LunarGeo: Lunar Geoscience & Mapping

• LunarGeo is a cluster of lunar geoscience and geospatial inference practices that encompass stereo imaging, DEM fusion, and calibrated mapping of the Moon’s surface, near-surface, and interior.
• It integrates methods ranging from high-resolution geochemical cartography and physically based simulations to advanced stereo sub-datasets for 3D reconstruction and precise pose estimation.
• The framework also supports deep-interior geophysics, near-surface stratigraphy, and emerging tomographic modalities—enabling improved navigation, hazard avoidance, and lunar structural analysis.

LunarGeo is a non-uniform designation used in recent lunar-science literature for several closely related enterprises: geospatially explicit lunar mapping, geophysical network design, geochemical cartography, physically based simulation, and supervised perception datasets. In MoonAnything, LunarGeo is the stereo-vision sub-dataset that provides dense depth maps and camera calibration for 3D reconstruction and pose estimation (Grethen et al., 1 Apr 2026). In other sources, the name appears in an in-depth report on the Lunar Geophysical Network (Haviland et al., 2021), a technical report on Chandrayaan-2 CLASS geochemical mapping (Kumar et al., 21 Aug 2025), and as a shorthand for the lunar tidal signature in geomagnetic records (Rosales et al., 2015). This suggests that “LunarGeo” is best understood not as a single standardized artifact, but as a cluster of lunar geoscience and geospatial inference practices spanning the Moon’s surface, near-surface, and interior.

1. Terminological scope and explicit usages

Several distinct uses of the term are explicit in the literature.

Usage of “LunarGeo”	Source	Technical content
Stereo sub-dataset in MoonAnything	(Grethen et al., 1 Apr 2026)	Stereo images, dense depth, calibration
“LunarGeo” synthesis for simulation	(Lebreton et al., 2024)	Unified multi-resolution DEM and rendering workflow
“LunarGeo” report on LGN	(Haviland et al., 2021)	Four-station lunar geophysical network
“LunarGeo” report on CLASS mapping	(Kumar et al., 21 Aug 2025)	Global X-ray line-ratio geochemistry
“LunarGeo” geomagnetic “tide”	(Rosales et al., 2015)	Lunar daily variation in geomagnetic data

In MoonAnything, LunarGeo is a geometric benchmark: 58 000 stereo pairs rendered at $512\times512$ px, with 38 000 South Pole pairs and 20 000 Tycho crater pairs, accompanied by depth, intrinsics, poses, baseline, and metadata (Grethen et al., 1 Apr 2026). In the SurRender study, “LunarGeo” denotes an overview built around a continuous global DEM at 20 m resolution, procedural fusion of local high-resolution tiles, and real-time rendering for precision navigation (Lebreton et al., 2024). In the LGN landing-site rationale, “LunarGeo” labels a mission-scale geophysical program centered on seismology, lunar laser ranging, heat flow, and magnetotellurics (Haviland et al., 2021). In the Chandrayaan-2 CLASS study, it labels a geochemical mapping framework based on O/Si, Mg/Si, Al/Si, Mg/Al, Ca/Si, and Fe/Si line-intensity ratios at 5.3 km/pixel (Kumar et al., 21 Aug 2025). In the Huancayo geomagnetic study, it denotes the coherent lunar tidal signature isolated from solar and seasonal sidebands (Rosales et al., 2015).

A common misconception is to treat LunarGeo as a single mission, dataset, or software package. The documented usage is broader and more heterogeneous. The shared denominator is rigorous lunar characterization with explicit geometry, calibrated observables, and inversion or reconstruction pipelines.

2. Geochemical and compositional cartography

One major LunarGeo strand is high-resolution compositional mapping of the lunar surface. Using a single-exposure Gaofen-4 lunar disk from 2018-07-28 04:49 UTC, Lu et al. generated seamless nearside maps of FeO, TiO$_2$, MgO, Al$_2$O$_3$, CaO, and SiO$_2$ at $\approx 500$ m/pixel. The method used only the near-IR band 5 at $760$–$900$ nm, with lunar effective $\lambda \approx 0.81\,\mu$m, and correlated reflectance with oxide wt% at twenty-two Apollo, Luna, and Chang’E-3 sample stations (Lu et al., 2020). The resulting regressions included, for example, $\mathrm{FeO}=49.652\cdot R^{-0.0943}$ with $_2$0, $_2$1 wt%, and $_2$2 with $_2$3, $_2$4 wt% (Lu et al., 2020). Mare averages were reported as FeO $_2$5 wt%, TiO$_2$6 $_2$7 wt%, MgO $_2$8 wt%, Al$_2$9O$_2$0 $_2$1 wt%, CaO $_2$2 wt%, and SiO$_2$3 $_2$4 wt%; highland averages were FeO $_2$5 wt%, TiO$_2$6 $_2$7 wt%, MgO $_2$8 wt%, Al$_2$9O$_3$0 $_3$1 wt%, CaO $_3$2 wt%, and SiO$_3$3 $_3$4 wt% (Lu et al., 2020).

