Earth · Deep Time — sources and model limits
This is an exploratory Earth sandbox. Its starting geography uses published data; its evolving conditions are qualitative approximations. Weather, seasons and geology use different compressed clocks. Dates, temperatures, currents and regional futures should be read as simulation outcomes rather than calibrated predictions.
Current validation limits — 18 September. Focused browser checks cover finite aquifer and sediment conservation, ice cold content, pressure-dependent boiling and complete save/load. Existing water, fire and lava operators pass. Current UI checks use Low and Mid; earlier five-tier and extended climate reports are historical. The prior 400-day Arabian/Australian greening case has not been rerun. Process-based drought controls improve vegetation coupling without a claim of globally corrected regional climate. Weather and ecology use resolved weather time; landscape processes use capped geological time.
Continents and plates. The 52 plate outlines come from Peter Bird’s PB2002 model (2003), converted by Hugo Ahlenius / Nordpil under the Open Data Commons Attribution License 1.0. Angular velocities use Argus, Gordon and DeMets’ NNR-MORVEL56 (2011). Four additional plate divisions are absent: Capricorn and Macquarie follow Australia, Lwandle follows Somalia, and Sur follows South America. Present-day rotations are extrapolated. Continuous, bounded cubic transport moves material without whole-cell jumps; local plate ownership is checked against the cumulatively rotated reference geometry. Local boundary strain accumulates and releases episodically. Rigid plate interiors remain coherent, while collision, subduction, rifting and uplift use local rules. Plate reorganization and detailed future mountain belts are unresolved. Fixed mantle-source locations in this reference frame produce illustrative hotspot tracks. Compare four published GPlates-derived future scenarios: Pangaea Ultima, Novopangaea, Aurica and Amasia. Their 50 actual published grids provide alternative references; the sandbox does not reproduce their complete future plate reorganizations.
Terrain, rivers and ice. Coastlines, river traces, lakes and glaciated areas use Natural Earth cartographic data, released into the public domain. Land elevation is resampled from NASA Earth Observatory / GEBCO imagery credited to Jesse Allen; its 0–6,400 m grayscale range clips higher peaks. Seafloor depth interpolates Natural Earth’s SRTM Plus-derived contours, capped at 10,000 m. These are generalized maps, not detailed elevation surveys.
Mapped river reaches seed shallow channels and water. Low retains 276 selected major river features; other tiers retain 570 major features and 46 major lakes. Matching five-tier terrain assets keep their directed initial grades. River cells were reduced by about 87%, and the old minor-stream cuts were removed from the terrain. The Columbia’s broad estuary connection is generalized. At Extreme detail an equatorial cell spans about 10 km; the source terrain is approximately 20 km, so channel widths, lake depths, discharge and sediment loads are schematic. Water follows evolving hydraulic head; erosion, deposition, glacier barriers and meltwater can alter drainage. Greenland and Antarctica use paired real NOAA ETOPO1 bedrock and ice-surface grids (2009), accessed through PacIOOS ERDDAP and sampled at 10 arcminutes. Grounded thickness comes from their difference; shelf thickness and grounding classification use hydrostatic approximations. Below-sea-level beds remain grounded beneath sufficient ice. The loaded bed is retained and relaxes toward an illustrative unloaded height with a 3,500-year response time. This is delayed local isostasy, not a full elastic-lithosphere model. The product is older than current BedMachine surveys. Outside those sheets, the alpine glacier prior includes historical terrain and estimated thickness. Ice-age starts are scenarios rather than historical reconstructions.
Layered weather. Seven pressure levels (20, 75, 200, 500, 700, 850 and 925 hPa) carry temperature, humidity, cloud, height and flow. Circulation uses a fixed coarse mesh; the lowest air layer exchanges heat, moisture and rainfall with every surface cell. Twelve 150-second circulation steps run inside each half-hour world update. Shared transport precedes midpoint rotation and pressure adjustment. One evolving base-height field uses the linearized hydrostatic pressure scale Rd × 288 K / g (about 8,430 m); immediate hydrostatic reconstruction couples upper pressure heights to the temperature column. A GPU reduction holds the global pressure reference fixed without changing local gradients. Hydrostatic thickness, radiation, rotation, bounded vertical parcel exchange and latent heating couple the layers. These are idealized dynamics; the pressure coupling, polar treatment and moist physics remain approximate. Background: hydrostatic pressure and height, UCAR atmospheric layers, the Held–Suarez idealized circulation benchmark, and Thatcher & Jablonowski’s moist benchmark. This sandbox does not implement or claim to pass either benchmark.
