Tectonic Chorus: A Real-Time Sonification of the Antarctic Ice Sheet

Tectonic Chorus is a planetary-scale instrument: 24 geophone stations and 12 weather-proof microphones are deployed in a grid across the Ross Ice Shelf, each powered by solar panels and small wind turbines. Data is pre-processed on Raspberry Pi 4s, compressed, and sent via Iridium satellite bursts to a base station at McMurdo Station, then to an AWS cloud instance. There, a custom machine learning model—a LSTM trained on decades of Antarctic seismic data—analyzes patterns of ice movement, crevasse formation, and meltwater flow. The model outputs control signals for a generative music engine built in SuperCollider, which maps seismic frequencies to granular textures and rhythmic pulses, creating a never-repeating symphony. The public can influence the composition by submitting "wishes" via a web interface, which subtly bias the ML model's mood parameters. The live stream is accessible as a 24/7 ambient broadcast, with curated highlights released monthly. The project combines art, science, and climate advocacy, making the invisible dynamics of the Antarctic ice sheet audible and emotionally resonant.

Extreme environments like the Antarctic ice sheet are remote, inaccessible, and silent to most people. Existing scientific sonifications are often utilitarian and not designed for aesthetic engagement. At the same time, climate change is rapidly altering these environments, but the public remains disconnected. Tectonic Chorus bridges this gap by turning real-time geophysical data into a living musical artwork that invites listeners to hear the ice's response to a warming world.

• Deploy 24 geophone stations and 12 microphones on the Ross Ice Shelf.
• Establish real-time, low-bandwidth data transmission via Iridium satellites.
• Develop a machine learning model that translates seismic and acoustic features into musical parameters.
• Build a generative music engine that produces a continuous, evolving composition without loops.
• Design a public web platform for live streaming and interactive influence.
• Collaborate with glaciologists to ensure scientific accuracy and data sharing.

• Not a scientific visualization tool for researchers.
• Not a static installation; it is a live, evolving broadcast.
• Not controlled by human performers in real time.
• No physical instrument construction beyond the sensor nodes.
• Not intended to replace field research or monitoring networks.

Hardware: Geophones (4.5 Hz, vertical component), weather-proof omnidirectional microphones (e.g., DPA 4098), Raspberry Pi 4 with custom hat for ADC, Iridium 9602 satellite modems, solar panels (40W each), small wind turbines (20W), sealed Pelican cases with heating elements.

Software: Python for sensor data acquisition and compression, PyTorch for ML model, SuperCollider for audio synthesis, Icecast for streaming, React for web interface, Node.js for backend API. AWS EC2 for cloud processing, S3 for data storage, Lambda for triggers.

ML Model: LSTM with attention, trained on 10 years of Antarctic seismic data from IRIS. Outputs control signals: tempo, density, pitch-shift, and filter cutoff.

The system is layered into four tiers:

1. Sensor Tier: Each node captures seismic (24-bit, 100 Hz) and acoustic (44.1 kHz) data. The Pi processes and compresses into 5-minute bundles (approx. 2 MB), then transmits via Iridium to the McMurdo gateway.
2. Gateway Tier: A server at McMurdo receives satellite bursts, reassembles packets, and forwards to AWS via internet.
3. Cloud Tier: AWS S3 buffers data; a Lambda triggers the ML inference server. The LSTM processes a sliding window of the last 30 minutes, outputting a control vector every 10 seconds.
4. Audio Tier: SuperCollider runs on a GPU-equipped EC2 instance. It renders 8 channels of audio: granular textures for seismic events, ambient pads for background hum, and processed microphone audio for meltwater streams. The mix is streamed via Icecast.

Public interface: Web app fetches live stream URL and allows users to select a ‘mood’ (calm, turbulent, experimental) which is fed back as a bias to the ML model.

• Extreme cold/failure: Enclosures may crack; batteries may freeze. Redundant heating elements and low-power sleep modes mitigate but not eliminate.
• Data latency: Iridium bursts have 2-5 minute delays. The ML must interpolate for real-time feel.
• Wind noise: Microphones may saturate during blizzards. Adaptive gain control and filtering in pre-processing.
• Funding gaps: Can't afford full deployment without grants or sponsorship.
• Permit denial: Antarctic activity requires permits; delays or denials could halt the project.

• How to differentiate between ice creaks and wind vibrations in the geophone signal?
• Should we include data from adjacent meltwater streams (summer only)?
• What is the best way to balance public influence (wishes) with scientific fidelity?
• Is 24 nodes enough spatial coverage for interesting musical variation?
• How to handle sensor drift over months of operation?

The project was fully planned with schematics, code prototypes, and preliminary permits from Antarctica New Zealand. However, the lead sound artist died in a climbing accident, and momentum stalled. The cost of shipping and logistics (~$500,000) was never secured. The plan remains as a detailed document and a GitHub repository of broken code.

This project also serves as a climate art piece: over years, the composition will change as the ice sheet melts and fractures. If built, it could be expanded to Greenland or deep-sea vents. The interactive features encourage community engagement with distant ecosystems.