Close Sources & method
This globe viewer starts in Normal imagery mode. It also includes two pre-calculated practice-only estimation modes built from the supplied screenshots. Approximation converts each valid embedded source-map pixel independently, at the native 1551 × 776 resolution, to the selected two-color palette (black → teal-green by default). No valid source pixels are spatially interpolated, blurred, smoothed, resized, or averaged. Identifiable blank gaps and annotation pixels alone receive a nearest valid donor value inside the layer’s data band. The top and bottom latitude cutoffs remain hard. Interpolation uses that precomputed approximation field to render a denser surface. Direct downloads of the published TIFFs could not be completed in the authoring session, so these modes are still stand-ins for the real numerical grids.
Genuine camera imagery and measured elevations
Normal imagery: The same 2048 × 1024 natural-color Moon texture used by Lunar Regolith Composition, from NASA SVS’s CGI Moon Kit . It is prepared for visualization from LROC color imagery, with adjusted exposure and white balance, filled polar coverage and small inpainted data gaps. It is a display texture; colors are never used to calculate oxide abundances. Both smooth and measured-relief imagery use the former LRC globe’s directional lighting and gamma response.
Returned samples: Apollo 11–17 (excluding Apollo 13), Luna 16/20/24 and Chang’e 5/6 use the landing locations, sample descriptions, six oxide values and individual citations retained from Lunar Regolith Composition. Each value describes a returned sample or selected site average; these point measurements do not imply uniform composition across a region. They remain separate from the atlas’s estimated fields and imported grids.
Measured terrain: LRO / Lunar Orbiter Laser Altimeter (LOLA) , LDEM_16 V3.1 , dataset LRO-L-LOLA-4-GDR-V1.0. Laser altimetry, not radar. Measurements 2009-07-13–2016-11-29; product created 2019-03-15. Native grid: 5760 × 2880, 16 px/°, ≈1.895 km latitudinal spacing. Signed Int16 DN × 0.5 m gives elevation relative to a sphere of radius 1,737,400 m about the lunar center of mass, in the Mean Earth / Polar Axis DE421 frame; this is not a geoid or sea-level datum.
The published global grid includes means and interpolation where necessary; a populated grid cell is not necessarily a direct laser return. No missing sentinel is declared and none occurs in these source bytes. This viewer retains one nearest source sample per 4 px/° cell (1440 × 720, ≈7.58 km latitude spacing), shifts the source 0–360° columns to −180–180°, and keeps half-meter values unchanged. Equidistant source-center ties consistently select the southeast sample, offset 0.03125° each way from the target center. Readouts bilinearly sample this retained DEM with longitude wrapping and latitude clamping. Exact poles use the adjacent row mean. A 720 × 360 cell mesh (0.5°, ≈15.2 km at the equator) samples this DEM to displace the globe; no procedural relief is used. The displayed limb, coordinates, pins and source-aligned place overlays use the same displaced geometry. Terrain no-data would omit triangles and return unavailable elevations; it is never substituted with a measured zero.
Vertical scale: 1× is true radial scale. 5× and 10× exaggerate elevation above/below the reference sphere, explicitly labeled; readouts and abundance data never change. Geometry applies to every imagery/abundance mode and imported layer. Native Approximation fields remain sampled with nearest neighbors, and Interpolation remains separate. GPU picking uses 16-bit angular coordinates (quantization ≤0.0055° longitude / 0.0028° latitude); the CPU fallback stores actual surface coordinates. Globe display raster is capped at 1024 × 1024. This is global context terrain, not a site-engineering DEM. Possible source artifacts at 45° latitude-band boundaries and the existing approximate screenshot/overlay registration remain limitations.
Full source URLs, processing, checksums and coverage metadata · Verification evidence and limits . All runtime assets are bundled with this Site; viewing makes same-origin asset requests, not remote source-data requests.
Reference and numerical archives Yang, C., Zhang, X., et al. (2023). Comprehensive mapping of lunar surface chemistry by adding Chang’e-5 samples with deep learning. Nature Communications, 14, 7554. 10.1038/s41467-023-43358-0 .
Author datasets: Figshare 24081438 and Figshare 24460114 . Inspect file names and metadata before choosing a layer or model variant. The dataset descriptions include MgO and Mg# as well as the five oxides shown here; Mg# is not an oxide wt% layer.
Preview provenance Embedded previews come only from the five graphics supplied in this conversation: map_1.png (TiO₂), map_3.png (FeO), map_5.png (Al₂O₃), map_4.png (CaO), and map_6.png (SiO₂). Each is cropped to the plotted world rectangle (pixels x=260…1810, y=122…897), approximately −180°…180° longitude and −90°…90° latitude. Polar title/annotation bands outside the illustrated data are suppressed. The colorbar is copied from the image.