The same general LunarGeo problem is addressed in X-ray fluorescence by Chandrayaan-2 CLASS. CLASS operates over $_3$5–$_3$6 keV with an energy resolution of $_3$7 eV at 5.9 keV, from a $_3$8 km polar orbit, recording 8 s frames that are re-binned to 96 s, and for low-SNR lines to 296 s, to yield an effective pixel scale of 5.3 km $_3$9 5.3 km after spatial gridding (Kumar et al., 21 Aug 2025). Using $_2$0 validated line detections, the study produced global maps of O/Si, Mg/Si, Al/Si, Mg/Al, Ca/Si, and Fe/Si. The Mg/Al map was identified as the ratio that best represents geochemical differences between the Procellarum KREEP Terrane, Feldspathic Highlands Terrane, and South Pole–Aitken Terrane; high Mg/Al $_2$1 marks mare basalt units and the PKT margin, whereas low Mg/Al $_2$2 corresponds to feldspathic highlands (Kumar et al., 21 Aug 2025). A Gaussian Mixture Model in $_2$3 space isolated three-component and five-component terrane partitions, and the proxy

$_2$4

was used to distinguish young, low-Mg# mare flows from older, high-Mg# highlands (Kumar et al., 21 Aug 2025).

These mapping programs converge on classical petrogenetic contrasts. Highlands are Al- and Ca-rich and consistent with plagioclase flotation, maria are Fe-, Mg-, and Ti-rich and consistent with mantle-derived basalt volcanism, and regional asymmetries can constrain impact geometry. At Tycho, the Gaofen-4 maps show that south and east ejecta are enriched in Al$_2$5O$_2$6, CaO, and SiO$_2$7, whereas the north and west are poorer in these crustal oxides; the study interprets this as a low-angle projectile arriving from the southwest (Lu et al., 2020).

3. Simulation infrastructures and supervised geometric datasets

A second major LunarGeo meaning is computational: the construction of unified geometric environments for rendering, perception, and guidance, navigation, and control. In the SurRender study, the baseline global model combined a Chang’e-2 20 m DEM with Kaguya/SELENE 118 m albedo, and fused 5 m LRO DEM tiles procedurally below $_2$8 km altitude. The result was a final continuous global DEM at 20 m resolution covering $_2$9 steradians (Lebreton et al., 2024). The software stack reprojected heterogeneous PDS and GeoTIFF tiles into a single cube-map and LOD hierarchy, stored as pyramidal “.BIG” data, and used conemaps, memory-mapped I/O, and a CPU-only path tracer to avoid VRAM limitations on multi-terabyte datasets. Reported performance was $\approx 500$0 at 15 Hz with 1 ray/pixel, and high-quality rendering at 100 rays/pixel in 5 s, with residual noise $\approx 500$1 LSB and sub-pixel rendering errors (Lebreton et al., 2024).

The rendering physics is explicit. Camera rays are intersected with a lunar sphere via

$\approx 500$2

heights are interpolated within a pyramid, and surface radiometry uses the rendering equation with a Hapke BRDF for lunar regolith (Lebreton et al., 2024). The workflow supports simulated descents from 1 500 km down to 20 km altitude and is intended for closed-loop GNC, hazard detection, and final inertial alignment (Lebreton et al., 2024).

MoonAnything formalizes the same agenda as a benchmark. Its LunarGeo sub-dataset comprises stereo images with corresponding dense depth maps and camera calibration, explicitly targeted at stereo matching, multi-view 3D reconstruction, and pose estimation (Grethen et al., 1 Apr 2026). South Pole scenes cover $\approx 500$3 km around the lunar south pole using a LOLA DEM at 5 m/px; Tycho scenes cover an approximately $\approx 500$4 km area using an Airbus-PixelFactory DEM at 1 m/px (Grethen et al., 1 Apr 2026). The camera model is ideal pinhole, without radial or tangential distortion. South Pole images use a $\approx 500$5 FoV, Tycho images a $\approx 500$6 FoV, with intrinsics $\approx 500$7, $\approx 500$8, $\approx 500$9 for $760$0 px (Grethen et al., 1 Apr 2026). The baseline is sampled between 2% and 22% of current altitude, under nadir, oblique, and dynamic trajectories (Grethen et al., 1 Apr 2026).

