What 7,954 Avalanche Accidents Tell Us — Aspect, Altitude & Month

01The dataset: 7,954 real accidents, not anecdotes

Across 7,954 real, geolocated avalanche accidents in the Alps and Pyrenees, the pattern is sharp: accidents and deaths concentrate on shady northerly aspects, in a narrow 2400–2800 m mid-altitude band, in the depth of winter — February most of all. That is the finding in one line; the rest of this page is the evidence, the caveats, and what to do with it.

Most avalanche advice is handed down as lore — "stay off north faces", "watch the mid-winter facets", "spring is for early starts". The lore is mostly right. But you do not have to take it on faith. Snow Trace maintains a structured database of 7,954 geolocated avalanche incidents, pooled from the official accident records of four alpine nations:

• France — data-avalanche.org (the reference French accident register).
• Austria — LAWIS, the national avalanche information service.
• Switzerland — the WSL Institute for Snow and Avalanche Research (SLF).
• Italy — AINEVA, the inter-regional snow-and-avalanche network.

Each record carries a position, an altitude, a slope aspect, a date, and — where the source reports it — the number of fatalities. Across the whole set there are 1,744 recorded deaths in 1,318 fatal incidents (the gap is multi-victim accidents). The records span roughly the last decade, the window in which terrain, equipment and the size of the backcountry crowd most resemble today.

This is not a survey of avalanches — most slides happen with nobody near them and are never logged. It is a register of accidents: avalanches that intersected people. That distinction matters for how you read it, and we come back to it at the end. But as a map of where the human cost of avalanches actually falls, it is about as good as it gets — and, because it is our own structured data, it is something a generic ski blog simply cannot reproduce.

Use the interactive accident profile below to slice the whole dataset three ways — by aspect, by altitude band, and by month — and to toggle between all incidents and the fatal subset. Then read on for what each axis is telling you.

02The accident profile, three ways

The instrument below is the heart of this page. It is the same accident-profile engine Snow Trace embeds, sliced per massif, into each of its regional ski-touring guides — here it shows the full, national picture.

Switch between the three tabs to see the same accidents projected onto three different axes:

• By aspect — a compass rose of where, on the 360° of slope orientation, accidents cluster.
• By altitude — the band of elevation that catches people.
• By month — the shape of the season's risk.

Toggle Fatal only to strip out the incidents nobody died in and watch how the shape changes — or, tellingly, how often it doesn't.

Fig. 01 — Switch axes and toggle the fatal subset: the danger is northerly, mid-altitude and deep-winter, and the fatalities follow the same shape.

03Aspect: the danger is northerly

The single sharpest signal in the whole dataset is aspect. Accidents are not spread around the compass — they pile up on the northerly half of the rose. The most accident-prone single aspect is N, and the N–NE–NW arc together dwarfs the sunny southern aspects.

Why? Two compounding reasons, both rooted in snow science:

1. Shaded slopes preserve weak layers. North-through-east aspects in the northern hemisphere get the least winter sun. Cold, dry snow there metamorphoses into faceted, sugary grains and surface hoar — the classic persistent weak layers that linger for weeks under later snowfalls. Sunny aspects warm, settle and bond; shady aspects keep their booby-traps. (See Aspect & Elevation and Understanding Snowpack Stability.)
2. Skiers chase the cold snow. The same shade that preserves weak layers also preserves good skiing — cold, light powder. So the slopes most likely to harbour a persistent weak layer are exactly the slopes powder-hungry skiers are most drawn to. The hazard and the temptation share an aspect.

Wind reshuffles this picture day to day — lee slopes load regardless of compass bearing — but averaged over thousands of accidents, the shady-aspect signal is unmistakable. If you take one number off this page, take this one: the danger lives on the cold side of the mountain.

04Why North-Facing Slopes Have the Most Avalanche Accidents

A useful way to hold the aspect finding is that it is the most portable number on this page. Altitude bands shift between ranges, and months move around with the weather — but the physics that loads the cold half of the compass (less winter sun, colder snow, weak layers that refuse to heal) does not care which valley you are standing in. That is why the shady-aspect signal survives being averaged across 7,954 accidents when almost every finer detail washes out.