Weak initial lows develop under the evolving circulation. Find a storm locates an existing cyclonic pressure depression; Seed a low is an explicit sandbox intervention. Clouds and night-time lightning use simulated condensation and rain. Low and middle cloud opacity converts condensate to pressure-weighted water path, with schematic effective sizes of 12 micrometres for liquid and 35 micrometres for ice. Cirrus uses a capped optical proxy; cloud microphysics and coverage remain approximate. A six-hour advected parcel-lift memory lets sustained convergence build clouds and subsidence clear them; terrain cooling is counted separately. The vapor-transport view integrates humidity times wind through the column. NOAA’s atmospheric-river description and CW3E’s AR scale explain the underlying transport concept. A bright instantaneous filament is not a complete atmospheric-river classification. Neither storm tracks nor hurricane intensities are calibrated, and no observed forecast is ingested.
Ocean depth, ice and heat. Four temperature layers span 0–50 m, 50–250 m, 250–1,000 m and deeper water, truncated at the seabed. Surface and simplified subsurface currents advect heat; an implicit column solve exchanges it vertically. Winds, thermocline displacement and upwelling interact with surface temperature. NOAA ENSO and upwelling inform those feedbacks. The model has ENSO-related mechanisms, without a calibrated or demonstrated irregular ENSO cycle; methane hydrates and detailed overturning circulation are unresolved. A phase-aware heat budget freezes and melts floating ice near its salinity-dependent freezing point, following the principles described by NSIDC and TEOS-10. Ice volume and concentration drift with water and wind. Thick shelves have lower face mobility than thin pack; draft and receiver-capacity limits prevent rapid unphysical pileups without deleting ice. This proxy does not resolve shelf attachment or iceberg shapes. Warm air can coexist with ice while latent heat is still being supplied.
Grounded ice, snow, freshwater ice and liquid share a finite phase-energy transaction. Snow densifies into firn; compaction over deep water adds floating ice. Slope, thickness and temperature control slow glacier deformation; motion abrades terrain. Grounding depends on draft and bed depth, and ice loading retains the delayed rebound model. Boiling now interpolates IAPWS saturation anchors against estimated elevation and steam pressure. Land ice stores capped internal cold content, following the USACE energy-budget principle. Three soil depths evolve independently with an apparent heat capacity near freezing. Marine ice retains an insulating phase approximation. Global steam condensation returns queued latent heat to low-level air. Approximate transport and atmospheric restoring prevent exact planetary energy conservation. Full glacier thermal profiles and supercritical/sub-triple-point water regimes are unresolved.
A separate dry convective adjustment moves heat and moisture between unstable adjacent layers from 925 through 500 hPa. Pair exchanges conserve represented sensible enthalpy and vapor; a schematic half-hour relaxation resolves gradual overturning. The general adjustment concept follows Manabe and Strickler (1964); the implementation is a reduced pairwise closure. Near-surface exchange uses a distinct 10 m wind diagnostic from the nominal 750 m wind, with water-relative velocity and a coefficient-consistent neutral profile. Atmospheric transport retains its 925 hPa wind. Extrapolating through this unresolved boundary layer is approximate, and does not implement the full height- and stability-aware COARE bulk algorithm.
Regional climate and time. Initial ocean temperatures have a marine latitude profile. Ocean fetch and actual crossed terrain initialize coastal moisture and rain shadows; these are schematic initial conditions, not mapped rainfall. Runtime precipitation consumes advected vapor, and orographic cooling uses only terrain above sea level. Root-zone storage and slower woody vegetation response help forests survive short dry intervals, consistent with the processes described in the Olympic National Park climate notes. Soil, climate and succession still determine vegetation classes; species and plant hardiness zones are not mapped. Weather uses 150-second circulation steps inside half-hour world updates, with 365.25 days per seasonal year. Ordinary surface/ocean heat exchange, snowfall and phase-energy changes share the resolved weather interval. Morphological ice flow, densification and erosion accelerate separately while moving existing material. Tilt sets daylight and seasonal heating. Precession moves perihelion relative to the 365.25-day seasonal clock, and orbital distance affects insolation, following the geometry described in NASA’s orbital-cycle overview; greenhouse gases, sulfate cooling, ice and vegetation albedo affect thermal targets. Solar expansion remains illustrative.