Ground-truth depth is generated by physically based ray tracing on the real DEM through the SurRender engine. The stereo relation is the standard

$760$1

with no added synthetic noise; depth uncertainty is attributed only to DEM sampling, stated as $760$2 m for the South Pole DEM and $760$3 m for the Tycho DEM (Grethen et al., 1 Apr 2026). Baseline experiments fine-tuning VGGT and MASt3R on the South Pole training set showed large improvements relative to unfine-tuned models on both seen and unseen test regions, underscoring the domain-specific character of lunar stereo and reconstruction (Grethen et al., 1 Apr 2026).

This computational branch of LunarGeo is significant because it closes the loop between orbital topography, physically grounded image formation, and downstream autonomy. A plausible implication is that lunar geospatial products are no longer merely cartographic outputs; they are operational assets for landing, hazard avoidance, and machine perception.

4. Near-surface structure, palaeoregoliths, and field geophysics

LunarGeo also encompasses methods for resolving the Moon’s near-surface architecture and stratigraphic record. A central target is the lunar palaeoregolith: ancient regolith trapped between successive mare basalt flows, preserving time-resolved records of surface exposure, galactic cosmic-ray fluxes, energetic events such as supernovae and gamma-ray bursts, and incursions of the interstellar medium (Crawford et al., 2010). Fresh basalt surfaces accumulate regolith by micrometeorite bombardment, impact gardening, and solar wind implantation. Typical growth rates on mare basalts today are 1–1.5 mm Myr$760$4, increasing to 3–5 mm Myr$760$5 at $760$6 Ga; a conservative long-term rate of 2 mm Myr$760$7 implies a soil $760$8 mm thick after 100 Myr (Crawford et al., 2010). Burial by younger flows thermally alters only the top of the old soil; for an overlying flow thickness $760$9–10 m, the thermal wave penetrates to 10–100 cm, so implanted ions and delicate phases below $900$0 remain preserved (Crawford et al., 2010).

The stratigraphy is quantitatively tractable. If $900$1 and $900$2 are the ages of underlying and overlying basalts, then the regolith formed during $900$3. A 0.5 m thick soil implies $900$4 Myr of exposure at 2 mm Myr$900$5 (Crawford et al., 2010). High-energy galactic cosmic rays penetrate deeply, but cosmogenic nuclide production is confined to the top meter, with attenuation approximated by

$900$6

where $900$7–2 m$900$8 depending on density and composition (Crawford et al., 2010).

Detection and sampling strategies link this stratigraphy to modern field geophysics. High-resolution imagery and spectral mapping identify mare units of different ages; orbiting or rover-borne GPR at 5–100 MHz can detect discrete reflectors at 1–20 m depth; thermal-inertia contrasts may reveal thin regolith layers through diurnal response; and robotic or human-assisted drills with modular coring bits can reach 10–100 m depth (Crawford et al., 2010). These goals align with the Artemis III geophysical white paper, which recommends a coordinated surface program of seismic, GPR, and electromagnetic measurements at the lunar South Pole (Schmelzbach et al., 2020).

The proposed seismic instrumentation comprises a broadband three-component 0.01–50 Hz velocity sensor and geophone mini-arrays with four high-frequency geophones at 2–100 Hz, linked by optical fiber for timing and telemetry (Schmelzbach et al., 2020). Controlled sources include an astronaut-operated hammer or low-mass gas-gun at 10–100 m offsets, and a final-stage ascent-vehicle impact at several kilometers (Schmelzbach et al., 2020). Targeted science cases include lobate-scarp fault monitoring, regolith stratigraphy in the top 10–100 m, ice detection in permanently shadowed regions through strong velocity contrasts, and shallow imaging of South Pole–Aitken basin structure (Schmelzbach et al., 2020). The GPR system is dual-frequency, with 500 MHz providing $900$9 m vertical resolution and $\lambda \approx 0.81\,\mu$0 m penetration, and 100 MHz providing $\lambda \approx 0.81\,\mu$1 m resolution and $\lambda \approx 0.81\,\mu$2 m penetration (Schmelzbach et al., 2020). Electromagnetic sounding with a 10 m square transmitter loop and three-axis receiver coils targets regolith, fractured bedrock, ice zones, and basin-scale resistivity structure (Schmelzbach et al., 2020).

