Search

In mountain territories, snow avalanches are a prevalent threat. Long-term risk management involves defining meaningful compromises between protection and overall sustainability of communities and their environment. Methods able to (i) consider all sources of losses, (ii) account for the high uncertainty levels that affect all components of the risk and (iii) cope for marked non-stationarities should be employed. Yet, on the basis of a literature review and an analysis of relations to Sustainable Development Goals (SDGs), it is established that snow avalanche risk assessment and mitigation remain dominated by approaches that can be summed up as deterministic, hazard oriented, stationary and not holistic enough. A more comprehensive paradigm relying on formal statistical modelling is then proposed and first ideas to put it to work are formulated. Application to different mountain environments and broader risk problems is discussed.

Anticipating risks related to climate extremes often relies on the quantification of large return levels (values exceeded with small probability) from climate projection ensembles. Current approaches based on multi-model ensembles (MMEs) usually estimate return levels separately for each climate simulation of the MME. In contrast, using MME obtained with different combinations of general circulation model (GCM) and regional climate model (RCM), our approach estimates return levels together from the past observations and all GCM–RCM pairs, considering both historical and future periods. The proposed methodology seeks to provide estimates of projected return levels accounting for the variability of individual GCM–RCM trajectories, with a robust quantification of uncertainties. To this aim, we introduce a flexible non-stationary generalized extreme value (GEV) distribution that includes (i) piecewise linear functions to model the changes in the three GEV parameters and (ii) adjustment coefficients for the location and scale parameters to adjust the GEV distributions of the GCM–RCM pairs with respect to the GEV distribution of the past observations. Our application focuses on snow load at 1500 m elevation for the 23 massifs of the French Alps. Annual maxima are available for 20 adjusted GCM–RCM pairs from the EURO-CORDEX experiment under the scenario Representative Concentration Pathway (RCP) 8.5. Our results show with a model-as-truth experiment that at least two linear pieces should be considered for the piecewise linear functions. We also show, with a split-sample experiment, that eight massifs should consider adjustment coefficients. These two experiments help us select the GEV parameterizations for each massif. Finally, using these selected parameterizations, we find that the 50-year return level of snow load is projected to decrease in all massifs by −2.9 kN m−2 (−50 %) on average between 1986–2005 and 2080–2099 at 1500 m elevation and RCP8.5. This paper extends the recent idea to constrain climate projection ensembles using past observations to climate extremes.

Mountain regions are subject to highly damaging hydrological and gravitational hazards. This exposure is due to their biophysical and societal characteristics. It is essential for a sustainable management of these risks to consider the natural risk as the result of complex interactions within the risk system, which is composed of a natural and a societal subsystem. In this way, it is possible to adopt a dynamic approach to the risk system by placing the phenomenon in an evolutionary context. We can therefore consider its trajectory according to the socio-environmental dynamics that influence it.Studying these risk systems over the long term is necessary to understand their evolution and to anticipate future ones in order to guarantee the sustainability of mountain socio-ecosystems, and requires an interdisciplinary approach between geography and history.The study of the trajectory of a multi-hazard system is being carried out in the Commune of Vallouise-Pelvoux, a high Alpine valley in the Écrins massif, France. This territory was chosen because it is subject to various risks that occur over a wide range of altitudes, its recent socio-economic development is mainly based on tourism, and it is marked by glacial recession, but also because we were aware of the availability of several sources allowing the production of event and multirisk chronologies.The first stage of the research consisted in the production of a multi-hazard event chronology over 420 years (1600-2020). This database was built from various resources. On the one hand, from existing databases produced by public services and organizations such as the French Forest Office (ONF) specifically the mountain land restoration service (RTM) or the departmental councils. On the other hand, archival research was carried out in the municipal archives of Vallouise-Pelvoux and the departmental archives of the Hautes-Alpes.After analysis of all available sources, the data collected was processed in various ways. Indeed, sources of different forms and origins requires standardization of the information to make it comparable and usable. The chronology was also subjected to a critical analysis : are the sources authentic? Reliable? What factors might influence them? Once this chronology of events in Vallouise-Pelvoux has been contextualized (changes in the natural and societal systems of the Commune), a first statistical analysis of the risks identified and the damage caused will be presented. In the future the data will be used to analyze the trajectory of the system.

