Skye Orbital 3D globe with Starlink satellites

SKYE ORBITAL

Skye Orbital is a real-time dashboard that visualizes over 13,000 active satellites and space debris on an interactive 3D globe, with visible-pass prediction and orbital close-approach detection. Client-side SGP4 propagation via satellite.js, WebGL rendering with CesiumJS, Cloudflare back end (Workers, D1, KV, Queues, Workers AI).

JavaScript CesiumJS WebGL Cloudflare Workers Cloudflare Queues Workers AI

PROJECT DETAILS

Objective

The goal was to make readable and accessible to a non-specialist audience something that usually stays an abstract table of coordinates: where a satellite is right now, when it will be visible from a given point, and which objects are dangerously converging with one another in low Earth orbit.

Technologies used

The front end is static (vanilla HTML/CSS/JS), with no framework or bundler: CesiumJS renders the 3D globe in WebGL (PointPrimitiveCollection and LabelCollection to keep it smooth with 13,000+ objects), and satellite.js computes SGP4 orbital propagation directly in the browser. The Cloudflare back end serves orbital elements (TLE) through Workers with a KV cache (2h TTL, respecting CelesTrak's rate limit) and Cron Triggers for periodic refresh; D1 stores the object catalog, detected close approaches and daily briefings. Close-approach detection runs on a 4-stage Cloudflare Queues pipeline (parse → bin-and-dispatch → precompute → scan → refine), and the daily natural-language briefing is generated via Workers AI.

Challenges and solutions

The main technical challenge was rendering over 13,000 tracked objects in real time without hurting performance: the fix was to delegate all position computation to the browser (SGP4 via satellite.js) and let the server only cache and serve the raw orbital elements. Close-approach detection needed a dedicated multi-stage pipeline to stay practical at that scale over a 72h window, with a 5km production threshold — deliberately presented as an indicative geometric signal, not an operational collision prediction.