RoamingEye 🛰️🌍
Earth's time machine. Scrub decades of NASA satellite data on a 3D globe and watch the planet change month by month: forests green and brown, oceans warm and cool, snow come and go. Click anywhere for that spot's full time series, and zoom into any region at 30-metre detail. All of it open data, free for anyone, anywhere.
RoamingEye is an open-source research instrument for planetary-scale observation. It turns the public satellite archives that humanity has already paid for — NASA's MODIS and Harmonized Landsat-Sentinel collections — into a fast, intuitive, browser-based globe that any researcher, educator, journalist, or curious person can use without an account, a license, or a fee.
Our thesis: the ability to watch the Earth change over time should not be locked behind expensive commercial platforms. The data is open. The tooling should be too.
🔭 Why this matters
Decades of Earth observation sit in open archives, but most of it is reached through specialist GIS software, API keys, and data-wrangling pipelines that exclude all but a handful of trained users. RoamingEye is an attempt to collapse that barrier: point, drag, and scrub — and the planet's recorded history is in front of you.
Statement of need. Researchers, educators, and students need a way to look at multi-decadal satellite records — to form hypotheses, check a site before pulling L3 granules, or teach a seasonal cycle — without a login, an SDK, or a GIS seat. Worldview is NASA-hosted and 2D; Google Earth Engine requires accounts and code. RoamingEye fills the gap in between: a zero-install, fully open-source, provenance-first 3D globe that runs in the browser, cites every dataset it renders, and exports reproducible, uncertainty-labelled time series.
It is built for, and by, the research community: every data source is open and cited, every imagery selection is provenance-tagged with the instrument and acquisition date, and the whole stack is MIT-licensed so any lab, classroom, or newsroom can fork and extend it.
✨ What it does
- 🌍 A real 3D Earth rendered with WebGL — grab to rotate, scroll to zoom from orbit down to the surface.
- 🔬 Native-resolution tile streaming, on by default — zoom in and the visible globe re-drapes itself with WMTS tiles chosen by screen-space error, up to each layer's native resolution (terrain reaches ~31 m), with parent-tile fallback so detail refines instead of popping.
- ⏳ A temporal scrubber — a ruler-style timeline that sweeps month-by-month across each layer's full published record, so you can watch the seasons turn and trends emerge: 2000 → present for the MODIS layers, and back to 1980 for the MERRA-2 reanalysis layers (2 m air temperature, aerosols).
- 🌱❄️🌡️ A rich set of scientific layers — 9 open NASA products across vegetation (NDVI, EVI), temperature (land surface, 2 m air, sea surface), water (precipitation, soil moisture), cryosphere (snow cover), and atmosphere (aerosols) — grouped in a clean picker and growing.
- 📚 An open-data Providers page — a built-in catalogue of the 37 agencies, archives, and platforms whose open data powers the project.
- 🧰 A reviewed open-software finder — browse Earth-science tools by domain, platform, and access path; every public recommendation links to its repository, documentation, SPDX evidence, and verification date.
- 🔎 Search any place — geocoded via OpenStreetMap; the globe traces the returned postcode, city, state, or country boundary and surfaces its latest month-over-month vegetation, rainfall, soil-moisture, and air-temperature signals.
- 📈 A point time-series probe — click anywhere on the globe and chart that layer's value at that point across its full record (26–46 years), with a provenance-stamped CSV download. Approximate by design (colormap inversion), honest about it everywhere.
- 🌋 A plate-tectonics context pack — Bird (2003) plate boundaries, ~1,200 Smithsonian GVP Holocene volcanoes colored by eruption recency, and live USGS seismicity (M4.5+, colored by depth) on one globe.
- 🧭 A live coordinate readout — hover anywhere to read latitude/longitude and the country/territory under the cursor.
- 🗺️ Toggleable overlays — coordinate grid, national borders, cities, and an atmosphere glow.
🧪 Built for research
RoamingEye is designed around real scientific workflows — see
docs/research-recipes.md for five step-by-step
walkthroughs (drought signals, LST trends, the plate-tectonics lecture view,
snowpack tracking, deforestation figures). For how the tool computes what it
shows and where it stops being trustworthy — the probe pipeline, area
weighting, uncertainty, the seasonal Mann-Kendall / Sen's slope trend test, and
the measured per-layer inversion accuracy — read METHODS.md.