Snow avalanches are a prevalent threat in mountain territories. Large-scale mapping of avalanche-prone terrain is a prerequisite for land-use planning where historical information about past events is insufficient. To this aim, the most common approach is the identification of potential release areas (PRAs) followed by numerical avalanche simulations. Existing methods for identifying PRAs rely on terrain analysis. Despite their efficiency, they suffer from (i) a lack of systematic evaluation on the basis of adapted metrics and past observations over large areas and (ii) a limited ability to distinguish PRAs corresponding to individual avalanche paths. The latter may preclude performing numerical simulations corresponding to individual avalanche events, questioning the realism of resulting hazard assessments. In this paper, a method that accurately identifies individual snow avalanche PRAs based on terrain parameters and watershed delineation is developed, and confusion matrices and different scores are proposed to evaluate it. Comparison to an extensive cadastre of past avalanche limits from different massifs of the French Alps used as ground truth leads to true positive rates (recall) between 80 % and 87 % in PRA numbers and between 92.4 % and 94 % in PRA areas, which shows the applicability of the method to the French Alps context. A parametric study is performed, highlighting the overall robustness of the approach and the most important steps/choices to maximize PRA detection, among which the important role of watershed delineation to identify the right number of individual PRAs is highlighted. These results may contribute to better understanding avalanche hazard in the French Alps. Wider outcomes include an in-depth investigation of the issue of evaluating automated PRA detection methods and a large data set that could be used for additional developments, and to benchmark existing and/or new PRA detection methods.

In mountainous regions, global warming will likely affect the frequency and magnitude of geomorphic processes. This is also the case for rockfall, one of the most common mass movements on steep slopes. Rainfall, snowmelt, or freeze-thaw cycles are the main drivers of rockfall activity, rockfall hazards are thus generally thought to become more relevant in a context of climate change. At high elevations, unequivocal relationships have been found between increased rockfall activity, permafrost thawing and global warming. By contrast, below the permafrost limit, studies are scarcer. They mostly rely on short or incomplete rockfall records, and have so far failed to identify climatically induced trends in rockfall records. Here, using a dendrogeomorphic approach, we develop two continuous 60-year long chronologies of rockfall activity in the Vercors and Diois massifs (French Alps); both sites are located clearly below the permafrost limit. Uncertainties related to the decreasing number of trees available back in time were quantified based on a detailed mapping of trees covering the slope across time. Significant multiple regression models with reconstructed rockfalls as predictors and local changes in climatic conditions since 1959 extracted from the SAFRAN reanalysis dataset as predictants were fitted to investigate the potential impacts of global warming on rockfall activity at both sites. In the Vercors massif, the strong increase in reconstructed rockfall can be ascribed to the recolonization of the forest stand and the over-representation of young trees; changes that are observed should not therefore be ascribed to climatic fluctuations. In the Diois massif, we identify annual precipitation totals and mean temperatures as statistically significant drivers of rockfall activity but no significant increasing trend was identified in the reconstruction. All in all, despite the stringency of our approach, we cannot therefore confirm that rockfall hazard will increase as a result of global warming at our sites.

International audience; In this study, we develop a novel method to automatically detect areas of snow avalanche debris using a color space segmentation technique applied to SAR image time series through January 2018 in the Swiss Alps. Debris avalanche zones are detected assuming that these areas are characterised by a significant and localised increase in SAR signal relative to the surrounding environment. We undertake a sensitivity study by calculating debris products by varying the D-M reference images (a stable reference image taken several weeks before the event). We examine the results according to the direction of the orbit, the characteristics of the terrain (slope, altitude, orientation) and also by evaluating the relevance of the detection with the help of an independent SPOT database ([1]) including 18,737 avalanche events. Small avalanches are not detected by SAR images and depending on the orientation of the terrain some avalanches are not detected by either the ascending or the descending orbit. The detection results vary with the reference image; best detection results are obtained with some selected individual dates with almost 70 % of verified avalanche events using the ascending orbit.

