Avalanche Statistics: What 6,659 Recorded Avalanches Tell Us by Aspect, Altitude & Month

01The dataset: 6,659 recorded avalanches, not anecdotes

Across 6,659 geolocated avalanche records from the Alps and Pyrenees — 3,210 accidents with people involved and 3,449 avalanches observed in France — the pattern is sharp: recorded avalanches concentrate on shady northerly aspects, in the 2400–2800 m mid-altitude band, in the depth of winter — January most of all.

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 6,659 geolocated avalanche records, pooled from four national sources that do not all record the same thing:

• France — data-avalanche.org: 3,449 avalanches observed and reported by its network of observers. Some involved people, but the export carries no field saying which and no casualty count — so these are observations, not accidents.
• Austria — LAWIS, the database of the Austrian avalanche warning services: 1,219 accidents with people involved, covering the last five years, with the number of deaths.
• Switzerland — the WSL Institute for Snow and Avalanche Research (SLF): 1,478 accidents with at least one person caught, with the number of deaths.
• Italy — AINEVA, the Interregional Snow and Avalanche Association: 513 accidents with people involved, with the number of deaths — drawn from AINEVA's whole public archive, of which we keep the last ten years.

Each record carries a position, an altitude, a slope aspect, a date and — for the Austrian, Swiss and Italian records — the number of deaths: 509 recorded deaths in 368 fatal accidents (the gap is multi-victim accidents), 107 in the Austrian records, 192 in the Swiss ones and 210 in the Italian ones. That is not a death toll for the Alps: the French records carry no casualties, and the windows differ — roughly the last decade for France, Switzerland and Italy, the last five years for Austria, the period in which terrain, equipment and the backcountry crowd most resemble today.

This is not a census of all avalanches — most slides happen with nobody near them and are never logged. It pools accidents (avalanches that intersected people) and observed avalanches (slides someone in France saw and reported), a distinction we come back to at the end. As a map of where avalanches meet the people who travel and watch the mountains, it is about as good as it gets — and, because it is our own structured data, a generic ski blog cannot reproduce it.

Use the interactive profile below to slice the dataset by aspect, altitude band and month, and to toggle between all records and the fatal subset (Austrian, Swiss and Italian accidents).

02The avalanche profile, three ways

The instrument below is the same avalanche-profile engine Snow Trace embeds, sliced per massif, in its regional ski-touring guides — here it shows the full picture. Its three tabs project the same records onto three axes:

• By aspect — a compass rose of where, on the 360° of slope orientation, recorded avalanches cluster.
• By altitude — the band of elevation where they released.
• By month — the shape of the season.

Toggle Fatal only to keep only the accidents in which someone died — and watch which parts of the shape hold and which do not.

Fig. 01 — Switch axes and toggle the fatal subset: recorded avalanches are northerly, mid-altitude and deep-winter; fatal accidents keep the northerly lean but sit lower, with a heavier high-altitude tail and a February rather than January peak.

03Aspect: the danger is northerly

The single sharpest signal in the whole dataset is aspect. Recorded avalanches are not spread around the compass — they pile up on the northerly half of the rose. The busiest single aspect is N, and the N–NE–NW arc alone — three of the eight aspects — holds 54% of the records with a known aspect, far more than the sunny southern aspects. The lean is the same in the accidents and in the French observations taken separately.

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 grows on the cold snow surface on clear nights — 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 records, 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 Dominate the Avalanche Record

The aspect finding is the most portable number on this page. Altitude bands shift between ranges and months move 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 stand in. That is why the shady-aspect signal survives averaging across 6,659 records when finer detail washes out.

Two cautions for anyone quoting it. "Northerly" is a band, not a bearing: treat the whole N–NE–NW arc as suspect, because a slope does not become safe by facing north-north-east instead of due north. And the rose is a default, not today's verdict: wind turns any aspect into a lee slope, and one storm from an unusual direction can load the sunny side and strip the shady one.

05Altitude: a crowded mid-altitude band

Recorded avalanches also concentrate in altitude. The peak band is 2400–2800 m — and the broader 2000–2800 m range holds about 66% of the records with a known altitude. The accident records alone sit a little lower than the French observations.

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 avalanches are recorded where the most people are — skiing and watching.
• 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 January is the statistical centre of gravity of the entire record.

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

The altitude finding is the easiest to misread. The 2400–2800 m band is where recorded avalanches peak, and the 2,000–2,800 m range holds about 66% of them — a statement about where avalanches were recorded, not about where danger is. Higher up on the glaciers the count falls, yet nobody sensible calls that terrain safe: the fatal accidents below carry a heavier high-altitude share than the record as a whole. The count falls because fewer people go there and the high snowpack behaves differently, not because the mountain forgives you.

If you pull one figure to set next to the aspect number, pair them rather than rank them: the centre of gravity of the whole record is a north-facing slope in the 2,000–2,800 m band, and the two axes only mean something read together.

07Month: the deep-winter peak

Finally, time. The season has a clear shape: records ramp up through early winter, peak in mid-winter (the busiest single month in the record is January), then taper through spring — with a distinct, separate spring-hazard story. The January peak comes from the French observations; among the accidents alone, the deep-winter months sit closer together, with February slightly ahead.

