About

The atmosphere is the oldest artist on Earth. It has been painting since before there were eyes to see it — sculpting light through ice crystals, dragging color across horizons, compressing entire landscapes into a single hour of sky. This project is an attempt to listen to what it's making.

The Work

Every day, fifty-four points across the planet are sampled for what the atmosphere is actually doing — not what it looks like on a forecast, but its raw physical state. Pressure gradients steep enough to bend trees. Winds crossing open water with nothing to slow them down. Temperature anomalies where the air is ten degrees warmer or colder than it should be. The ten most extreme conditions become source material for original artwork.

The data passes through the aesthetic lens of eleven artists who understood that abstraction is not the absence of meaning but its distillation. Sam Francis knew that empty space is as charged as color. Rothko proved that two rectangles can hold grief. Frankenthaler showed that letting paint find its own path is an act of trust. Hilma af Klint painted what she believed was invisible. Lesley Tannahill builds palimpsests where every mark remembers the marks beneath it. These sensibilities are not styles to imitate — they are ways of seeing that transform the same pressure reading into radically different compositions.

The result is a growing archive of the planet's most volatile moments, translated into color and form. No two pieces share the same atmospheric conditions. None can be repeated. Each one is a fossil of weather that no longer exists.

The System

The pipeline runs at dawn UTC. Weather flows in from the Open-Meteo API (NOAA's GFS model) and is scored on a composite index — pressure anomaly weighted heaviest, then wind, temperature deviation, precipitation, humidity. The highest-scoring conditions are sent to an AI model on Amazon Bedrock, which interprets the atmospheric data as original digital artwork — either high-resolution PNG via Flux 1.1 Pro or vector SVG via Claude. Every piece is rendered in seven possible canvas formats, from square to panoramic, chosen by the conditions themselves. Everything is archived permanently.

A parallel system extracts color palettes from Copernicus Sentinel-2 satellite imagery — the actual colors of Earth as seen from orbit. Deserts, glaciers, river deltas, volcanic islands. A seasonal record of what the planet looks like from 786 kilometers up.

How the Art Is Made

Artistic Interpretation, Not Imitation

No artist's work was used to train the model. The AI's understanding of these artists comes from art criticism, exhibition catalogs, and the public discourse around their practices — the same way a human art student learns by reading about artists before they ever pick up a brush. Each artist is described to the model in a few sentences that capture their visual philosophy: Rothko's luminous floating rectangles, Kusama's obsessive dot fields, Mondrian's primary-color grid logic. The model doesn't see their paintings. It reads their ideas, then writes SVG code — shapes, gradients, transforms, filters — from scratch. The atmospheric data is the only input. The artist's sensibility is the lens.

Weather Data: 54 Points, 6 Variables

Every run samples 54 points across 10 latitude bands, from 70°N (Arctic) to 60°S (Subantarctic). At each point, six atmospheric variables are measured at the current hour:

The composite visual interest score weights these: pressure anomaly (30%), wind (25%), temperature deviation (20%), precipitation (15%), humidity (10%). The top 10 scorers — with at least 15° of geographic separation to ensure diversity — become the day's artwork subjects.

Why GFS?

The weather data comes from NOAA's Global Forecast System (GFS), a 0.25° resolution atmospheric model updated every six hours. GFS was chosen over alternatives like ECMWF IFS (higher accuracy but rate-limited), ICON (better European resolution), or HRRR (3km US-only) because it offers the best tradeoff for this project: free global coverage at resolution sufficient for macro-scale weather drama, with reliable hourly data across all 54 scan points simultaneously. The art interprets continental-scale atmospheric patterns — pressure systems spanning oceans, temperature fronts crossing mountain ranges — where GFS excels. The gfs_seamless variant blends forecast and analysis data for best-available current conditions.

Quality Scoring

After each artwork is generated and rendered to PNG, an AI art critic evaluates it on four dimensions: composition, color harmony, complexity, and emotional impact. Each piece receives a quality score from 1 to 10. This score feeds into the archive, influences print pricing, and helps surface the strongest work. The critic sees only the final image — it has no access to the weather data or artist prompt that produced it.

Weather Forecasting

Every evening at 20:00 UTC, a separate system scans the 24-hour GFS forecast for all 54 scan points, scoring each for predicted visual drama using the same composite formula. This produces a ranked list of tomorrow's most promising atmospheric conditions — a weather drama forecast that anticipates where the most striking art will come from before the pipeline runs.

Dynamic Pricing

Print prices are not static. Each artwork's price reflects three signals: the quality score from the AI critic (higher-rated work commands a premium), the rarity of the weather conditions that produced it (extreme atmospheric events are scarcer), and scarcity (as editions sell, remaining prints increase in value). The base price can increase up to 2× for exceptional pieces born from rare weather.

Natural Scarcity

Every artwork is generated from a one-time atmospheric event — a specific combination of pressure, wind, temperature, and precipitation at a specific location and moment. That weather will never occur again. Unlike digital assets where scarcity is manufactured by capping supply, the scarcity here is physical: the atmosphere produced it once, and it's gone. The artwork is a fossil of a moment that no longer exists.

This is closer to photography than to minting — capturing something real, unrepeatable, and time-bound. The atmospheric data is archived alongside every piece, timestamped and verifiable against NOAA's public records. The weather happened. The art it produced is the only one of its kind. Limited edition prints (5 per size) add a physical layer of scarcity on top of a natural one. When the last print of a piece sells, that's it — the weather is gone, the edition is closed, and the work exists only in the hands of its collectors.

The Artist

I'm James Tannahill. I grew up staring at maps the way other people stared at posters — tracing coastlines, memorizing the shapes of countries I'd never visit, trying to understand how a flat sheet of paper could contain an ocean. That impulse never went away. It just found new material.

Weather is the map I can't finish reading. It is planetary-scale, constantly rewriting itself, and fundamentally invisible at any useful resolution. You can stand in a storm and still only experience a pinhole of what the system is doing. This project is my way of seeing more of it — using the language of the painters who taught me that seeing is not passive, that abstraction is a form of attention, and that the most honest response to something overwhelming is to let it move through you and onto the canvas.

The data is real. The locations are real. The weather happened. What the machine makes of it — and what you make of that — is where the art lives.

FAQ