Following the projected increase in extreme precipitation, an increase in extreme snowfall may be expected in cold regions, e.g. for high latitudes or at high elevations. By contrast, in low/medium elevation areas, the probability to experience rainfall instead of snowfall is generally projected to increase due to warming conditions. In mountainous areas, despite the likely existence of these contrasted trends according to elevation, changes in extreme snowfall with warming remain poorly quantified. This paper assesses projected changes in heavy and extreme snowfall, i.e. in mean annual maxima and 100-year return levels, in the French Alps as a function of elevation and global warming level using a recent methodology based on non-stationary extreme value models. This methodology is applied to an ensemble of 20 adjusted GCM-RCM pairs from the EURO-CORDEX experiment under the scenario RCP8.5. available for each of the 23 massifs of the French Alps from 1951 to 2100, and every 300 m of elevations. Results are provided as relative or absolute changes computed w.r.t. current climate conditions, at the massif scale and averaged over all available massifs. Overall, mean annual maxima are projected to decrease below 3000 m and increase above 3600 m, while 100-year return levels are projected to decrease below 2400 m and increase above 3300 m. At elevations in between, values are on average projected to increase until +3 °C of global warming, and then decrease. At +4 °C, average relative changes in mean annual maxima and 100-year return levels respectively vary from −26 % (−7 kg m−2) and −15 % (−11 kg m−2) at 900 m, to +3 % (+3 kg m−2) and +8 % (+13 kg m−2) at 3600 m. Finally, for each global warming level, we compute the elevation threshold that separates contrasted trends, i.e. where the average relative change equals zero. This elevation threshold is projected to rise between +1.5 °C and +4 °C: from 3000 m to 3350 m for mean annual maxima, and from 2600 m to 3000 m for 100-year return levels. These results have implications for the management of risks related to extreme snowfall.

The United Nations 2030 Agenda and Sustainable Development Goals (SDGs) define a path towards a sustainable future, but given that uncertainty characterises the outcomes of any SDG-related actions, risks in the implementation of the Agenda need to be addressed. At the same time, most risk assessments are narrowed to sectoral approaches and do not refer to SDGs. Here, on the basis of a literature review and workshops, it is analysed how SDGs and risks relate to each other’s in different communities. Then, it is formally demonstrated that, as soon as the mathematical definition of risks is broadened to embrace a more systemic perspective, acting to maintain socioenvironmental systems within their sustainability domain can be done by risk minimisation. This makes Sustainable Development Goals and risks ‘‘the Yin and the Yang of the paths towards sustainability’’. Eventually, the usefulness of the SDG-risk nexus for both sustainability and risk management is emphasized. 2030 Agenda Environmental risks Planet boundaries Risk quantification Sustainability science Systemic approach

The spatial and temporal quantification of rockfall frequency remains a major challenge in mountain environments, especially also in terms of rockfall management. Approaches that have been used traditionally to quantify rockfall frequency include historical records, remote sensing, or in situ monitoring, but have been shown repeatedly to suffer from a lack of completeness or to rely on rather short time series (of a few years) that are, in addition, limited to small areas. As such, they normally cannot meet the stringent requirements of hazard and risk analyses. Here, we propose a new procedure coupling field analysis of rockfall deposits in mitigation structures and growth disturbances in tree-ring series with three-dimensional (3D), process-based rockfall modeling to estimate rockfall frequencies for individual cliff compartments. This procedure has been tested on a slope in the Swiss Alps (La Fory, Sembrancher VS), where rockfall triggered from a 38 hm $$^{2}$$ cliff to threaten a 1-km-long section of the road to the Grand St. Bernard tunnel and the local railway line. Based on 84 rockfall deposits retrieved from 420-m-long protection barrier and growth disturbances in trees over the 1994–2017 period, we estimate rockfall hazard to between 0.043 and 0.348 events yr $$^{-1}$$ at various locations on the slope. In total, we used 68,680,000 rockfall simulations to translate hazard values into rockfall frequencies at the cliff level. Converging release frequencies between tree-ring analyses (0.99–1.35 events yr $$^{-1}$$ hm $$^{-2}$$ ) and values obtained from deposits in the protection barrier (0.82 events yr $$^{-1}$$ hm $$^{-2}$$ ) confirm the reliability of our procedure. Despite remaining limitations, our approach enables precise quantification of rockfall release frequency in a holistic and reproducible manner both in space (a few hm $$^{2}$$ ) and time (several decades), and thus yields the data needed for hazard and risk mapping. We therefore recommend, where applicable, to include this procedure in future rockfall management strategies.