Two things are worth saying out loud for anyone quoting this. First, "northerly" is a band, not a single bearing: the worst single point on the rose sits inside a broad N–NE–NW arc, and the sensible reading is to treat that whole cold quadrant as suspect rather than to fixate on one bearing. A slope does not become safe because it faces north-north-east instead of due north. Second, the rose is a default, not a verdict for the day. Wind turns any aspect into a lee slope, and a single storm from an unusual direction can load the sunny side and strip the shady one. The climatology tells you which way to lean before you know anything else; it does not tell you which slope is loaded this afternoon.

The blunt version — the one worth putting on a slide — is this: over the long average, the danger lives on the cold side of the mountain, and it does so for reasons of snow physics that are unlikely to change from one winter to the next.

05Altitude: a narrow killing band

Accidents also concentrate in altitude. The peak band is 2400–2800 m — and the broader 2000–2800 m range accounts for the large majority of incidents with a recorded altitude.

This is not because higher is automatically safer. It is the intersection of three things:

• Exposure. This is the altitude band where most ski-touring terrain lives in the Alps and Pyrenees — above the treeline, below the big glaciated peaks. The most accidents happen where the most people are.
• Snowpack. It is high enough to hold cold, faceting snow through mid-winter, but low enough that the pack is shallower and more variable than on the high glaciers — and a shallow snowpack facets faster and is easier to trigger from.
• Treeline effect. Right around the treeline, slopes are open enough to slide but tempting enough to ski, and wind-loading is often at its most erratic.

Read the altitude axis alongside aspect, not instead of it: a north-facing slope at 2,600 m in February is the statistical centre of gravity of the entire accident record.

06Why 2,000–2,800 m Is the Avalanche Accident Band

The altitude finding is the one most easily misread, so it is worth stating carefully. The 2,400–2,800 m band is where accidents peak, and the broader 2,000–2,800 m range holds the large majority — but that is a statement about where accidents are, not about where danger is. Move a thousand metres up onto the glaciers and the accident count falls, yet nobody sensible calls that terrain safe. The count falls because fewer people are there and because the high snowpack behaves differently, not because the mountain forgives you.

The honest one-liner for this axis: the mid-altitude band is the accident band because it is where the terrain, the snowpack and the crowd all overlap. It is high enough to hold cold, faceting snow deep into winter; low enough that the pack is thinner and more variable than on the high glaciers, which lets it facet faster and tip more easily; and it is the elevation where most ski touring actually happens, above the trees and below the serious glaciation.

If you are pulling a single figure to set next to the aspect number, pair them rather than rank them. The centre of gravity of the whole record is not "north" and it is not "2,600 m" — it is a north-facing slope in the 2,000–2,800 m band, and the two axes only mean something when you read them together.

07Month: the deep-winter peak

Finally, time. The season's risk has a clear shape: it ramps up through early winter, peaks in mid-winter (the busiest single month in the record is February), then tapers through spring — with a distinct, separate spring-hazard story.

• December–March carries the overwhelming majority of accidents. This is when persistent weak layers are deepest and most active, storms are frequent, and the touring crowd is largest.
• The mid-winter peak is a persistent-weak-layer season: the dangerous slides are the ones triggered remotely on buried facets, days or weeks after the snow that loaded them. (See Avalanche Problem Types.)
• April–May never reaches the winter peak, but the character of the hazard changes completely: wet-snow and loose-wet avalanches driven by the daily melt–freeze cycle. The management shifts from "avoid the buried layer" to planning the tour so you are off the slope before it warms. (See Spring Corn & the Melt–Freeze Cycle.)
• Summer–autumn incidents are rare and mostly involve high-altitude glaciated terrain.

The lesson is not "ski in spring instead". It is that the same danger level means a different problem in January than in April — and the calendar is your first clue to which.