Teaching with it?
docs/teaching/ndvi-phenology-lab.md
is a ready-to-run 60–75 minute classroom lab — "when does the Earth turn
green?" — with learning objectives, a student worksheet, an assessment rubric,
instructor answer notes, and the honest limits written up as teachable
material. No accounts, no install, works on a Chromebook. Free to adapt (MIT).
A few examples it already supports:
| Field | What you can observe |
|---|
| Vegetation phenology | Green-up and senescence cycles via monthly NDVI/EVI across years. |
| Deforestation & land-use change | Step a 30 m study patch over a forest frontier across consecutive years. |
| Urban expansion | Watch a city's footprint grow in high-resolution true colour. |
| Drought & agriculture | Compare vegetation vigour between wet and dry years over a region. |
| Snow & cryosphere | Track seasonal snow advance/retreat with monthly snow-cover composites. |
| Disaster & event assessment | Pull the clearest pre/post imagery for a study area after a fire or flood. |
| Point time series | Click a site → 26-year chart + CSV for drought, greening, or LST trends. |
| Plate tectonics & geohazards | Plate boundaries + Holocene volcanoes + live seismicity on the terrain. |
Every high-resolution scene is labelled with its instrument and acquisition date (e.g. Sentinel-2 · HLS S30 · 30 m · 2024-08-05) so observations are reproducible and citable.
🛰️ Data & provenance
100% open, public-domain or open-licensed data:
| Source | Product | Native resolution | Coverage | License |
|---|
| NASA GIBS (MODIS/Terra) | Monthly vegetation (NDVI, EVI) | 1 km | 2000 → present | Public domain |
| NASA GIBS (MODIS/Terra) | Monthly snow cover | 2 km | 2000 → present | Public domain |
| NASA GIBS (HLS) | High-res true colour (Sentinel-2 S30 / Landsat L30) | ~30 m | 2013/2015 → present | Public domain |
| Natural Earth | National borders, populated places | 1:110m | — | Public domain |
| OpenStreetMap (Nominatim) | Geocoding & administrative boundaries | — | — | ODbL |
See DATA_SOURCES.md for the full catalogue, layer identifiers, and scientific notes.
An honest note on resolution. Open data tops out at roughly 10 m (Sentinel-2) to 30 m (Landsat/HLS) for recent, frequently-revisited imagery. True sub-metre "street-level" imagery only exists in commercial archives, which are neither free nor global. RoamingEye deliberately stays within the open ecosystem — and its tiled-streaming engine renders each layer's full native resolution wherever you zoom, by default.
🚀 Quickstart
Requirements: Node.js 20.19+ (or 22.12+) and npm.
git clone https://github.com/zkWizard/RoamingEye.git
cd RoamingEye
npm install
npm run dev # → http://localhost:5173
The dev server is also exposed on your local network, so you can open the printed http://<your-ip>:5173 address on a phone or tablet.
npm run build # type-check + production build
npm run test # unit tests (Vitest)
npm run test:e2e:smoke # required critical-path smoke (Playwright)
npm run test:e2e # full browser suite (Playwright)
🏗️ How it works
A lightweight, dependency-honest stack — no heavy GIS frameworks:
- Three.js for WebGL rendering, with 3DTilesRendererJS streaming a satellite-imagery and terrain globe (Esri World Imagery over AWS Terrarium elevation) beneath the NASA layers, and draping those layers on the terrain up close.
- NASA GIBS WMS for imagery, streamed straight into GPU textures (the service is CORS-open, so there's no backend).
- Pure, unit-tested logic for the geodesy, timeline, scene selection, and geocoding; rendering and DOM kept separate from it.
- TypeScript, Vite, Vitest, Playwright, ESLint/Prettier, and GitHub Actions CI.
For a contributor's tour of the codebase, see ARCHITECTURE.md.
🤝 Contributing
We are actively looking for collaborators — Earth scientists, remote-sensing specialists, graphics engineers, designers, and data wranglers all have a place here.
- 💬 Just want to ask something? Start a discussion — Q&A for "how do I…?" and "is this layer trustworthy for…?", Ideas for anything you wish it did. No bug report required, and you don't need to have contributed anything first.
