Is SLABS machine learning or AI?

SLABS is a statistical model: a generalised additive model, a flexible form of logistic regression with four inputs, fitted to Swiss accident and GPS-track data. Every effect can be plotted and checked, and a gradient-boosting model tested on the same data did no better. It is not a neural network.

Does SLABS show where an avalanche would run out?

No. SLABS estimates how likely a skier at a spot is to trigger an avalanche that catches someone, from the slope around that spot and the bulletin. It has no runout or avalanche-path model, and spontaneous avalanches are left out. For runout, the SLF's CAT and ATH maps and ATES-based maps are built for that.

Does SLABS know today's conditions?

Only through the bulletin. Three of its four inputs are the bulletin's danger level, critical elevation and critical aspects, which is why BEYOND asks for the conditions before it runs. Nothing else about the snowpack, wind, new snow or tracks on the slope goes in. If the bulletin is off for your slope, SLABS is off too.

Does SLABS cover off-piste next to the lifts?

The model was built on Swiss ski-touring data. Accidents involving lift-served off-piste were excluded and track points within 200 m of a piste were removed, so the data describes touring terrain. We found no published test of SLABS for freeride terrain next to resorts or outside Switzerland. Treat its colours there as a starting point, not a verdict.

How well does it work?

On the Swiss data it separated accident spots from travelled spots better than the older reduction methods (AUC 0.943 against 0.873 for the Graphical Reduction Method). Staying on green would have prevented 80% of those accidents, so about one in five happened on green. No method catches every accident.

SLABS avalanche terrain analysis, explained

Contents

Short answer

SLABS is a statistical model published in 2024 by Skitourenguru with researchers from SLF, VITO and NTNU. It estimates where a skier is more likely to trigger a slab avalanche from four inputs: the steep part of the slope area within about 40 m (the 85th percentile of its angles), the bulletin's danger level, and the elevation and aspects the bulletin marks as critical. It has no runout model and does not see the snowpack on your slope.

Where SLABS comes from

SLABS stands for "Screening the Likelihood of Avalanches on Backcountry Ski tours". It was published in May 2024 by Bart Degraeuwe (VITO), Günter Schmudlach (Skitourenguru), Kurt Winkler (SLF) and Jochen Köhler (NTNU) in Cold Regions Science and Technology, open access under a CC BY licence. Skitourenguru has used it since winter 2024/25 to rate about 10,000 ski tours in the Alps every day; before that it used the older QRM method.

The authors compared 1,250 avalanche accidents on ski tours in Switzerland (winters 2001/02 to 2020/21, from the SLF accident database) with 8,558 ordinary ski tours recorded as GPS tracks (57,800 km) and asked what separates the places where people were caught from the places where people ski without incident.

The model is a generalised additive model, a flexible form of logistic regression with four inputs, fitted in R. Every effect in it can be plotted and checked, which is why Skitourenguru calls it a white-box method. A gradient-boosting model (XGBoost) tested on the same data in 2026 did no better. So SLABS is a statistical model fitted to accident data; it is not a neural network, and calling it "AI" would overstate it.

The four inputs

Only one of the four inputs is pure terrain. The other three come from the official avalanche bulletin, which is why SLABS changes from day to day and why BEYOND asks you for the conditions before it runs.

Input	What it is	Where it comes from
MSA40	The steep part of the slope area around you (the 85th percentile of its slope angles), bounded by terrain folds and forest, with a control radius of about 40 m	Terrain (elevation model)
Danger level	The danger level in the bulletin for the region, on the 1 to 5 scale	Avalanche bulletin
Elevation relative to the critical elevation	How far above or below the elevation the bulletin names as critical you are	Avalanche bulletin
Aspect overlap	How much of the slope faces the aspects the bulletin names as critical	Avalanche bulletin

What the colours mean

SLABS does not give a probability you can take at face value. Because many accidents are never reported, the authors use the model only for relative risk: how much more or less likely an accident is at this spot than on average. The range runs from about a thousand times lower to almost a hundred times higher than average.

The relative risk is cut into three classes: green, orange and red. The limits are set so that, in the Swiss data, staying out of orange and red would have prevented 80% of the accidents, and staying out of red 60%. The authors write that "there are no objective criteria to set such limits". About one in five accidents in the data happened at a green spot.