PII S0022143022001034; International audience; This work proposes a holistic quantitative snow avalanche risk assessment that evaluates, at reasonable computational costs and for various types of buildings, the impact of forest cover changes on the probability distribution of runout distances, impact pressures and subsequent risk estimates. A typical case study of the French Alps shows that, from a completely deforested to a completely forested path, avalanche risk for a building located downslope decreases by 53–99%, depending on how forest cover is accounted for in avalanche statistical–dynamical modeling. Local forest cover data inferred from old maps and photographs further demonstrates that a 20–60% risk reduction actually occurred between 1825 and 2017 at the site because of the afforestation dynamics, with significant modulations according to the considered building technology. These results (1) assert the protective role of forests against snow avalanches, (2) highlight the potential of combining nature-based solutions with traditional structural measures to reduce risk to acceptable levels at reasonable costs, (3) suggest a significant decrease in risk to settlements in areas that encountered similar forest cover changes and (4) open the door to the quantification of long-term avalanche risk changes as a function of changes of all its hazard, vulnerability and exposure drivers in various mountain context.

Des 1990, l’OMS (1999) a percu l’importance des determinants environnementaux de la sante, ce qui l’a amenee a elargir sa definition de la sante de 1946 : « la sante environnementale comprend les aspects de la sante humaine, y compris la qualite de la vie, qui sont determines par les facteurs physiques, chimiques, biologiques, sociaux, psychosociaux et esthetiques de notre environnement. Elle concerne egalement la politique et les pratiques de gestion, de resorption, de controle et de prevention des facteurs environnementaux susceptibles d’affecter la sante des generations actuelles et futures. »Partant du constat de l’impact de la qualite des ecosystemes sur la sante humaine, le concept d’exposome a emerge. Celui-ci est defini comme la totalite des expositions a des facteurs environnementaux que subit un organisme humain, depuis sa conception jusqu’a sa fin de vie en passant par le stade du developpement in utero (Wild, 2005). La notion de « One health » elaboree par Zindsstag et collaborateurs (2011) insiste en outre sur la bidirectionnalite des relations et appelle a developper des approches integrees pour garantir une sante environnementale globale (humaine, vegetale, animale et ecosystemique). A partir de differents exemples, nous questionnons ici le niveau d’integration disciplinaire et spatio-temporelle necessaire a la prise en compte et a l’operationnalisation du concept de sante environnementale globale. Nous montrons ainsi que la sante environnementale offre un cadre d’actions pour repondre au besoin urgent du developpement d’une gouvernance des risques multidimensionnelle et integree.

Snow avalanches are a prevalent threat in mountain territories. Large-scale mapping of avalanche prone terrain is a prerequisite for land-use planning where historical information about past events is lacunar. To this aim, the most common approach is the identification of Potential Release Areas (PRAs) followed by numerical avalanche simulations. Existing methods for identifying PRAs rely on terrain analysis. Despite their efficiency, they suffer from i) a lack of systematic validation on the basis of adapted metrics and past observations over large areas and ii) a limited ability to distinguish PRAs corresponding to individual avalanche paths. The latter may preclude performing numerical simulations corresponding to individual avalanche events, questioning the realism of resulting hazard assessments. In this paper, a method that well identifies individual snow avalanche PRAs based on terrain parameters and watershed delineation is developed, and confusion matrices and accuracy scores computed both in terms of PRA numbers and areas are proposed to test and evaluate it. Confrontation to an extensive cadastre of past avalanche limits from different massifs of the French Alps used as ground truth leads to high accuracy rates, between 89.8 % and 93.5 % in numbers and 96.2 % and 97.1 % in areas. This shows the applicability of the method to the French Alps context. A sensitivity study is performed, highlighting the most important steps to reach high accuracy in PRA detection, among which the strong role of watershed delineation to identify the right number of individual PRAs. Outlooks for further progresses are discussed. Notably, the proposed data and evaluation framework could be used for additional developments of the method, and to benchmark existing and/or new PRA detection methods.

