| 14 Jul 2026
How exposed is a mountain road to snow avalanches? A corridor-scale assessment in the southern Andes
Abstract. The International Route 115-CH (Pehuenche Pass, ~36° S, Chile) is a key Andean border crossing that is closed repeatedly each winter by snow and adverse-weather hazard. Yet no calibrated, corridor-scale assessment of avalanche exposure exists for any mountain road in the southern Andes, where systematic avalanche records are largely absent. We combine back-analysis of documented avalanches with corridor-wide simulation in the open-source AvaFrame model (com1DFA). After calibrating friction on two documented paths, we simulate 208 release areas along the 61 km corridor under three snow-depth scenarios. Simulated avalanches reach up to 40 % of the corridor. Of all road crossings, 77 % exceed the 30 kPa vehicle-damage threshold and 41 % exceed the 100 kPa vehicle-destruction threshold, and avalanches typically arrive within a minute of release (median 25 s), leaving little time for evacuation. Mapping these arrival times, the first such analysis for the route, reveals the few segments where preventive closure can substitute for permanent structural protection. Exposure is highly non-uniform: six hot-spot segments, about 7 km (11 %) of the route, concentrate the highest path convergence and define where mitigation would most reduce collective risk. Key limitations are the friction calibration anchored on two documented paths, the 12.5 m terrain model, and the exposure-based framing: results are conditional on avalanche release and do not represent annualized risk. However, using only open-source tools and public data, the workflow is reproducible and transferable to other data-scarce mountain corridors in the Andes and beyond.
Competing interests: Felix Oesterle is one of the developers of the open-source AvaFrame simulation framework used in this study, as declared in the competing interests section of the manuscript. No other competing interests are present.
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.
Topic
The authors tackle the practically relevant question of how meaningful hazard/risk analyses can be elaborated in mountain regions without comprehensive records of earlier avalanche activity. They focus on linear infrastructure and use the Chilean sector of the mountain road over Pehuenche Pass as a case study. An important aspect for future wide adoption of their approach is the use of modern open-source software like quasi-3D flow simulation codes. The authors' approach comprises (i) identification of potential release areas (PRAs – already provided for the case study), (ii) an estimate of fracture depth as a function of return period and altitude, based on measurements from a weather station in the study area, (iii) calibration of a dynamical avalanche flow model against recorded events, (iv) simulation of the run-out from the PRAs with these parameters, including some sensitivity assessment, and (v) analysis of the number of paths hitting each road segment, the corresponding impact pressures and arrival times to identify the hazard hotspots and potential mitigation measures.
General interest, topical fit for NHESS, novelty
The problem of assessing snow avalanche hazard in data-scarce regions arises in the majority of mountain areas on Earth where snow avalanches can occur – even in Europe, there are mountain regions where avalanche events are not routinely registered. The manuscript should therefore be of interest to many readers of NHESS, even though a methodology developed for snow avalanches cannot directly be applied to other hazards like landslides or debris flows. Moreover, the authors' emphasis on open-source tools is a perfect fit for an open-access journal like NHESS.
The proposed procedure combines several elements, each of which is well-known and has been used and described by other authors in various contexts. The main achievements in this manuscript are (i) recognizing the need for natural hazard/risk assessments in areas lacking observational data and the feasibility of doing this in a more rational and technically up-to-date way using modern open-source tools, (ii) proposing a detailed method adapted to the available data, which should also be applicable in other areas, (iii) demonstrating the method in a relevant case study, and (iv) a critical discussion of the limitations, advantages and shortcomings of the proposed method. From a scientific point of view, this manuscript represents a fairly limited degree of novelty, but it has significant practical value for Chile, Argentina and other countries with exposed traffic routes or settlements but limited economic resources for mitigation of natural hazards.
Major issues
There are two main aspects of this work that leave me – and presumably many future readers as well – a little unsatisfied after reading it. The authors are clearly aware of these weak points, and their arguments for their choices are understandable. Yet, I think there would be room for improvement.
The first point concerns the difference between hazard and risk. One of the authors' goals is to provide a rational basis for optimizing the cost-effectiveness of future mitigation measures along the corridor. This is clearly a problem requiring knowledge of the spatial distribution of risk. The authors seem to have aimed for a middle ground:
While each of these assumptions is rather crude, it is possible that the errors they cause cancel to some degree. The first assumption may overestimate the release probability by an order of magnitude. (It needs to be admitted, though, that the well-known Swiss guidelines for hazard mapping also make this assumption.) Second, if persistent weak layers are a frequent problem in this part of the Andes, the fracture height may often be much larger than the new-snow depth. This may even affect the choice of friction parameters in the run-out calculation as the latter are assumed to vary discontinuously with volume. Third, there likely is a pronounced dependence of release height on aspect due to wind loading or depletion of the release area. Accounting for this is, admittedly, a difficult task, but experience of the locals might give valuable hints. In addition, if the weather station near the road records wind speed and direction, one could determine the dominant direction(s) of strong and persistent winds and modify the snow and new-snow heights manually for each PRA, accounting for the angle between slope aspect and wind direction. With about 200 PRAs, this looks like a manageable effort.
Incidentally, I may have misunderstood the criterion used by the authors for identifying run-out simulations that affect the road (lines 242–247), or it is overly cautious: A buffer width of 50 m from the road centerline combined with a peak-pressure threshold of 10 Pa may mean that especially small avalanches releasing close to the road but with short run-out will count as hits even though they never would be able to reach the road in reality. The effect of such a criterion on the overall hazard picture may also depend quite sensitively on the stopping criteria implemented in the numerical model because many Voellmy-type models exhibit creeping unless the simulation is stopped by time, velocity or momentum-fraction criteria.