In this near-surface sense, LunarGeo is both archival and prospective. It is archival because buried soils may preserve records of spiral-arm crossings, nearby supernovae, and dense interstellar clouds; it is prospective because seismic, radar, EM, excavation, and coring architectures are already being specified for crewed and robotic campaigns.

5. Deep-interior geophysics and precision geodesy

At planetary-interior scale, LunarGeo is represented most directly by the Lunar Geophysical Network. LGN is a four-station, long-lived surface array with a 6–10 year goal, proposed for launch in 2030 (Haviland et al., 2021). Its primary objectives are to identify and characterize any partial-melt layer atop the core–mantle boundary, determine the size, state, and composition of the lunar core, constrain mantle heterogeneity, map crustal thickness and heat-production variations among major terranes, and assess present seismo-tectonic activity (Haviland et al., 2021). Each lander deploys a broadband 0.01–1 Hz VBB seismometer plus a short-period reference sensor, a set of three Next-Generation Lunar Retroreflectors, a 3 m-deep heat-flow probe, and a magnetotelluric sounder (Haviland et al., 2021). The network architecture places three sites on the nearside—P-5, Schickard Basin, and Crisium Basin—and one on the farside at Korolev Basin (Haviland et al., 2021).

The geophysical specifications are explicit. The VBB sensor targets a noise floor $\lambda \approx 0.81\,\mu$3 m/s$\lambda \approx 0.81\,\mu$4 over 0.01–1 Hz, a dynamic range of at least 120 dB, and direct detection of core phases ScS, PKP, and PcP on single records (Haviland et al., 2021). The heat-flow probe measures thermal conductivity $\lambda \approx 0.81\,\mu$5 and temperature gradient $\lambda \approx 0.81\,\mu$6, with heat flow given by

$\lambda \approx 0.81\,\mu$7

The magnetotelluric package measures horizontal electric and magnetic fields to recover the impedance tensor $\lambda \approx 0.81\,\mu$8, with investigation depth scaling as $\lambda \approx 0.81\,\mu$9 (Haviland et al., 2021). Network optimization emphasizes mean inter-station separation $\mathrm{FeO}=49.652\cdot R^{-0.0943}$0, terrane-interior siting, regolith thickness $\mathrm{FeO}=49.652\cdot R^{-0.0943}$1 m for the heat probe, and low local magnetic anomalies (Haviland et al., 2021).

Relative to Apollo, the expected gain is large. For PKP sampling at 180–270$\mathrm{FeO}=49.652\cdot R^{-0.0943}$2, LGN records PKP from 100% of deep-moonquake events with at least one event per 5$\mathrm{FeO}=49.652\cdot R^{-0.0943}$3 bin, whereas Apollo recorded only 55%; LGN yields at least two events per 5$\mathrm{FeO}=49.652\cdot R^{-0.0943}$4 in 89% of bins and at least three in 72%, compared with 16% and 11% for Apollo (Haviland et al., 2021). ScS detections are expected to exceed 80/yr at nearside nodes and PKP detections 30/yr at the farside node, assuming VBB noise $\mathrm{FeO}=49.652\cdot R^{-0.0943}$5 m/s in the RMS 0.07–0.14 Hz band (Haviland et al., 2021).

Precision geodesy extends this interior program. Deployment of new lunar retro-reflector arrays, active laser transponders, and radio beacons at the south pole is proposed as a way to transform lunar laser ranging into a precision geodetic tool (Viswanathan et al., 2020). The science objectives include probing free core nutation and mantle precession angles, refining Love numbers $\mathrm{FeO}=49.652\cdot R^{-0.0943}$6 and $\mathrm{FeO}=49.652\cdot R^{-0.0943}$7, improving tests of the Equivalence Principle and $\mathrm{FeO}=49.652\cdot R^{-0.0943}$8, and tying the lunar body frame to the ICRF through differential VLBI (Viswanathan et al., 2020). South-pole placement is geometrically important because monthly tidal displacements are 11.2 cm at the equator and 4.5 cm at the pole, and a station within $\mathrm{FeO}=49.652\cdot R^{-0.0943}$9 of the south pole can still maximize Earth visibility at roughly 40–50% (Viswanathan et al., 2020). The passive CCR design uses fused-silica prisms with 5 cm base diameter and mass $_2$00 g each, arranged on a common baseplate; active transponders are specified at $_2$01 kg and 5 W; radio beacons at $_2$02 kg (Viswanathan et al., 2020). Passive south-pole LLR is expected to achieve single-epoch RMS $_2$03 mm, while active transponders aim for sub-cm one-way range accuracies (Viswanathan et al., 2020).

