repetition: Some sections are longer than necessary and occasionally repeat similar points across different sections the manuscript. Certain details could be reduced without affecting the core message, results, or interpretation. In my opinion streamlining the text would improve clarity and strengthen the paper overall, but the current level of detail is also acceptable if the authors prefer it as is.

TECHNICAL COMMENTS
Abstract: Clear and effective. It concisely conveys the key methods and findings.

Line 70-73: Sentence is confusing and should be clarified.

Line 90: What metric did Castabrunet use for future avalanche activity?

Line 133: Wind can also be a primary driver of snowpack variability by aspect, which I assume is also true for this study area.

Line 167: “Strong” activity is vague. Please specify (e.g., higher avalanche counts).

Fig 2 and 3. Captions should clarify that these refer to the Haute-Maurienne subset, as “EPA dataset” could be interpreted as covering all of France.

Line 188: Elevation bands are not described as ranges. Are the bands centered around the reported values?

Line 195: Briefly define/explain quantile mapping.

Line 201: Add a sentence describing the final weather forcing datasets (e.g., temporal resolution and how it is structured across elevation and aspect).

Line 206: Clarify whether “50 layers” refers to snow stratigraphy layers and whether “5 variables” refers to snowpack properties.

Line 226: It would be worth noting in the discussion the limitations of using a single strength–stress ratio to represent persistent weak layer avalanches. These types of avalanche are among the most complex to model and forecast, and a simplified metric is likely to miss important processes. A brief acknowledgment of this limitation would strengthen the interpretation of results.

Sect 2.3.2: It appears predictors are computed daily at 18:00. This should be explicitly stated.

Sect 2.4.2: This section does not clearly explain how the different metrics are combined into a loss function. While nₘᵢₙ and nₘₐₓ are intuitive based on the observation windows, the role of ΔT is unclear. Even with Fig. 4, its interpretation and purpose in normalizing residuals are difficult to follow. Clarifying how ΔT functions within the loss formulation would improve readability.

Line 276: Winter seasons are often labeled by the year they end, though conventions vary.

Line 405: Sentence is unclear and should be revised.

Line 419: Please be more specific about the uncertainty in older observations. If the concern is potential underreporting in earlier periods, this could be stated more directly rather than described as generic uncertainty.

Line 560: Consider starting a new paragraph when transitioning to future results.
Citation: https://doi.org/10.5194/egusphere-2026-336-RC1
- AC1: 'Reply on RC1', François Doussot, 22 Jun 2026
  
  Thank you for the relevant comments. Detailed responses to all comments are provided in the attached document.
  
  Citation: https://doi.org/10.5194/egusphere-2026-336-AC1
RC2:
'Comment on egusphere-2026-336', Bert Kruyt, 28 May 2026
This work analyzes trends in avalanche activity over long periods, and does so by training a machine learning model with observed avalanches and the corresponding weather and snowpack data, in order to predict avalanche activity from weather and snow simulations outside the observational record. Although I am by no means an expert on machine learning, this appears as an appropriate use of such algorithms.

Especially the attempt to asses the changes in large avalanche cycles is novel, although there are some limitations in the methodology w.r.t this application. However, I do feel that with some elaboration on-, and possibly quantification of these effects, this paper warrants publication.

Specifically:
fig 2 shows an increase in avalanche activity, including the period 2006-2022. Arguably climate change was already taking effect here, yet rather than decreasing the records increase. This increase is insufficiently explained in the methods, as well as the effect it can have on the results.

ln 195: Quantile mapping assumes distributions in historical data are preserved in future data. As such, they tend to impose past changes on future data, and are generally not well suited for the analysis of extremes. The authors would improve the paper by mentioning this in the methods, as well as the way the ADAMONT dataset aims to mitigate this. Now the reader has to wait for the discussion of this -in my opinion- important aspect.

In the discussion of these shortcomings due to quantile mapping (sect 4.2), it could be elaborated upon how these shortcomings might affect the results (more than just that they affect the results).

fig 5: from the figure it appears if the model underestimates the extremes in 2017 on all metrics. This should be related to the shortcomings due to quantile mapping mentioned above.

ln330; temporal linear trends are calculated (% per decade). Why the assumption that the changes should be linear in time, since warming or CO2 increase are not?

fig 6: Not clear in black and white.( I don't know if that is still a requirement in this day and age, but:) This could be improved by different linestyles for observed and simulated trendlines (e.g. dashed and dot-dashed).

In general, many studies have noted or predicted a decrease in avalanche activity due to climate change. However the question why is not always elaborated on sufficiently. The obvious explanation is that the winters are simply getting shorter, and as a result, less avalanches occur over a season. Section 4.3 (and possibly the introduction) could benefit from mentioning this point explicitly. IMHO the interesting question is whether the decrease in avalanche activity can be solely explained by the decrease in winter duration, or if there are other dynamics at play. I hate to be the reviewer suggesting that his own paper be cited, but I'm going to do it anyway, as I feel that in this case mentioning this work (arc.lib.montana.edu/snow-science/objects/ISSW2023_O6.04.pdf ) is relevant.