Predicting avalanche activity from meteorological and snow cover simulations is critical in mountainous areas to support operational forecasting. Several numerical and statistical methods have tried to address this issue. However, it remains unclear how the combination of snow physics, mechanical analysis of snow profiles and observed avalanche data improves avalanche activity prediction. This study combines extensive snow cover and snow stability simulations with observed avalanche occurrences within a Random Forest approach to predict avalanche days at a spatial resolution corresponding to elevations and aspects of avalanche paths in a given mountain range. We develop a rigorous leave-one-out evaluation procedure including an independent test set, confusion matrices, and receiver operating characteristic curves. In a region of the French Alps (Haute-Maurienne) and over the period 1960–2018, we show the added value within the statistical model of considering advanced snow cover modelling and mechanical stability indices instead of using only simple meteorological and bulk information. Specifically, using mechanically-based stability indices and their time-derivatives in addition to simple snow and meteorological variables increases the recall from around 65 % to 76 %. However, due to the scarcity of avalanche events and the possible misclassification of non-avalanche days in the training data set, the precision remains low, around 3.5 %, due to the scarcity of avalanche days. These scores illustrate the difficulty of predicting avalanche occurrence with a high spatio-temporal resolution, even with the current cutting-edge data and modelling tools. Yet, our study opens perspectives to improve modelling tools supporting operational avalanche forecasting.

In a context of climate change, trends in extreme snow loads need to be determined to minimize the risk of structure collapse. We study trends in 50-year return levels of ground snow load (GSL) using non-stationary extreme value models. These trends are assessed at a mountain massif scale from GSL data, provided for the French Alps from 1959 to 2019 by a meteorological reanalysis and a snowpack model. Our results indicate a temporal decrease in 50-year return levels from 900 to 4200 m, significant in the northwest of the French Alps up to 2100 m. We detect the most important decrease at 900 m with an average of −30 % for return levels between 1960 and 2010. Despite these decreases, in 2019 return levels still exceed return levels designed for French building standards under a stationary assumption. At worst (i.e. at 1800 m), return levels exceed standards by 15 % on average, and half of the massifs exceed standards. We believe that these exceedances are due to questionable assumptions concerning the computation of standards. For example, these were devised with GSL, estimated from snow depth maxima and constant snow density set to 150 kg m−3, which underestimate typical GSL values for the snowpack.

Comprehensive assessments of global glacier mass changes based on a variety of observations and prevailing methodologies have been published at multi-annual intervals. For the years in between, the glaciological method provides annual observations of specific mass changes but is suspected to not be representative at the regional to global scales due to uneven glacier distribution with respect to the full sample. Here, we present a simple approach to estimate and correct for this bias in the glaciological sample and, hence, to provide an ad hoc estimate of global glacier mass changes and corresponding sea-level equivalents for the latest years, i.e. about -300±250 Gt in 2016/17 and -500±200 Gt in 2017/18.

Snow avalanches result from a complex interaction of weather and terrain, where past weather as well as internal snow cover processes play an important role. Snow cover models account for these processes and simulate the snow cover at a level of detail that allows to describe snow instability. That information on snow instability is required to assess avalanche climates rather than snow climates – which represents a classification based on weather observations. Running the model SURFEX/Crocus with long-term meteorological data (S2M reanalysis) covering the winter seasons between 1958 and 2020, we eventually derived the avalanche problem types for the 23 massifs representing different regions of the French Alps at daily resolution. This allows us to create a climatology based on avalanche problem types and study differences between mountain regions in the French Alps.In a first step, we applied a commonly used snow climate classification to understand how in the different Alpine regions seasonal characteristics fluctuate under continental or coastal influence. In some regions the majority of the winter seasons had coastal characteristics, whereas in other regions almost half of the seasons were classified continental. In a second step, we add snow instability information by simulating avalanche problem types. This allows us to explore snow instability patterns that result from the meteorological forcing that the snow climate classification summarizes but which the snow climate classification cannot fully capture. Across the regions persistent weak layers and wet snow were the most common avalanche problems types on days when the model expected natural release. Moreover, we applied a k-means clustering to the frequencies of new snow, persistent and wet snow avalanche problem types to identify similarities between in the Alpine regions. We assessed the clustering performance and found that 4 clusters separate well our data which includes information on duration and frequency of avalanche problem types. The clusters coincide with the geographic location of the regions, i.e. Northern, Southern, inner-alpine or front-range regions have specific characteristics that manifest in frequency of avalanche problem types.We showed how avalanche problem types can be used as an additional descriptor to the snow climate classification to include snow instability specific information. As avalanche problem types identified spatial differences between different snow climate regions, we are now ready to analyze how those changes evolve with time in the different regions.

