An interactive globe that turns the planet inside-out. Instead of borders and terrain, it renders a third kind of map: what we actually know about the Earth's interior, layered on top of what theory predicts, slice by slice with depth — from the crust to the inner core.
| 660 km — cold slabs ring the Pacific | core–mantle boundary — the African LLSVP |
|---|---|
 |
 |
| the measured data as a filled 3-D volume | the theoretical model alone — a blurry estimate |
 |
 |
Two layers are composited in 3-D, sharing one camera:
| layer | what it is | how it looks |
|---|---|---|
| The scan | A classified slice of seismic tomography at the chosen depth — fast (cold, sinking) vs slow (hot, rising) shear-velocity anomalies. | Crisp. Colour encodes the anomaly (or the feature class); opacity encodes how well the region is resolved. Where coverage is thin, the scan fades and the model behind shows through. |
| The model | The smooth, radially-symmetric theoretical reference Earth (PREM). | Blurry — rendered to an offscreen buffer and Gaussian-blurred: an estimation, not an observation. Where the scan goes blind (the deep core), the estimate brightens to take over. |
| 3D bodies | Each feature (slab, plume, hot pile, cratonic keel, ULVZ) interpolated into a translucent, fuzzy-outlined body spanning its real depth range. | Slabs as dipping sheets, plumes as conduits, the two LLSVPs as basal piles. The body at your current depth brightens. Hover any body for what it is + its data sources; click to isolate it (the orbit pivot flies to its centre, the rest fade, a faint Earth stays for context) — Esc/back to return. |
| Relief surface | A translucent blue-marble Earth with hill-shaded topography (+ optional country borders). | Geographic orientation you can see through to the interior — toggle it or fade it with the opacity slider. |
| Theory in gaps | Where seismic coverage is too thin to resolve, the expected field is shown faintly under a diagonal hatch. | Makes "no data here — this is the model's guess" explicit, instead of just fading to nothing. |
Each depth also reports its temperature (K and °C, flagged as a modelled estimate with a ± and "not measured"), pressure, density, the resolved-coverage %, and a plain "what we know here" note.
Reading the colour: it is seismic-wave speed = the rock's temperature & stiffness, not motion. Blue = fast = cold/stiff (usually sinking); red = slow = hot/soft (usually rising); neutral = ordinary average mantle (measured, nothing unusual); hatched/faint = no readings, theory's guess only.
A Feature glossary button defines every type (craton, ridge, slab, plume, LLSVP, ULVZ); the Footprints toggle projects each body's outline radially onto the surface so you can see where it sits on the map; and hovering or extracting a body links straight to the public datasets & papers behind it.
A built-in "How we know" panel lists the public datasets & methods behind all of this (tomography models, normal modes, gravity, geoneutrinos, the core field, …) with links.
It is deliberately not binary "scanned / not-scanned." The classification is normalised to a diverging colour scale with transparency, so the map reads as a continuous field of confidence and feature type.
ES modules + fetch need to be served over HTTP (not opened as file://):
python3 -m http.server 8123
# then open http://127.0.0.1:8123/Everything is self-contained — three.js is vendored under vendor/, the
geographic data under data/. No network calls at runtime.
- drag orbit · scroll zoom
- depth slider / ticks dive to any depth or jump to a boundary (Moho, 410, 660, D″, CMB, ICB…)
- ▶ dive auto-descend to the core · ↑/↓ nudge depth · space dive
- Colour by ΔVs (velocity) or feature class
- toggles for the scan layer, the theoretical model, coastlines, feature tags, auto-rotate
- sliders for scan opacity, model haze (blur), and ΔVs gain
The radial profile — velocities, density, pressure, layer boundaries — is the real PREM reference model (Dziewonski & Anderson, 1981). The lateral anomalies are a hand-built, geographically-faithful synthesis of well-established features from published global tomography (the African & Pacific LLSVPs, subducted slabs, plume conduits, cratonic roots — cf. S40RTS, SEMUCB-WM1, GyPSuM), not a pixel-exact re-render of any single dataset. Treat it as an illustrative map of what we know is down there, not a measurement.
The site is plain static files at the repo root and uses only relative paths,
so it works under a project subpath (user.github.io/repo/):
- Push this repo to GitHub.
- Settings → Pages → Build and deployment → Deploy from a branch, pick
main// (root). - Open the published URL. (
.nojekyllis included so all files are served verbatim.)
Public: https://terrascan.bowd.io/ — served as a static site from GitHub Pages (no build step; three.js is vendored, everything loads over relative paths).
index.html markup + import map
css/style.css the HUD
js/earthModel.js PREM radial model, layers, depth↔radius helpers
js/tomography.js feature dataset → per-depth ΔVs + coverage DataTexture
js/geo.js coastlines, land-mask rasteriser, lat/lon → 3D
js/shells.js theoretical onion shells + the scan-slice shaders
js/postfx.js render-to-target, separable blur, starfield, composite
js/ui.js all DOM controls & readouts
js/main.js scene, OrbitControls, depth logic, the render loop
vendor/ three.js + OrbitControls (pinned r160)
data/ Natural Earth coastlines & land (preprocessed by build-geo.mjs)
Dev helpers (not needed at runtime): build-geo.mjs regenerates the compact geo
JSON; shoot.mjs renders verification screenshots with Playwright.
- Coastlines & land polygons: Natural Earth (public domain).
- Reference model: PREM / IRIS.
- Rendering: three.js.