flightyco2

A simple flight-emissions calculator for users of Flighty.

Upload your Flighty CSV export, get a per-year and per-flight breakdown of your carbon footprint computed with Google's Travel Impact Model and DEFRA 2024 factors. All processing happens client-side, the CSV never leaves your browser.

Try it out here!

Methodology

• TIM 3.0.0: faithful port of Google's Travel Impact Model, validated against its published worked example to within 0.07%. Reads per-aircraft EEA fuel-burn curves, route-specific distance adjustments (GAIA), CORSIA well-to-wake conversion, seat-area class allocation.
• DEFRA 2024: UK Gov GHG conversion factors, verified against the published spreadsheet. Used as a transparent fallback when TIM lacks data for a flight (no aircraft string, unmapped ICAO); the per-flight result records which method ran.
• Non-CO₂ effects: 1.9× radiative-forcing uplift on by default, toggleable in the UI.

Full methodology and source list is in the in-app methodology drawer.

Stack

• Astro 6 static + Svelte 5 islands + TypeScript 6 + Tailwind v4
• Biome + Vitest + PapaParse + Zod
• Deployed on Cloudflare Pages

Local development

Requires Node 20+ and npm.

git clone https://github.com/dupontcyborg/flightyco2.git
cd flightyco2
npm install

npm run dev      # Astro dev server
npm test         # Vitest
npm run build    # Production build
npm run lint     # Biome

Tests run on a fresh clone with no private data. The end-to-end regression uses sample_data/canonical-fixture.csv.

Reference data (airports, aircraft, fuel-burn, etc.) is regenerated from upstream sources via npm run data (and per-source variants npm run data:airports, data:gaia, data:eea, data:tim, data:defra, data:seats). See data/README.md for upstream attribution and licenses.

Airline brand logos

public/airlines/*.svg contains airline brand marks used solely to identify the carrier of each flight in the user's own uploaded CSV — i.e. nominative use, no endorsement implied, no resale. The trademarks remain the property of their respective owners; this project claims no rights to them.

AI disclosure

This project was built with substantial use of large language models. Specifically:

• Architecture and design: human (me, @dupontcyborg, a senior software engineer).
• Implementation: predominantly LLM-assisted. Most TypeScript, Svelte components, and tests were drafted by my dear friend Claude :)
• Review: me again.

Bugs and typos are most likely my own, in all fairness to Claude.

License

MIT. Contributions welcome.