Volcanoes and mantle sources. Selected names, locations and principal rock types follow the Global Volcanism Program, Smithsonian Institution, Volcanoes of the World v5.4.0 (7 August 2026), compiled by E. Venzke; see GVP terms and citation guidance. The catalogue links individual records. USGS Hawaiian volcano evolution and the Columbia River Basalt Group inform the examples. Silica, viscosity, magma supply, eruptions, ash and sulfur conversion are simplified. Deep submarine lava is hidden beneath the ocean in the living surface view and can be inspected in Rock & crust. Mantle convection and hotspot pulses are invented sandbox dynamics; no universal hotspot cycle or future eruption date is implied. The Columbia flood-basalt province is historical, with no automatic future eruption inferred.
Visual and interaction inspiration: David A. Roberts’ [SH18] Humanity, The Powder Toy, and Tropical Tidbits’ weather-map presentation.
Simulation storage uses seventeen full-resolution 32-bit fields plus coarse atmospheric buffers. This avoids small heat, ice and climate changes disappearing through half-precision rounding. Numerical regression checks cover layered temperatures, physical transport across the five detail tiers, latent freezing/melting/boiling, polar melt, regional climate, delayed rebound, small plate displacements and seasonal daylight. These are behavior checks, not a claim of calibrated Earth-system skill.
Regional refinement. The clock starts at one hour per second. Weather follows selected speed up to four days per second, then remains resolved while geological time accelerates. A March-equinox calendar displays both hemispheres’ seasons and actual axial tilt; the night view fades out at one million years per second. Ocean-fetch initialization now includes equatorward moisture paths and subtropical aridity, informed by NWS low-level jet guidance and the Bureau of Meteorology’s subtropical-ridge explanation. These are schematic initial conditions. Plant-available water excludes the wilting reservoir; growing-season warmth allows boreal trees to survive cold winters. Incoming vertical air displaces recipient air and carries its humidity and compressional heating; bounded mixing preserves a constant humidity tracer and avoids adding vapor to a fixed air mass. Terrain-adjusted temperature uses matching interpolation weights. Moving parcels use the dry adiabatic lapse rate with explicit latent heating, following NWS parcel thermodynamics. Dry, bare soil receives a bounded sensible-heating anomaly consistent with the energy-partition mechanism in NASA’s land-surface model description; its 6 K amplitude is illustrative and uncalibrated. Seasonal climate memories determine biome recruitment, while temporary moisture pulses can grow vegetation within that regime. Established living forest retains its class through a cold season; loss of woody cover permits tundra succession. Major channels are carved once and conditioned separately at each detail; directed bare-bed grades are preserved, while ice or lake levels can still cause local barriers. Warm mapped valleys exclude misplaced historical alpine ice, avoiding invented large meltwater pulses. Real ETOPO sheets remain intact. Coarse riverbed heights remain approximate. Freshwater now has its own temperature, salinity/load and explicit freezing/melting/boiling state. A shared-face head limit stabilizes deep lakes while allowing rivers to drain.
v6 surface materials. Ten substrate families and nineteen terrestrial vegetation classes follow the separate soil/vegetation organization used by GFS Noah, Noah soil/vegetation tables and ECMWF IFS physical processes. Their coefficients here are illustrative, not copied operational parameter tables. Persistent substrate, soil wetness and vegetation affect thermal storage, infiltration, erosion and albedo. Actual skin temperature drives thermal ignition and rock melting. Char is separate from basalt, and cooled lava buries existing cover before later weathering and deposition.
Finite inland water and ice. The Caspian is a closed brackish basin, with fresher northern water. Its schematic initial surface is −28 m; other major lake depths are approximate. Dissolved salt follows liquid flow and remains after evaporation. Phase changes use a single water/ice/steam transaction. The distinction between drifting sea ice and grounded glacier ice follows NSIDC sea-ice processes and NASA glacier motion; ageing alone does not turn floating pack into a land ice cap.