• December–March carries 81% of the records. 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 loose and wet slab 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 records are rare and skew to high 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 January Tops the Avalanche Calendar

The calendar is the axis people most want to turn into a rule — "ski early season", "wait for spring" — and the one that punishes rules hardest. December through March carries 81% of the record, and the deaths follow the same season a little later: December–March holds 80% of the fatal accidents, and the busiest fatal month is February.

The useful reading is not "avoid January". It is that the type of problem tracks the calendar even when the danger level does not. In deep winter the trigger and the cause are separated in time: a slide on facets buried weeks earlier, under a slope that feels settled underfoot. By April the same posted rating means wet snow, managed by the clock rather than by a column test. January being the peak is a fact about weak layers, traffic and observers overlapping, not a month to cross off the calendar.

09Fatalities: the same northerly lean, a different altitude and calendar

Toggle the profile to Fatal only — and read it with one caveat first. The fatal subset is 368 accidents drawn from the Austrian, Swiss and Italian records — the only sources in our dataset that carry deaths. So the comparison is partly French observations versus Alpine fatal accidents, not one population split in two.

With that in mind, the aspect signal holds: 62% of fatal accidents sit on the N–NE–NW arc, and N is again the single deadliest aspect. Shaded slopes preserve persistent weak layers, and those slides tend to be deep, wide and triggered from terrain that looked manageable.

The other two axes do not simply sharpen. Fatal accidents peak in the 2000–2400 m band rather than 2400–2800 m, and they carry a heavier high-altitude tail: 13% of fatal accidents with a known altitude released above 3,200 m, against 6% of all records. And the calendar shifts: December–March holds 80% of fatal accidents (against 81% of all records), but the busiest fatal month is February, not January. There is no "safe corner" of the distribution — quiet bars on the full record can still hold deaths.

Across the Austrian, Swiss and Italian records there are 509 deaths. Survival, when an avalanche does bury 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 avalanche statistics honestly

Data like this is powerful and easy to misuse. Four 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 these records — and the French observations add a third factor: where observers are. 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 avalanches are not a forecast. This page describes the long-run climatology of recorded avalanches. 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 records 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 avalanche 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, or a day out with a mountain guide.

Fig. 02 — North has 1,257 recorded avalanches to South's 435: read as counts, North looks 2.9x worse. But a 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 2.9:1 — 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 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. 6,659 geolocated records of two kinds: 3,210 accidents with people involved — LAWIS (Austria, 1,219), the WSL/SLF (Switzerland, 1,478) and AINEVA (Italy, 513) — and 3,449 avalanches observed and reported in France through data-avalanche.org, which carries no casualty field. Each record has a position, a date, an altitude, a slope aspect, a source tag, a place label and — for the Austrian, Swiss and Italian records — a death count. Exact duplicates are removed before counting. The charts are plain aggregations of those fields: no modelling, no weighting, no smoothing.

*What each record is not. The register does not carry the posted danger rating for the day, the trigger, the time of day, the burial depth, or a normalised per-country* breakdown. So the danger level most accidents happen at, or the share triggered by the party, are figures from the avalanche-research literature, not from these 6,659 records — and we keep the two separate.

What it can tell you. With this many records, the gross patterns are trustworthy: the northerly concentration, the mid-altitude band and the deep-winter peak are large, stable signals that show up in the accidents and in the French observations alike.

What it cannot tell you. It cannot give you a rate, because there is no matching denominator: we know where avalanches were recorded, 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 avalanches were recorded 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 source is built from avalanches someone witnessed and reported, each with its own thresholds and completeness, over different windows (about ten years for France, Switzerland and Italy, five for Austria). Slides in remote terrain, minor burials with no injury and avalanches in lightly travelled ranges are under-counted. Thin data in a corner of the rose or the calendar means few people or few reports there, not proven safety.

Why this is not a risk map. This page describes the past, aggregated, with no denominator and known reporting gaps: a climatology of recorded avalanches, 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, thin inner-Alpine snowpack of the Queyras facets differently from the deeper, milder snowpack of the northern Pre-Alps, open to Atlantic westerlies; the Pyrenees have their own quirks again. That is why the same avalanche-profile engine runs per massif — sliced to a region's bounding box — inside Snow Trace's regional guides, such as the Vanoise or the Écrins, so the signal you act on is the one for the terrain under your skis, not a continental average.

Then hold your route against it. Snow Trace route topos give each line's main aspect and high point: in Belledonne, Dôme de la Coche via the Pas de la Coche faces NW and tops out at 2,134 m, inside the shaded arc; Croix de Chamrousse via Lac Achard faces S at 2,250 m; Croix de Belledonne via the Doménon lakes faces W at 2,926 m.

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 avalanche 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; there is nothing to buy and nothing to license. A few asks, so the numbers travel accurately.

Quote the headline as what it is: 6,659 geolocated avalanche records from France, Austria, Switzerland and Italy — 3,210 accidents with people involved (AT/CH/IT) and 3,449 avalanches observed in France, 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 January peak — please carry the one-line caveat with them: these are record counts, a climatology, not a per-slope risk and not a forecast. If you cite deaths, say they are the 509 in the Austrian (last five years), Swiss and Italian records only — the French records carry none.

A suggested credit line: "Snow Trace avalanche dataset — 6,659 recorded avalanches (FR observations, AT/CH/IT accidents) — snow-trace.com." Link back to this page where you can, so readers can slice the profile themselves and read the limitations in full.