The second issue – calibration of the run-out model com1DFA – is the crux of the problem, as the authors emphasize. A satisfactory solution to the problem would require substantially more well-documented events. If there are multiple events in a path (here: two events each in Campanario and La Biblioteca), dividing them into one set for calibration and one for validation is generally a good strategy, but it becomes questionable if each subset contains only a single event. Nevertheless, this works well at La Biblioteca with two (humid) avalanches of similar size released under similar conditions, and the resulting friction coefficient μ = 0.31 looks reasonable. The authors' approach is, however, problematic for the Campanario avalanche because one event is a large wet-snow avalanche and the other a small dry mixed-snow event. It is therefore not too surprising that μ = 0.20 fitted both events.
Given the volumes of the four events, the authors infer that "large" avalanches (> 10,000 m3) should be modeled with μ = 0.20 and "small" ones with μ = 0.31. While there is reasonable empirical support for diminishing μ with increasing release volume, the authors' conclusion rests on very weak foundations: First, the second Campanario avalanche with a volume of 9,100 m3 is only about three times larger than the small La Biblioteca events and falls below the proposed threshold of 10,000 m3, so it should be simulated with μ = 0.31 according to the authors' recipe. Second, it is well known that dry-snow avalanches can have significantly lower friction than wet-snow ones of comparable volume. Third, an abrupt increase of μ by 0.11 at the transition from "large" to "small" avalanches is not physically plausible and may even induce distortions in the simulation results if there are many events near the threshold volume.
In my opinion, the conclusions one can tentatively draw from these four events and general experience with numerical run-out models are the following:
The calibration proposed by the authors has three parameters, μsmall = 0.31, μlarge = 0.20 and Vthr = 10,000 m3. If one wishes to stick to a three-parameter prescription, the three tentative conclusions above could be captured by a formula like
μwet(V) ~ μwet,1 − d · lg(V / 1 m3) and μdry(V) ~ μdry,1 − d · lg(V / 1 m3)
with μwet,1 ~ 0.50, μdry,1 ~ 0.44 and d ~ 0.06. This gives μ ~ 0.29 for the two events at La Biblioteca, μ ~ 0.21 for the wet-snow avalanche at Campanario and μ ~ 0.20 for the dry-snow event. Of course, this interpolation is purely speculative, but it reproduces the values found in the back-calculations fairly well, extrapolates reasonably to very large and very small events, and it avoids the spurious jump at 10,000 m3. A more detailed calibration using AIMEC will modify μwet,1, μdry,1 and d somewhat, but it should be interesting to compare the overall picture with this calibration to the one obtained in the manuscript.
Minor issues
A number of small remarks can be found in the annotated manuscript. Those remarks are an important part of this report and should be considered carefully by the authors. In particular, the figures will need some work to make them easier to read.
Presentation: language, figures, citations
The manuscript is clearly and suitably structured. I appreciate that the authors clearly point out potential weaknesses of their approach instead of shoving such issues under the rug. The language holds a high standard throughout and is easy to read. Tables collect the properties of events used in the analysis, simulation parameters, and impact-pressure statistics.
The figures illustrate important points of the work and are generally informative. However, they require some careful improvement because many are hard to read due to small size or too much information; see the annotated manuscript for more specific comments.
The reference list is carefully compiled, including DOIs for the readers' convenience. the choice of references is adequate – perhaps somewhat on the copious side. The number of self-citations is commendably low. For one or two small remarks, see the annotated manuscript.
Recommendation to the editor
Giving a clear recommendation is not easy in this case. On the one hand, the authors have identified a definite practical problem affecting many countries, especially in less developed regions of the planet, and they developed a methodology that should be applicable or readily adaptable in many places and by many consultants and authorities. Moreover, with small exceptions, the paper is well written and many of the limitations of this approach are discussed. In addition, scientific publications on snow avalanches in the Andes are preciously few. All these are strong points favoring acceptance of the manuscript.
On the other hand, the novelty of this manuscript is limited to combining mostly well-known, simple approximations to complex processes and quantities like avalanche release probability, fracture height or calibration of friction coefficients with one of the more modern open-source run-out simulation programs. The prescription for choosing the friction parameters for each avalanche path – a crucial step in this endeavor – has some shortcomings and internal inconsistencies. The analysis of the simulation results, while focusing on the spatial distribution and severity of the avalanche hazard plus the feasibility of non-structural mitigation measures, is solid in most respects, but it stops short of a realistic evaluation of the risk, which could be done with moderate extra work, at least at the level of individual risk.
Weighing the criteria of scientific novelty vs. practical relevance and considering the improvements that can be made with a realistic amount of work, I recommend to reconsider the manuscript after a major revision. Such a revision should, in a suitable way, make the calibration of friction parameters logically consistent and more realistic. What can be done with regard to release heights and probabilities will depend on the availability of different types of data and of expert advice, so this must be left to the authors. It is, however, desirable, that the analysis be made ready for use in a risk analysis or, even better, include an estimate of individual risk for a car or bus passenger passing this road in winter. I believe that this would make the paper more interesting and compelling and at the same time increase its impact.
Stord, Norway, 2026-08-07
Dieter Issler