README.md

---
license: mit
tags:
  - geospatial
  - taxi
  - new-york-city
  - osrm
size_categories:
  - 1M<n<10M
---

# Carbon Clock — Manhattan · Trip data

Trip data for [**The Carbon Clock — Manhattan**](https://github.com/NirmitSachde/the-carbon-clock-manhattan), a 24-hour animated visualization of NYC traffic, emissions, and air quality.

This dataset hosts the largest trip-tier binary that exceeds GitHub Pages' practical cross-origin limits. The smaller tiers (25 K / 100 K / 500 K) live in the repo itself.

## File: `trips-2m.bin`

- **2,933,898 trips** — every yellow-taxi ride that started and ended in Manhattan during **January 2025**
- Source: [NYC TLC Yellow Taxi Trip Records](https://www.nyc.gov/site/tlc/about/tlc-trip-record-data.page)
- **2,933,750 trips routed through OpenStreetMap roads via OSRM v5** (148 kept their straight-line endpoints — 99.995% success)
- Geometry simplified with Douglas–Peucker (30 m tolerance) — preserves every turn, drops collinear noise
- 20,489,760 total waypoints, average 7.0 per trip
- Format: custom binary `CCM1` — Uint32 offset table + RGBA color per trip + Float32 [lng, lat, seconds-since-midnight] per waypoint
  - ~40% the size of equivalent JSON
  - Parses ~10× faster (zero JSON.parse cost; the points decode directly to a Float32Array that's pushed straight to the GPU as a Deck.gl `TripsLayer` data buffer)
- Size: ~257 MB

## How it's used

The frontend at https://nirmitsachde.github.io/the-carbon-clock-manhattan/ fetches this file when the user selects the "Everything · 2.9 M" tier from the dropdown. Hugging Face's CORS headers make the cross-origin fetch work without any proxy.

## License

MIT for the binary format and processing pipeline. The underlying taxi data is from NYC TLC and remains public domain.