08Why February Is the Deadliest Month for Avalanches

The calendar axis is the one people most want to turn into a rule — "ski early season" or "wait for spring" — and it is the one that punishes rules the hardest. February is the busiest single month in the record, and December through March carries the overwhelming majority of accidents, but the useful reading is not "avoid February." It is that the type of problem tracks the calendar even when the level of danger does not.

Here is the version worth quoting. In deep winter the accidents are a persistent-weak-layer story: slides triggered on buried facets, often remotely, often days or weeks after the snow that loaded them fell. That is what makes mid-winter both the busiest and the least intuitive season — the trigger and the cause are separated in time, so the slope that catches you can look, and feel, settled underfoot. By April and May the count has dropped, but the hazard has not disappeared; it has changed species, into wet-snow and loose-wet slides driven by the daily melt–freeze cycle, managed by the clock rather than by a column test.

So the honest takeaway from the month axis is a translation, not a permission slip: the same posted danger rating means "trust nothing you cannot see" in January and "be off the slope before it warms" in April. February being the peak is a fact about weak layers and traffic overlapping — not a month to cross off the calendar.

09Fatalities follow the same pattern — only sharper

Toggle the profile to Fatal only and the picture barely moves — it just intensifies. The fatal subset clusters on the same northerly aspects, in the same mid-altitude band, in the same deep-winter months as the full incident set. The factors that make a slope likely to produce an accident are the same factors that make that accident likely to be deadly.

That alignment is itself a finding. It means there is no "safe corner" of the accident distribution that fatalities avoid — the deadliest accidents are not freak outliers in unusual terrain, they are the typical accident, on the typical slope, taken to its conclusion. The persistent-weak-layer slide on a shady mid-winter face is both the most common accident and the most lethal one, because those slides tend to be deep, wide, and triggered from terrain that looked manageable.

Across the record there are 1,744 fatalities. Survival, when an avalanche does catch you, is overwhelmingly a question of how fast your partners dig you out — a function of the rescue gear you carry and the practice behind it. (See The Avalanche Survival Curve.)

10How to read accident statistics honestly

Data like this is powerful and easy to misuse. Three cautions, so you draw the right conclusions:

1. This is exposure plus hazard, not hazard alone. A slope catches people when it is both dangerous and visited. The mid-altitude, north-facing, mid-winter peak is partly a story about where the snowpack is treacherous — and partly a story about where skiers go. You cannot fully separate the two from accident data. The honest reading is: this is where the combination of risk and traffic produces the most harm — which is exactly the combination you, as a skier, are part of.

2. Past accidents are not a forecast. This page describes the long-run climatology of accidents. It does not tell you anything about today. A south-facing slope is not "safe" because the rose is lighter there — on a sun-driven spring afternoon it can be the only dangerous slope on the mountain. Always read your **local bulletin** for the current aspects, elevations and problems. The statistics tell you which way to lean by default; the bulletin tells you what is true now.

3. Absence in the data is not safety. Aspects, altitudes and months with few accidents may simply be places and times few people go. Thin data is not the same as low risk.

4. Aspect and altitude are not the whole of terrain. These charts say nothing about slope angle — the most direct terrain trigger of all. A north-facing slope at 2,600 m is only a candidate; whether it can actually slide is mostly a question of how steep it is. No accident chart substitutes for measuring the steepness of the slope under your skis — or for learning to assess the snowpack yourself. (See Slope Angle & Avalanche Terrain.)

Used correctly, the value of this page is calibration. It sets your priors: when nothing else is known, expect the danger to be northerly, mid-altitude and mid-winter. Then let the bulletin, the weather and your own observations update those priors for the day in front of you — the habit of running that update deliberately is exactly what a structured avalanche course is for.

Fig. 02 — North has 2,284 recorded accidents to South's 415: read as counts, North looks 5.5x worse. But an accident count is hazard times traffic, and everyone chases the cold north powder. Drag in the missing north:south traffic ratio and the risk-per-outing verdict crosses over at 5.5:1 — below it North looks deadlier, above it South does. Same real counts, opposite conclusion, because the denominator that would turn a count into a rate is one the accident record simply does not carry.

11Methodology and limitations

If you are going to cite this page, cite it accurately. Here is exactly what is under the hood, and what it will and will not support.