From the local to global level, indicators and reports are produced and published to support the transition towards sustainable development. Building from two European-level science–policy workshops, this perspective essay discusses the types of risks involved with such sustainability reporting. The analysis is rooted in the framework of the UN 2030 Agenda and sustainable development goals (SDG). As a globally adopted framework, it provides an example of how risks are either recognised and framed, or non-recognised. Well recognised risks include data availability for SDGs and siloed preparation of indicators, while risks receiving less attention are ritualistic reporting lacking a critical evaluation of the limitations of the SDG framework itself. These different risks are likely to reinforce each other. A specific risk is a too narrow focus on one-way communication aiming to inform individual policy decisions. Risks related to SDGs are best managed with iterative, integrative and interactive knowledge production fostering holistic understanding.

At present, a quantitative basis for delimiting realistic rockfall runout zones on the basis of trajectory simulation data is generally missing. The objective of this study is to come up with standardized reach probability threshold values (RPTV) to separate "realistic" from "unrealistic" simulated rockfall runouts. We therefore compared reach probability values (Preach) simulated with Rockyfor3D for 458 mapped, fresh rockfall blocks (silent witnesses SW) on 18 different sites with a volume >= 0.05 m3 and estimated occurrence frequencies up to 300 years. We analysed which block, slope and forest characteristics influenced Preach of the SW based on a linear mixed effects model. The results indicate that the limit of a realistic runout zone lies in the range where simulated Preach values are between > 1 % and approximately 3 %. We conclude that RPTV can be defined to values lying in the range from 1.2 % to 2.5 % depending on the defined block volume and the encountered cumulative basal area in a forested transit zone. Where possible, the defined RPTV should be compared and validated by field recordings of SW.

International audience

Anticipating risks related to climate extremes often relies on the quantification of large return levels (values exceeded with small probability) from climate projection ensembles. Current approaches based on multi-model ensembles (MMEs) usually estimate return levels separately for each chain of the MME. By contrast, using MME obtained with different combinations of general circulation model (GCM) and regional climate model (RCM), our approach estimates return levels together from the past observations and all GCM-RCM pairs, considering both historical and future periods. The proposed methodology seeks to provide estimates of projected return levels accounting for the variability of individual GCM-RCM trajectories, with a robust quantification of uncertainties. To this aim, we introduce a flexible non-stationary generalized extreme value (GEV) distribution that includes i) piecewise linear functions to model the changes in the three GEV parameters ii) adjustment coefficients for the location and scale parameters to adjust the GEV distributions of the GCM-RCM pairs with respect to the GEV distribution of the past observations. Our application focuses on snow load at 1500 m elevation for the 23 massifs of the French Alps, which is of major interest for the structural design of roofs. Annual maxima are available for 20 adjusted GCM-RCM pairs from the EURO-CORDEX experiment, under the scenario RCP8.5. Our results show with a model-as-truth experiment that at least two linear pieces should be considered for the piecewise linear functions. We also show, with a split-sample experiment, that eight massifs should consider adjustment coefficients. These two experiments help us select the GEV parameterizations for each massif. Finally, using these selected parameterizations, we find that the 50-year return level of snow load is projected to decrease in all massifs, by −2.9 kN m−2 (−50 %) on average between 1986–2005 and 2080–2099 at 1500 m elevation and RCP8.5. This paper extends to climate extremes the recent idea to constrain climate projection ensembles using past observations.

This article discusses the issue of snow avalanches monitoring at regional scale in the French Alps using SAR observations from Sentinel-1. A color space segmentation technique applied to SAR image time series has recently been used to detect areas of avalanche debris and successfully evaluated in the Swiss Alps. The objective of this paper is to generalize the detection of avalanche debris to all the French massifs and to develop indicators derived from our satellite product Sentinel-1 that can provide information about the avalanche activity at the spatial scale of the massifs. Evaluations are also planned using model outputs, in-situ measurements and very high resolution optical satellite observations.

International audience; Determining avalanche activity corresponding to given snow and meteorological conditions is an old problem of high practical relevance. To address it, numerous statistical forecasting models have been developed, but intercomparisons of their efficiency on very large datasets are seldom. In this work, an approach combining random forests with class-balancing is presented and systematically compared with competing methods currently described in the avalanche literature. On more than 50 years of daily avalanche observations, in the 23 massifs of the French Alps, the competing classifiers are evaluated on their ability to distinguish three classes of avalanche activity: non-avalanche days, days with moderate activity, and days with high activity. Moreover, the variables of higher importance in the random forest classifiers are shown to be coherent with current avalanche literature and a clustering based on these variable importance separates massifs which are known to have different avalanche activities. Our approach opens perspectives to support operational avalanche forecasting.