The pre-calculated field inverts each source pixel against that layer’s embedded colorbar and retains the existing manually transcribed legend min/max values. Approximation textures are lossless PNGs, exactly one output pixel per embedded source pixel. A mask separates the verified source place-name footprints and red reference markers from the abundance field, including their antialiased edges. The five screenshots help identify ink obscured by bright backgrounds. Low-abundance dark pixels and ambiguous colored stripes outside these footprints are retained. Place names and reference points are rendered as independent optional overlays with adjustable opacity; they never enter abundance sampling, interpolation or numerical imports. The original embedded screenshots retain the source lettering. The displayed names were checked visually and cross-referenced against the USGS/IAU lunar nomenclature catalogue . Overlay positions follow the existing screenshot registration; official catalogue coordinates are stored separately, and joined source labels retain both feature names. Gap filling changes only masked pixels and wraps longitude without crossing either latitude boundary. The native band edges are measured separately for each image. Interpolation bilinearly upsamples this completed scalar field 2×, with periodic longitude and clamped latitude inside the band. Pixel-center sampling is consistent between the map, globe and inspector. Browser downscaling can omit sub-screen pixels but does not blend them in Approximation. The smooth globe re-displays the same image or imported grid. Measured terrain optionally displaces this surface with the separately documented LOLA DEM. Graticules and pins are interface overlays. The projection and image registration are suitable for this visual exercise, not exact site-scale sampling.
Import actual abundance numbers Run the included converter on locally available numerical GeoTIFFs, assigning the correct oxide explicitly:
python prepare_oxide_grid.py \
--layer FeO "/path/to/FeO.tif" \
--layer TiO2 "/path/to/TiO2.tif" \
--step-deg 0.5 --output oxide_grids.json Then choose Import oxide grid JSON . Imports take priority for their matching layers; other layers keep the selected Approximation or Interpolation behavior. Multiple files for the same oxide may be passed as additional --layer FeO FILE arguments for a tiled mosaic. Files must be scalar abundance data, not RGB pictures. Missing georeferencing is rejected unless you explicitly supply geographic bounds.
The converter averages available source samples into a coarse longitude–latitude grid (0.5° by default), retains missing coverage as no-data, and records metadata. The displayed point uses the cell containing the selected coordinates; it is not a new in-situ measurement or the value of a native high-resolution pixel. Averaging occurs within the output grid cell in the chosen geographic projection, not as a geodesic area-integrated resource estimate. Quantization is 0.01 wt% for compact transport; the viewer displays one decimal place, which is not a claim about model accuracy.
Material per tonne of regolith processed One tonne means 1,000 kg. For an oxide at w wt%, its mass is 10 × w kg. The ideal element yield is this oxide mass multiplied by the element’s share of the oxide’s formula mass; the remaining stoichiometric mass is oxygen. FeO uses Fe/(Fe + O), TiO₂ uses Ti/(Ti + 2O), Al₂O₃ uses 2Al/(2Al + 3O), CaO uses Ca/(Ca + O), and SiO₂ uses Si/(Si + 2O). Imported MgO also contributes Mg/(Mg + O). Atomic weights follow CIAAW : O 15.999, Fe 55.845, Ti 47.867, Al 26.9815384, Ca 40.078, Si 28.085, Mg 24.305. At 5.8 wt% FeO, 58 kg of oxide contains about 45.08 kg Fe and 12.92 kg oxygen.
The O₂ readout sums the oxygen mass from each available compound separately. Combining oxygen atoms into O₂ changes the molecule count, not the mass. Missing values stay unavailable; partial coverage is explicitly a subtotal. Unmapped oxides are excluded. The independently mapped oxide fields are used as supplied without renormalizing their sum. Calculations use unrounded sampled values, including imported grids where present; colors and terrain exaggeration do not affect yields. These are theoretical full-reduction inventories with 100% separation and recovery, not predictions of a particular processing plant.
Scientific limits The study’s maps are remotely inferred, surface-sensitive oxide-equivalent abundances. The presence of a reported oxide does not by itself establish a deposit of the free oxide mineral, excavation yield, process recovery, or subsurface composition. Missing cells remain unavailable. The viewer does not force the five layers to sum to 100% or infer missing MgO, water, or other unmapped compounds. The material column calculates ideal elemental and oxygen inventories from the available oxide-equivalent abundances; actual extraction and recovery are not modeled. Approximation and Interpolation remain visualization-derived practice estimates from screenshots, not author-published values, and they inherit labeling artifacts, stretch choices, and any errors in reading the legend limits. Coverage and artifacts must be assessed in the original files.
Data handling The original abundance payload remains embedded unchanged. The viewer loads bundled camera, DEM and renderer assets from the same Site. Source links open only when clicked. Imports are processed locally in memory. Reloading the page discards them; export preserves the current coordinate, values (if available), and source metadata.