PERMOS published the Swiss Permafrost Bulletin 2025 this year (Noetzli and Pellet, DOI 10.13093/permos-bull-2026), and buried in a network’s routine yearbook is a lesson in which instrument to trust. The hydrological year it covers — October 2024 to September 2025 — was the warmest in Switzerland above 1000 m asl since measurements began in 1864: +1.53 °C over the 1991–2020 mean, +2.73 °C over the 1961–1990 reference.

And yet almost every cheap surface indicator read calmer. The winter was very mild and snow-poor, and snow is a blanket. Without it the ground radiated to a clear sky, so mean annual ground surface temperature came in 0.1–2 °C below 2024 at most PERMOS sites — still above the 2012–2021 average. Rock glaciers, which creep faster when their ground is warm, slowed by 13% across the Swiss Alps, regionally from −5% in the Lower Valais to −22% in the Engadine.

←TODAY: the boreholes are logging as you read this, and the 20 m record was set by warmth that arrived years ago. →3012: the mountains a Zurich-3012 still stands on are the ones whose deep records were read honestly, not the ones whose survey marks looked quiet. Fulcrum: a single number only lies when you forget how long it took to reach the sensor.

Then the boreholes. Permafrost temperatures at 10 m and 20 m stayed at high levels, with new record highs at 70% of them: 17 of 24 boreholes at 10 m, 13 of 14 at 20 m. One year, two instruments, opposite answers — and only one is telling you about the trend.

Why the deep sensor wins

PERMOS states the physics plainly: changes in ground surface temperature penetrate downward with increasing delay and attenuation. In the Swiss Alps the depth of zero annual amplitude — where the seasonal swing has damped out entirely — sits at roughly 15–20 m. Below it, the ground answers to multi-annual change with delays of years to decades. So a 20 m borehole is at or below that depth: it physically cannot be reporting this winter.

The movement record says the same about a different quantity. Saibene, Gärtner-Roer, Beutel and Vieli put a shape-array chain down a 40 m borehole on the Murtèl rock glacier and read its deformation for nearly eight years (The Cryosphere, 2026). Surface movement averages 12 cm per year — but the shear zone at 26.5–28 m contributes 56% of it, a horizon a survey mark on top will never see. Only the active layer, about 3.5 m thick, carries any seasonal cycle. And the control is heat, not water: ground heat flux through the active layer was the most statistically relevant variable, while warm-phase precipitation showed no significant correlation with peak velocity.

Atelier: Anything built high and anchored into this ground — cable-car pylons, mountain huts, rockfall barriers, rock anchors, via ferrata fixings — was designed on the assumption that the ground holds still or creeps predictably. A surface reading is not a foundation reading: a survey mark measures the sum of everything beneath it, dominated by a shear horizon it cannot see. One good year is not a trend either — a 13% slowdown is a weather result, not a reprieve. Since 2022 rock glacier velocity has been an Essential Climate Variable under the Global Climate Observing System, so the number now has a keeper. Monday move: for any high-altitude asset you monitor, check whether its only instrument is a surface geodetic mark, and price a single deep borehole thermistor string against the cost of trusting the wrong reading.

Hack: Trace the annual temperature wave down through the ground and watch it damp and arrive late — the fact that lets 10 m and 20 m set records in a year the surface cooled. The damping depth d below is illustrative, not a PERMOS value; the shape is the point.

import math
d = 4.0                       # damping depth, m — ILLUSTRATIVE, not a PERMOS number
for z in (0, 3.5, 10, 20):    # surface · active-layer base · PERMOS 10 m · 20 m
    print(f"{z:>4} m  swing x{math.exp(-z/d):.3f}   arrives {z/d*12/(2*math.pi):+.1f} months late")

By 20 m the seasonal swing is a few per cent of the surface swing — which is exactly what “depth of zero annual amplitude at 15–20 m” means on the page.

PERMOS also reports eight rock slope failures from permafrost areas in 2025, with estimated volumes between 1,000 m³ and 6 million m³, all starting in north-facing slopes and all between May and October. The largest was the series of failures from the Kleines Nesthorn above Blatten, which deposited 6 million m³ on the Birchgletscher and provoked its failure; that event destroyed a Valais village and displaced its community. PERMOS records it as a rock slope failure originating in a permafrost area and publishes no causal attribution — and neither will this desk.

If you build in the Alps, ask what your monitoring actually reaches — and read the deep record before you trust the calm one.