What each record is. The dataset is 7,954 geolocated avalanche accidents pooled from four national registers: data-avalanche.org (France), LAWIS (Austria), the WSL/SLF (Switzerland) and AINEVA (Italy). Each incident carries a longitude and latitude, a date (as a day count), an altitude, a slope aspect, a source tag, a name or place label, and a fatality count where the source reports one. The aspect, altitude and month charts on this page are plain aggregations of those fields — no modelling, no weighting, no smoothing. What you see is a count of where the records fall.

*What each record is not. Deliberately, so you do not over-read the page: the register does not carry the posted danger rating for the day, the trigger (who or what released the slide), the time of day, the burial depth, or a normalised per-country* breakdown. So when this page talks about the danger level most accidents happen at, or the share triggered by the party, those are figures from the wider avalanche-research literature — not something we can derive from these 7,954 records. We keep the two clearly separate.

What it can tell you. With this many accidents, the gross patterns are trustworthy. The northerly concentration, the mid-altitude band and the deep-winter peak are large, stable signals that survive being averaged across four countries and roughly a decade. If a pattern is that big and that consistent, it is telling you something real about where the human cost of avalanches falls.

What it cannot tell you. It cannot give you a rate, because there is no matching denominator: we know where accidents happened, not how many skier-days passed on each aspect, altitude or month without incident. So the charts show counts, never probabilities. A tall bar means "many accidents happened here," which blends genuine hazard with plain foot traffic — you cannot read a bar as "this is how likely you are to be caught."

Selection and reporting bias. Every register is built from accidents that were witnessed, survived to be reported, or serious enough to be recorded, and the four national sources do not all record with the same completeness or the same thresholds. Accidents in remote terrain, minor burials with no injury, and slides in lightly-travelled ranges are all under-counted relative to well-populated, well-documented ones. Thin data in a corner of the rose or the calendar means few people or few reports there — not proven safety. Absence of accidents is not evidence of low risk.

Why this is not a risk map. Put those cautions together and the conclusion is firm. This page is a description of the past, aggregated, with no denominator and known reporting gaps. It is a climatology of accidents, useful for setting your default expectations. It is not a forecast, not a probability model, and not a map you can point at to call a slope safe or dangerous today. For that there is one source only: the current avalanche bulletin for your massif.

12From the national picture to your massif

The national rose is a starting point, not the end. Snowpack character varies enormously across the range: the cold, continental snowpack of the Queyras facets differently from the maritime snowpack near the coast; the Pyrenees have their own quirks again. That is why the same accident-profile engine on this page runs per massif — sliced to a region's bounding box — inside each of Snow Trace's regional ski-touring guides, so the aspect-altitude-month signal you act on is the one for the terrain under your skis, not a continental average.

Pair the statistics with the live picture: check the current snow depth and the official bulletin for your massif before every tour, and read the accident profile as the baseline those daily signals adjust. The numbers tell you where the risk has always concentrated. The bulletin tells you whether today is one of those days.

13How to cite this

Journalists, guides, instructors and fellow skiers are welcome to use these figures — that is what the page is for. Snow Trace is free and community-first, so there is nothing to buy and nothing to license. A few asks, so the numbers travel accurately.

Quote the headline as what it is: 7,954 geolocated avalanche accidents from France, Austria, Switzerland and Italy, spanning roughly the last decade, pooled from data-avalanche.org, LAWIS, the WSL/SLF and AINEVA. When you cite the patterns — northerly aspects, the 2,000–2,800 m band, the deep-winter and February peak — please carry the one-line caveat with them: these are accident counts, a climatology, not a per-slope risk and not a forecast. That single clause is the difference between using the data well and misleading a reader into treating a quiet corner of the chart as a safe one.

A suggested credit line: "Snow Trace avalanche accident dataset — 7,954 incidents, FR/AT/CH/IT — snow-trace.com." Link back to this page where you can, so readers can slice the profile themselves and read the limitations in full. And if you want the same profile for a single massif rather than the whole range, every regional guide on Snow Trace carries one.