Apologies that my review took a while. It is an interesting paper that I recommend be published given some small revisions.
Citation: https://doi.org/10.5194/egusphere-2026-336-RC2
- AC2: 'Reply on RC2', François Doussot, 22 Jun 2026
  
  Thank you for the relevant comments. Detailed responses to all comments are provided in the attached document.
  
  Citation: https://doi.org/10.5194/egusphere-2026-336-AC2

Status: closed

RC1:
'Comment on egusphere-2026-336', Simon Horton, 21 Apr 2026
GENERAL COMMENTS
This manuscript analyzes trends in the number of avalanches in a French alpine valley from 1950 to 2100 using historical data and future climate scenarios. The authors combine weather datasets, a snow cover model, and a machine learning approach to predict the daily number of avalanches by aspect sector, calibrated against a long-term observational record.
The study examines changes in total seasonal avalanche counts, monthly distributions, and the most active week each winter. A key strength is moving beyond the binary “avalanche day” metric commonly used in previous climate studies to predicting daily avalanche counts, providing a more informative measure of changes in hazard severity with clear relevance for planning and risk management.
Overall, the manuscript is clearly written and structured, uses high-quality data and methods, and presents and interprets results that are relevant to the natural hazards community. It is well suited for NHESS. I recommend publication after addressing the comments below.

SPECIFIC COMMENTS
Predictors across elevation bands: The rationale for selecting predictors from multiple SAFRAN elevation bands is not fully clear. As most predictors relate to avalanche release, conditions at start-zone elevations should be more relevant than those near runout elevations. Taking certain variables taken from a single elevation (e.g., wind direction, persistent weak layers) and others from multiple elevations accentuates an inconsistency. Snowpack conditions at lower elevations are likely secondary influences (e.g., sufficient snow depth to cover ground roughness, snow available for entrainment, or weak layers for step-downs). Mixing these predictors with those from start zone elevations may weaken model performance and, as noted in the discussion, adds complexity due to elevation-dependent climate change effects. A more consistent approach that either directly targets start-zone elevations or more clearly justifying the inclusion of lower elevations would improve interpretability.

Results by aspect: It would be interesting to report model performance by aspect (e.g., mean absolute error for each aspect) to better understand the skill of the model chain. Since the model predicts avalanches by aspect, this also raises the question of whether past and future trends differ by aspect (for example is there a stronger decrease on south-facing slopes than north-facing slopes?). While a full additional analysis may be beyond scope of this paper, briefly reporting performance by aspect and discussing potential aspect-dependent trends would add value.

Underpredicting biggest cycles: This is an interesting result that leads to questions about whether it is a limitation of the physical models, the selected predictors, or the extreme value statistics models. This could deserve a bit more direct attention in the discussion.

Length and repetition: Some sections are longer than necessary and occasionally repeat similar points across different sections the manuscript. Certain details could be reduced without affecting the core message, results, or interpretation. In my opinion streamlining the text would improve clarity and strengthen the paper overall, but the current level of detail is also acceptable if the authors prefer it as is.

TECHNICAL COMMENTS
Abstract: Clear and effective. It concisely conveys the key methods and findings.

Line 70-73: Sentence is confusing and should be clarified.

Line 90: What metric did Castabrunet use for future avalanche activity?

Line 133: Wind can also be a primary driver of snowpack variability by aspect, which I assume is also true for this study area.

Line 167: “Strong” activity is vague. Please specify (e.g., higher avalanche counts).

Fig 2 and 3. Captions should clarify that these refer to the Haute-Maurienne subset, as “EPA dataset” could be interpreted as covering all of France.

Line 188: Elevation bands are not described as ranges. Are the bands centered around the reported values?

Line 195: Briefly define/explain quantile mapping.

Line 201: Add a sentence describing the final weather forcing datasets (e.g., temporal resolution and how it is structured across elevation and aspect).

Line 206: Clarify whether “50 layers” refers to snow stratigraphy layers and whether “5 variables” refers to snowpack properties.

Line 226: It would be worth noting in the discussion the limitations of using a single strength–stress ratio to represent persistent weak layer avalanches. These types of avalanche are among the most complex to model and forecast, and a simplified metric is likely to miss important processes. A brief acknowledgment of this limitation would strengthen the interpretation of results.

Sect 2.3.2: It appears predictors are computed daily at 18:00. This should be explicitly stated.

Sect 2.4.2: This section does not clearly explain how the different metrics are combined into a loss function. While nₘᵢₙ and nₘₐₓ are intuitive based on the observation windows, the role of ΔT is unclear. Even with Fig. 4, its interpretation and purpose in normalizing residuals are difficult to follow. Clarifying how ΔT functions within the loss formulation would improve readability.

Line 276: Winter seasons are often labeled by the year they end, though conventions vary.

Line 405: Sentence is unclear and should be revised.

Line 419: Please be more specific about the uncertainty in older observations. If the concern is potential underreporting in earlier periods, this could be stated more directly rather than described as generic uncertainty.