Taken together, LGN and next-generation south-polar geodesy define a deep-interior LunarGeo program in which seismic, thermal, electromagnetic, and rotational observables are estimated jointly, rather than in isolation.

6. Navigation, particle probes, and emerging tomographic modalities

A final LunarGeo sense is operational and exploratory: inference architectures for positioning, interior flux measurements, and unconventional tomography. For north-polar surface positioning, Gong and Dempster propose a single-satellite navigation system in a polar low lunar orbit with semi-major axis $_2$04 km, eccentricity $_2$05, inclination $_2$06, argument of perilune $_2$07, and orbital period $_2$08 h, giving a signal-available window of about 12 min per pass (Gong et al., 4 Apr 2025). Doppler shift is modeled through the accumulated delta-range and its time derivative, with finite-difference evaluation of $_2$09, followed by a three-step geolocation algorithm: an algebraic initial estimate, a constrained 2D Gauss–Newton solver on the lunar surface, and an unconstrained weighted 3D Gauss–Newton refinement (Gong et al., 4 Apr 2025). Monte Carlo results over random receiver sites at latitudes 70–90$_2$10 reported Step 1 mean error $_2$11 km, Step 2 $_2$12 km, and Step 3 $_2$13 km after one pass; with two consecutive passes the 99%-ile drops to hundreds of meters for the better ephemeris case, and ten passes yield sub-10 m performance (Gong et al., 4 Apr 2025). The dominant error term is ephemeris error, and the study explicitly recommends multi-pass processing to resolve the one-pass mirror ambiguity (Gong et al., 4 Apr 2025).

For the Moon’s radiogenic interior, geoneutrino predictions provide another LunarGeo observable. A refined lunar interior model with five geochemical reservoirs and a core radius of 380 km predicts integrated $_2$14 fluxes of $_2$15 cm$_2$16 s$_2$17 and $_2$18 cm$_2$19 s$_2$20, with a PKT/FHT ratio of 8.63 (Hu et al., 2 Mar 2026). At the PKT site, the flux breaks down into $_2$21 cm$_2$22 s$_2$23 from $_2$24U, $_2$25 cm$_2$26 s$_2$27 from $_2$28Th, and $_2$29 cm$_2$30 s$_2$31 from $_2$32K (Hu et al., 2 Mar 2026). Proposed detection channels are inverse beta decay on protons, elastic scattering on electrons, and the radiochemical reaction $_2$33 (Hu et al., 2 Mar 2026). The IBD rate at PKT is 20.58 kt$_2$34 yr$_2$35, and the study concludes that a 25 kt·yr IBD detector buried in a 50 m deep lava tube in PKT can measure total geoneutrino flux to 4% and Th/U to 27%, while a $_2$36He assay offers unique access to $_2$37K (Hu et al., 2 Mar 2026).

The newest tomographic direction uses gravitational waves. In a perturbative framework for calibrated GW forcing, small radial perturbations $_2$38, $_2$39, $_2$40, and interface shifts $_2$41 map to first-order shifts in lunar normal-mode eigenfrequencies through kernels $_2$42, $_2$43, and $_2$44 (Yan et al., 13 May 2026). GW-driven surface amplitudes depend on the overlap integral

$_2$45

and the study shows that including calibrated GW amplitudes alongside frequencies can reduce estimation errors on the Moon’s elastic parameters by about an order of magnitude relative to frequency-only inversion (Yan et al., 13 May 2026). A complementary analysis of the Lunar Gravitational Wave Antenna emphasizes that long-duration lunar GW inference is a geometric problem: choosing an origin that minimizes timing uncertainty can reduce the relevant time range by roughly an order of magnitude, and in the GW250114 case the shift was from $_2$46 s to $_2$47 s (Tissino et al., 3 Jun 2026). Two minutes before merger, the study reports that LGWA would have measured the chirp mass to a precision of 0.0002 solar masses and constrained sky position to 65 square degrees (90% HPD area) (Tissino et al., 3 Jun 2026).

These developments show that LunarGeo has expanded well beyond classical surface geology. It now includes geometric navigation, neutrino flux estimation, and GW-calibrated inverse problems, all of which use the Moon as an observational platform and as an object of structured inference. A plausible implication is that future LunarGeo frameworks will be intrinsically multimodal, linking maps, simulations, subsurface sounding, interior networks, and dynamical observables into a single quantitative lunar reference system.