The order and pace never change: one note a week, a full turn every eight weeks. As the total ice shrinks, the highest and lowest notes (top) fall silent first and the middle notes (bottom) hold longest. Dashed rings are silent notes. Pale halos are the upper octave that open water adds.
Cello: thick ice is low and warm, thin ice is high and bare. The bass sounds while the year's mean ice volume (a model) is above 14 thousand km³. Open water is heard in the organ, not the cello.
One bar is one ice year. Each bar is 16 seconds, October to September, 48 bars from 1979 to 2026, and every week of ice is one pulse.
The wheel. Eight organ notes turn in the same order at the same pace, one note a week: C, G, C, E♭, F, G, B♭, C. They come from a minor pentatonic scale, which has no semitones and no tritone (the harshest intervals), so whichever notes are left, none of them clash in that way.
The ice decides how many can sound. In a full March all eight; in the 1980s Septembers, six; in the 2012 record low, two. A note that drops out is a rest, not a quieter note, so the loop keeps its length and pace, and the notes that remain keep the pitch and touch they had in 1979. The highest and lowest go first and the middle ones hold longest, so the loop gets a gap that grows and the range narrows toward the middle. The loss follows the ice as a whole, not particular seas.
The cello is thickness. It plays one note every two months, and its register follows ice volume (from a model): thick ice is low and warm, thin ice is high and bare, so over the record it climbs. Each note swells in with a bow instead of a strike.
The bass is the floor. A low C sits under everything while the ice is thick, and it is gone by the 2020s. The cello and bass also rise and fall with each year's extent.
Open water is light. When the sea opens, the organ gains its upper octave and the room grows wider and brighter. Thinning and opening are told apart: the cello says how thick the ice is; the organ and the room say how much water is open.
Time bends a little. Weeks press slightly in freeze-up and drag in melt, but every bar is exactly 16 seconds. After the last bar come six October days of refreeze: the wheel restarts thin, the cello holds an open fifth, and it is left unresolved. The method is that of process music, except that the process here is subtraction.
The wheel and the voices panel are drawn from the rules that made the sound. The number of organ notes follows total ice extent, a satellite record (R). The cello's register and the bass follow ice volume, which is a model (M). Warming is not in this piece; the Round, the main piece of the pair, will carry it. Land snow, green and melt are a model (M) run at 29 points, about 650 km apart: they show the timing of the seasons, not local detail, and are not a satellite product. Days with no satellite grid repeat the nearest one. The disc at the pole is not measured. The satellites' orbits do not pass over the pole, so each sensor leaves a round area unseen: about 600 km in radius until August 1987, about 305 km until the end of 2007, about 95 km until the end of 2024 and about 140 km since (measured from the grids used here). The map draws it as full ice inside a dashed circle, and NSIDC's extent figures count it as ice too. The disc shrinks in 1987 and 2008 because the sensors changed, not the ice. The extent record starts on 26 October 1978, so the first weeks of the 1979 bar repeat that day's value. Before mid-1987 extent was recorded every other day, and there are no data from 2 December 1987 to 13 January 1988; those days are interpolated.
Acknowledgement. This piece is made from public satellite and climate records of the Arctic Ocean and its sea ice. Sea ice is part of the homeland of Inuit and of other Indigenous peoples of the Arctic. The Inuit Circumpolar Council's 2026 Iqaluit Declaration reaffirms "our right to our ancestral homelands across Inuit Nunaat", and its 2018 Utqiaġvik Declaration affirmed Inuit rights across "lands, waters, ice, air space and resources". This piece uses no Inuit knowledge and no Inuit data, and the Inuit Circumpolar Council and SIKU have not reviewed it. Inuit-led work on sea ice, such as SIKU, is the place to learn from people who live with and travel on the ice.
Sea ice (R): NSIDC Sea Ice Index, Version 4 (G02135), Fetterer, Knowles, Meier, Savoie, Windnagel and Stafford, doi:10.7265/a98x-0f50; weekly concentration grids at half resolution and in 5 % steps, each year's maximum and minimum grids at the full 25 km in 5 % steps, and the daily total extent. Derived by us; NSIDC has not reviewed this. Ice volume (M): PIOMAS v2.1, Polar Science Center, University of Washington (Zhang and Rothrock 2003; Schweiger et al. 2011, doi:10.1029/2011JC007084). The picture's files carry only each year's mean volume and the cello's register steps; the series is not republished and is available from the Polar Science Center. PIOMAS for March to September 2026 repeats 2025, so the 2026 cello register and bass follow that extension. Terrain (R): GEBCO Bathymetric Compilation Group 2025, The GEBCO_2025 Grid, doi:10.5285/37c52e96-24ea-67ce-e063-7086abc05f29; not for navigation, and GEBCO does not endorse this page. Land snow, green and melt (M): weather data by Open-Meteo.com (CC BY 4.0, non-commercial use), generated using Copernicus Climate Change Service information 2026 (ERA5; Hersbach et al. 2020). Contains modified Copernicus Climate Change Service information 2026. Neither the European Commission nor ECMWF is responsible for any use that may be made of the Copernicus information or data it contains. Glaciated areas and coastlines: Natural Earth (public domain). Sound: pipe organ and cello from VSCO 2 CE, Versilian Studios (CC0). Drawing library: d3 7.9.0, Mike Bostock (ISC). Licence. The audio and the picture's derived files are CC BY-SA 4.0; the code is MPL-2.0. The land layer file (arctic_land.json) comes from Open-Meteo's free API, which is for non-commercial use: to reuse it commercially, regenerate it under a commercial plan. Each dataset above keeps its provider's terms. Made by Claude Code (Sonnet 5.5) and reviewed by GW.
What is measured and what is modelled. Sea ice: NSIDC Sea Ice Index v4 concentration grids, one every 7 days (R); the pole hole is drawn as ice and is not measured; days with no grid repeat the nearest one. Land height and sea depth: GEBCO_2025 Grid (R, a compiled terrain model; not for navigation; GEBCO does not endorse this page). Land snow, green and meltwater: a model (M) run on ERA5 air temperature and snowfall at 29 points, about 650 km apart, then blended: it shows the timing of the seasons, not local detail, and it is not a satellite vegetation product. Greening is kept faint on purpose and is left off permanent ice (Natural Earth glaciated areas, public domain; small glaciers are missing from it). Rings and strip: NSIDC extent, NOAA Barrow CO2, NOAA ocean temperature, NASA GISTEMP (64 to 90 N); 2026 is partial. Sound: VSCO 2 CE samples and synthesis. Claude Code (Sonnet 5.5) built this; GW has not reviewed it.