Line 560: Consider starting a new paragraph when transitioning to future results.
Citation: https://doi.org/10.5194/egusphere-2026-336-RC1
- AC1: 'Reply on RC1', François Doussot, 22 Jun 2026
  
  Thank you for the relevant comments. Detailed responses to all comments are provided in the attached document.
  
  Citation: https://doi.org/10.5194/egusphere-2026-336-AC1
RC2:
'Comment on egusphere-2026-336', Bert Kruyt, 28 May 2026
This work analyzes trends in avalanche activity over long periods, and does so by training a machine learning model with observed avalanches and the corresponding weather and snowpack data, in order to predict avalanche activity from weather and snow simulations outside the observational record. Although I am by no means an expert on machine learning, this appears as an appropriate use of such algorithms.

Especially the attempt to asses the changes in large avalanche cycles is novel, although there are some limitations in the methodology w.r.t this application. However, I do feel that with some elaboration on-, and possibly quantification of these effects, this paper warrants publication.

Specifically:
fig 2 shows an increase in avalanche activity, including the period 2006-2022. Arguably climate change was already taking effect here, yet rather than decreasing the records increase. This increase is insufficiently explained in the methods, as well as the effect it can have on the results.

ln 195: Quantile mapping assumes distributions in historical data are preserved in future data. As such, they tend to impose past changes on future data, and are generally not well suited for the analysis of extremes. The authors would improve the paper by mentioning this in the methods, as well as the way the ADAMONT dataset aims to mitigate this. Now the reader has to wait for the discussion of this -in my opinion- important aspect.

In the discussion of these shortcomings due to quantile mapping (sect 4.2), it could be elaborated upon how these shortcomings might affect the results (more than just that they affect the results).

fig 5: from the figure it appears if the model underestimates the extremes in 2017 on all metrics. This should be related to the shortcomings due to quantile mapping mentioned above.

ln330; temporal linear trends are calculated (% per decade). Why the assumption that the changes should be linear in time, since warming or CO2 increase are not?

fig 6: Not clear in black and white.( I don't know if that is still a requirement in this day and age, but:) This could be improved by different linestyles for observed and simulated trendlines (e.g. dashed and dot-dashed).

In general, many studies have noted or predicted a decrease in avalanche activity due to climate change. However the question why is not always elaborated on sufficiently. The obvious explanation is that the winters are simply getting shorter, and as a result, less avalanches occur over a season. Section 4.3 (and possibly the introduction) could benefit from mentioning this point explicitly. IMHO the interesting question is whether the decrease in avalanche activity can be solely explained by the decrease in winter duration, or if there are other dynamics at play. I hate to be the reviewer suggesting that his own paper be cited, but I'm going to do it anyway, as I feel that in this case mentioning this work (arc.lib.montana.edu/snow-science/objects/ISSW2023_O6.04.pdf ) is relevant.

Apologies that my review took a while. It is an interesting paper that I recommend be published given some small revisions.
Citation: https://doi.org/10.5194/egusphere-2026-336-RC2
- AC2: 'Reply on RC2', François Doussot, 22 Jun 2026
  
  Thank you for the relevant comments. Detailed responses to all comments are provided in the attached document.
  
  Citation: https://doi.org/10.5194/egusphere-2026-336-AC2

François Doussot, Léo Viallon-Galinier, Nicolas Eckert, and Pascal Hagenmuller

Viewed

Total article views: 1,818 (including HTML, PDF, and XML)

HTML	PDF	XML	Total	BibTeX	EndNote
1,074	650	94	1,818	152	186

HTML: 1,074
PDF: 650
XML: 94
Total: 1,818
BibTeX: 152
EndNote: 186

Views and downloads (calculated since 03 Feb 2026)

Month	HTML	PDF	XML	Total
Feb 2026	667	347	64	1,078
Mar 2026	295	173	16	484
Apr 2026	51	61	6	118
May 2026	27	26	2	55
Jun 2026	6	11	2	19
Jul 2026	23	20	4	47
Aug 2026	5	12	0	17

Cumulative views and downloads (calculated since 03 Feb 2026)

Month	HTML	PDF	XML	Total
Feb 2026	667	347	64	1,078
Mar 2026	295	173	16	484
Apr 2026	51	61	6	118
May 2026	27	26	2	55
Jun 2026	6	11	2	19
Jul 2026	23	20	4	47
Aug 2026	5	12	0	17

Viewed (geographical distribution)

Total article views: 1,778 (including HTML, PDF, and XML) Thereof 1,778 with geography defined and 0 with unknown origin.

Country	#	Views	%

Latest update: 17 Aug 2026

Short summary

Avalanches are sensitive to climate warming, but long and reliable records are rare. We combined avalanche observations with weather and snow simulations in the Haute-Maurienne valley (French Alps). This allowed us to reconstruct past avalanche activity and explore future changes. The results show a strong long-term decline in avalanche occurrence, especially in spring, while extreme events decrease more slowly. This study provides quantitative insights to support mountain risk management.


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