Radial values: real PREM. Lateral features: a faithful synthesis of published tomography.
The scan — what you're seeingⓘ
14 real tomography models, combined live.
Colour = how fast seismic waves travel = the rock's temperature & stiffness. Motion (sinking / rising) is inferred from that — it is not a direct movement reading.
slow / hotaveragefast / cold
Blue = fast → cold, stiff rock, usually sinking (slabs)
Red = slow → hot, soft rock, usually rising (plumes, piles)
Hatched / faint = no readings — theory's guess only, not measured
subducted slab (cold)
LLSVP — giant hot pile
ULVZ — melt at the core
mantle plume (hot)
cratonic root (cold)
ridge / rift (hot)
Same data, coloured by what the feature is rather than its speed.
At this depthⓘ
The scan picks out
—
Conditions
—temperatureⓘ
—pressure
—density
—shear speed baseline
how well we've scanned here—
solid scan coverage
What we know here
—
Data pipelineⓘ
Model inputs
clustering
threshold0.50
smooth0.30
agreement min0.40
visualization
A THIRD MAP OF THE EARTH
You're looking inside the planet.
Ordinary globes show the outside — coastlines, borders, terrain. This one looks inward. As you slide the depth control down, you cut a horizontal slice through the Earth and see two things stacked on top of each other:
① THE SCAN — sharp
What earthquakes reveal: how fast seismic waves travel at this depth — i.e. how hot/cold & stiff the rock is. (Colour shows temperature, not motion; we infer sinking/rising from it.)
Red = slower → hotter, softer, rising Blue = faster → colder, stiffer, sinking Faint = we've barely scanned here
② THE MODEL — blurry
Behind the scan glows the textbook Earth: the smooth, layered reference model (PREM). It's deliberately out of focus — a prediction, not an observation.
Where the scan has no coverage (especially the deep core), the blurry model brightens and takes over. That hand-off is the whole idea.
③ THE 3-D BODIES — how they're derived
Data bodies come straight from the measurements: the 14 tomography models are averaged on a (lon, lat, depth) grid, lightly smoothed, kept only where the models agree, then meshed into 3-D form (surfaces / volume / points / wire) at a chosen anomaly threshold — blue = fast/cold, red = slow/hot. Tune all of this in the Data-pipeline panel.
Extracted features are an interpretation, not derived from those models: hand-curated named structures (slabs, LLSVPs, plumes, cratons, ridges) placed at their published locations from the literature. So a plume won't sit exactly on the data's red blob — they're two different things. The survey-link dial fades each feature toward how strongly the measured ensemble actually supports it there.
Drag to orbit · scroll to zoom · click a boundary to jump there · hover a body for info, click to isolate it. Coastlines: Natural Earth. Reference model: PREM (Dziewonski & Anderson 1981). Built with three.js.
HOW WE KNOW
Everything public that maps the Earth's inside
We have never drilled past ~12 km. Everything deeper is inferred — mostly from earthquakes, plus gravity, magnetism, tiny particles, and lab physics. These are the public datasets & models behind that inference.
gathering sources…
FIELD GUIDE
What are these features?
A quick taxonomy of what lives inside the Earth. Blue = fast = cold/stiff; red = slow = hot/soft. "Motion" is inferred from temperature, not measured.
THE SHALLOW SKIN — KARST & CAVES
Mapped caves vs the unmapped frontier
The deep layers above are seismic inference. This last layer is the opposite: the top few km of soluble rock, where humans have physically crawled and dived. It rides on the surface like the coastlines.
Karst belts — glowing patches over carbonate/evaporite outcrop (after the World Karst Aquifer Map). These stand for the unmapped frontier: caves almost certainly thread this rock, yet only a sliver has ever been surveyed. Tint shows regime — coastal/flooded, alpine, continental.
Flooded / underwater caves — cenotes, blue holes, springs, sumps. Water-filled karst (e.g. Yucatán's Ox Bel Ha, the Hranice Abyss, Taam Ja').
Dry long caves — air-filled passage, sized by surveyed length (Mammoth, Jewel, Optymistychna…).
Deepest caves — vertical depth records (Krubera-Voronya −2,224 m, Veryovkina, the Arabika Massif).
Dots are real surveyed systems; their size encodes length or depth. Hover for stats & source; click to look a system up. Belts & cave figures: WOKAM (Goldscheider et al. 2020), the UIS/Bob Gulden long- & deep-cave lists, and the underwater-cave literature — current to 2025–26.
REAL 3-D CAVE SURVEYS
Fly to a surveyed cave
These caves carry real survey data — pick one to swing the globe to it and open its panel, then descend into the actual passages. (They also appear as pulsing rings on the globe in the Cave scans layer.)
CONFIGURATIONS
Presets
Save the whole current view — depth, every layer, all dials, the data pipeline & enabled models — under a name. Click one to restore it; ★ sets the default that loads automatically. Or copy a share link to send a view to someone.