- 🗣️ Used RoamingEye in your research or teaching? Give feedback — three questions, two minutes, rough notes welcome. Open to a 15-minute chat about your workflow? Say so in the form; what researchers tell us directly shapes the roadmap.
- 📘 Read
.github/CONTRIBUTING.md — setup, workflow, and the (one-line, no-CLA) DCO sign-off.
- 🏛️
GOVERNANCE.md explains the trust ladder and how decisions get made.
- 🌱 New to the project? Browse good first issues.
- ➕ Want to add a dataset?
docs/adding-a-data-layer.md walks through it.
- 🧰 Want to improve the software finder?
docs/agent-fleet.md explains the six review-gated catalog agents, Fleet status view, and editorial approval flow.
- 🚩 The flagship engine: shipped — and its follow-ons (tile-edge skirts, polar handling, Sentinel-2 at 10 m) are great graphics projects. See the .
Every change lands through a reviewed, CI-gated pull request. Be kind — see the Code of Conduct.
👀 Why are there so many open pull requests?
Because most of them are opened automatically — and none of them are in your way.
Alongside the catalog agents above, RoamingEye runs a fleet of scheduled Earth-science specialists — Geologist, Biologist, Meteorologist, Marine Biologist, Environmental Scientist, and Geospatial & Remote-Sensing Engineer. Each one opens small, self-contained pull requests from a codex/<lane>-<task> branch, and Dependabot opens the rest. docs/fleet-expansion-program.md documents that program — what each lane owns, what a specialist must produce to complete a cycle, and how its work is validated and integrated.
If you're here to contribute, the queue length is not something you need to read or work around:
🗺️ Roadmap
🎯 Goals — 2026
Last updated: 2026-07-11. Tracked daily as work lands; full detail and
the feature-level Now / Next / Later live in ROADMAP.md.
Now / Next / Later at a glance
- Now — 🚩 absolute probe accuracy (#170), release & DOI residuals, research/NASA outreach.
- Next — more scientific layers (fire/thermal, surface water), tile-edge polish for HD streaming, Sentinel-2 at 10 m.
- Later — true 3D elevation terrain (GEBCO/SRTM), change-point detection & region comparison, annotation & collaboration, offline field mode.
🎓 Citing RoamingEye and its data
Work made with RoamingEye involves three citable objects — the tool, the
imagery service, and the datasets. Doing all three right takes two minutes:
1. The tool. The repository ships a CITATION.cff that
GitHub renders as a ready-made citation ("Cite this repository" in the
sidebar, APA and BibTeX). When citing a specific observation, include the
view URL from the address bar — it encodes the layer, month, and camera
position, so readers reproduce exactly what you saw — and the acquisition
details shown in the app (e.g. the study-region chip's
Sentinel-2 · HLS S30 · 30 m · 2026-05-11). CSV exports already carry all of
this in their # provenance headers, including # view_url and
# tool_version.
2. The imagery service. NASA asks users of GIBS to include this
acknowledgment verbatim:
We acknowledge the use of imagery provided by services from NASA's Global
Imagery Browse Services (GIBS), part of NASA's Earth Science Data and
Information System (ESDIS).
3. The data. GIBS renders datasets that carry their own DOIs and
citations — NASA's data-use guidance
asks publications to cite them directly (each DOI resolves to a landing page
with the full citation; CSV exports name theirs in # data_product /
# data_doi):
This table is generated from the same layer configuration the app runs on and
is drift-guarded by a unit test — if a layer's source product changes, CI
fails until the docs follow. The in-app Data providers page (footer link)
shows the same list.
A Zenodo DOI (for stable, versioned citations) is planned — maintainers: mint
one by connecting the repo at zenodo.org
and cutting a release; the DOI badge then goes here.
📄 License & attribution
- Code: MIT.
- Imagery & data: retain their own licenses (see the table above). Imagery courtesy of NASA EOSDIS GIBS; vector data © Natural Earth; geocoding © OpenStreetMap contributors.
🙏 Acknowledgements
RoamingEye stands entirely on the shoulders of the open Earth-observation community — the scientists and engineers at NASA, ESA, USGS, the OpenStreetMap project, and Natural Earth, who put planetary-scale data in the public domain. This project is simply a friendlier window onto their work.