How BEYOND runs it

In BEYOND you draw an area of up to 8 × 8 km and enter today's conditions. The analysis then draws its layers on the 3D terrain: the SLABS risk, the MSA40 steepness, the overlap with the critical aspects, the terrain folds that bound a slope, the size of the release area, and the plain slope angle. You can inspect any point, change the opacity and run it again with other conditions. Plan curvature is an experimental extra.

When you draw a route, the map shows the SLABS risk along it, next to the elevation profile coloured by slope and the steepest section. Skitourenguru does the same for its route ratings; the difference is that BEYOND runs it for the area and the line you choose.

The SLABS layer in BEYOND is our own implementation of the published method. The map marks its output as experimental: it does not replace the official avalanche bulletin, training or judgement on the ground. Skitourenguru runs an operational variant, SLABS-1LR, which adds a "one level lower" rule for slopes outside the critical aspects and elevations. We do not use the ATHM terrain tiles or the accident dataset of Skitourenguru; those need their consent.

How it compares to other avalanche terrain maps

ATES (Avalanche Terrain Exposure Scale, version 2, 2025) is a static rating of terrain in five classes from 0 to 4, independent of the day's danger, and it includes exposure to avalanche paths and runout.

AutoATES v2 (2024) computes ATES classes per pixel from an elevation model and forest data, with a runout simulation. topo^alpin's avalanche terrain layer is computed with it. AutoATES has no snowpack or bulletin input.

The SLF's CAT and ATH maps in White Risk are static maps of release areas and runout, built for avalanches up to size 3. The SLF says they consider the terrain, not the current avalanche situation.

Skitourenguru's own ATHM is a static terrain map in four classes, from slope, slope size, slope shape and forest. It is the closest thing to a static "avalanche terrain" layer; SLABS is not static.

So SLABS differs in two ways: the bulletin goes in, and runout stays out. A red SLABS spot means a skier there is more likely to trigger a slab under today's bulletin; it does not say how far a slab would run or what lies below.

How well it works, and who checked it

On the Swiss data, SLABS separated accident spots from travelled spots better than the older rules it was tested against: an AUC of 0.943, against 0.926 for QRM, 0.873 for the Graphical Reduction Method and 0.858 for the Professional Reduction Method. At the same prevention rate it leaves more terrain open. This is a test on the data the model was built from, not an independent test in another region.

In 2025 the safety research group of the German Alpine Club (DAV) compared Skitourenguru's route ratings with the assessments of mountain guides at 565 rated spots in two Tyrolean areas over nine days. It found no systematic weaknesses, but at 8 of 44 clearly dangerous spots Skitourenguru showed green, and it tended to underestimate acute situations. Those ratings were computed with the predecessor method, QRM, not with SLABS; the study was done in cooperation with the developer, and DAV supports Skitourenguru financially. The Swiss accident prevention office BFU contributed to the publication of the paper and is Skitourenguru's main partner; it did not run a validation itself.

SLABS is peer-reviewed (2024). The DAV comparison of 2025 tested the predecessor method, and BFU ran no validation, so neither has validated SLABS itself.

What it doesn't show

There are accident points in green areas. The authors: "No method can guarantee absolute safety."
No runout: SLABS does not model how far an avalanche runs, what lies below a slope, or terrain traps.
Spontaneous avalanches are not in the model. SLABS is about avalanches that a skier triggers.
The snowpack on your slope, wind loading, recent tracks and danger signs such as whumpfing or shooting cracks are not inputs. If the bulletin is off for your slope, SLABS is off too.
Danger level 4 rests on only nine accidents in the data, so the model is weak there; level 5 was not considered. Wet-snow conditions are poorly covered.
The data is Swiss ski touring: accidents involving lift-served off-piste were excluded, and track points within 200 m of a piste were removed. We found no published test of SLABS outside Switzerland or for freeride terrain next to resorts.
Terrain comes from an elevation model. Glaciers can cause errors, and features smaller than the resolution are not there.
The published method and the operational variant differ in details (SLABS-1LR, the slope radius). BEYOND follows the paper.

SLABS: what the avalanche terrain analysis shows, and what it doesn't