Convection and currents. Pressure-weighted heat/moisture exchanges follow the mechanism described in ECMWF’s mass-flux convection teaching material. Warm-water supply, shear and land friction matter for tropical development, consistent with NOAA hurricane guidance. Frontal ascent follows moving thermal gradients. The numerical tests show spontaneous cloud and circulation development, not validated hurricane climatology. Wet ocean boundaries, wind stress, Coriolis and density-related subsurface pressure replace simple coastal zero-value derivatives and prescribed deep return flow. NOAA’s currents overview explains the basin-scale mechanisms; these remain reduced dynamics, not observed-current assimilation.
Mountain belts and volcanic construction. Broad shortening belts, displaced overriding-plate arcs and buoyancy-dependent accretion follow the distinctions in NPS subduction-zone geology. Seamounts may subduct; accretion is not universal, and obduction specifically means oceanic crust/upper mantle emplacement onto a continent. Current geometry is approximate. Geological-time magma construction raises seamounts and islands. Simultaneous local eruption plumes, ice/water quenching, volume-scaled CO₂ and sulfur emissions couple volcanoes to climate; the 200,000-year carbon-removal timescale is a sandbox parameter.
Display and performance. Low/Mid/High/Ultra/Extreme use 512/1024/2048/3072/4096-wide grids. Larger grids add simulation cells by resampling existing source geography. The independent column shows three soil temperatures and aquifer storage. Crust and mantle temperatures remain explicitly unresolved. Debug distinguishes allocated GPU bytes from optionally measured JS heap. The GPU backend remains WebGL2. All current production passes fit the guaranteed sixteen fragment-texture inputs.
Surface heat and sustained moisture supply. The open-water exchange uses the relative-wind and humidity-deficit bulk mechanism described in Bonino et al. (2022), with fixed illustrative coefficients rather than the complete COARE algorithm. Liquid availability bounds evaporation. Surface radiation includes absorbed sunlight and a humidity-dependent longwave sky, following the Brutsaert emissivity relation and its cloud extension discussed by Tian et al. (2023). Transmission and cloud coefficients are schematic. Radiation, latent heat and ocean mixing share weather time. Ice cooling uses an insulating slab approximation; positive radiation reaches surface melt. Colder middle-tropospheric targets are informed by NOAA AOML tropical soundings, with a dry-stability floor and the simulation's virtual pressure-level datum. Heat and vapor now use matching vertical air-exchange faces.
Surface evaporation pressure. Saturation at exposed water uses local hydrostatic surface pressure, combining the evolving 925 hPa height anomaly with surface elevation relative to sea level. Sea-floor depth does not change marine air pressure. The nominal 288 K scale height and well-mixed near-surface humidity are reduced approximations. Atmospheric condensation still uses each pressure level. This follows the surface-pressure distinction in the COARE input documentation and sea-surface saturation function; the sandbox does not implement full COARE.
Wind-driven upwelling. The surface ocean uses depth-integrated Ekman transport and its spherical divergence, following the mechanism in MIT's upper-ocean notes and NOAA PMEL's tropical-Pacific dynamics review. Closed coastal faces allow offshore transport to create coastal upwelling. A three-degree equatorial regularization prevents division by zero and is explicitly approximate. The prior internal-current divergence source and extra thermocline multiplier were removed because they erased tropical warm pools in the coupled test. Entrainment now uses the diagnosed water-volume rate; the full equatorial wave and ENSO system is not resolved.
Groundwater, ecology and material transport. USGS water budgets guide finite recharge, aquifer storage, lateral flow and baseflow. Material retention, wilting point, root depth and seasonal leaves control plant water use. Drought, fuel and recovery vary by vegetation structure. Plant/soil carbon, ocean exchange, weathering and burial modify a bounded CO₂ feedback. These are illustrative parameters, not mapped hydrogeology or calibrated ecology. Finite loose material moves by wind, wet-slope failure and tidal/wave shoreline exposure. Atmospheric dust, surge and resolved waves are unfinished. Salinity and thermal stratification affect ocean mixing; its four thermal slabs remain fixed. Visible eruptions deplete finite magma storage; rifting thins crust. Autonomous plate splitting and merging remain unresolved.