Global Lunar Chemical Maps

• Global lunar chemical maps are quantitative, spatially resolved representations of the Moon’s elemental and mineralogical variations that elucidate its evolutionary history.
• They integrate diverse methodologies including XRF, gamma-ray spectroscopy, hyperspectral imaging, and in situ analyses calibrated with lunar samples.
• Advanced machine learning and statistical clustering enhance these maps, supporting targeted resource assessment and informed lunar exploration.

Global lunar chemical maps are quantitative, spatially resolved representations of elemental and mineralogical variation across the lunar surface. These maps are fundamental to planetary science as they provide insight into the origin, evolution, and present composition of the Moon, and form the basis for comparative planetology, geological interpretation, and resource assessment. They are constructed using a variety of remote sensing, in situ, and sample-return techniques—most prominently X-ray and gamma-ray spectroscopy, hyperspectral imaging, and neutron mapping—each addressing specific chemical and spatial sensitivities. Recent advances combine large-scale observational datasets, improved calibration with ground-truth samples, machine learning, and statistical clustering to deliver increasingly high-resolution and geochemically informative maps.

1. Measurement Techniques and Instrumentation

Global lunar chemical mapping draws upon a diverse suite of instruments and platforms:

• X-ray Fluorescence (XRF) Spectroscopy: Orbital instruments such as Chandrayaan-2 CLASS (Kumar et al., 21 Aug 2025, Narendranath et al., 2013), Chang’E-2 XRF (Dong et al., 2015), and the Apollo 15/16 XRS (Gloudemans et al., 2021) utilize solar-induced XRF to quantify surface abundances of major rock-forming elements (primarily O, Na, Mg, Al, Si, Ca, Fe). XRF probes the upper few hundred microns of regolith, providing near-surface elemental maps with spatial resolutions now reaching 5.3 km/pixel (Kumar et al., 21 Aug 2025).
• Gamma-Ray and Neutron Spectroscopy: Lunar Prospector mapped the global distribution of elements such as Fe, Ti, Th, and K, with gamma-ray measurements sampling depths of a few centimeters and neutron data revealing information about hydrogen (and by proxy, regolith hydration and composition). Enhanced spatial resolution has been achieved with advanced pixon image reconstruction techniques (Wilson et al., 2018).
• Hyperspectral and Multispectral Imaging: Visible to mid-infrared spectral datasets collected by the Moon Mineral Mapper (M³) and instruments on Gaofen‑4 provide compositional and mineralogical clustering via analysis of spectral absorption features. Machine learning approaches, notably variational autoencoders combined with k-means clustering, have enabled unsupervised global mineralogical mapping (Thoresen et al., 5 Nov 2024).
• Optical and UV-Vis Photometry: Large-area imaging (e.g., Gaofen‑4, ~500 m resolution) can be quantitatively calibrated using ground-truth lunar samples to generate spatially seamless major oxide maps by correlating reflectance features with chemical composition (Lu et al., 2020).
• In Situ and Isotopic Analysis: New instrument platforms, such as the Luna-27 DLS-L spectrometer, will directly measure regolith-derived volatiles’ isotopic ratios (D/H, 18O/17O/16O, 13C/12C), further characterizing local and potentially global volatile processes (Meshcherinov et al., 2023).

Each technique offers distinct penetration depths, spatial resolutions, elemental sensitivities, and methodological constraints, necessitating careful integration and cross-calibration.

2. Data Processing, Calibration, and Map Generation

Sophisticated data analysis and calibration pipelines are central to transforming raw spectral data into reliable chemical maps:

• Calibration with Ground-Truth Samples: High-fidelity quantitative maps are produced by statistically correlating spectral responses with known chemical abundances measured from Apollo and Luna soil samples (Lu et al., 2020). For example, reflectance is converted to compositional data using regression models (linear, power, exponential, etc.), optimized for each element based on sample-site cross-validation.
• Spectral Deconvolution and Background Subtraction: For XRF data, spectra are decomposed into Gaussian components corresponding to Kα lines of major elements (Mg, Al, Si) and background, often using χ²-minimization or other least-squares fitting algorithms (Dong et al., 2015, Kumar et al., 21 Aug 2025). Accurate background modeling—subtracting noise, cosmic-ray, and scattered solar contributions—is critical, especially given solar variability and instrumental artifacts.
• Resolution Harmonization: Comparing and integrating datasets with disparate spatial resolutions (e.g., gamma-ray data at ~45 km, LOLA laser albedo at tens of meters), requires image processing techniques such as standard and adaptive Gaussian blurring. Adaptive filters are spatially tuned to bridge gaps, enabling coherent data fusion (Strukova et al., 8 Jul 2024).
• Open-Source Pipelines: Recent efforts (e.g., CLASS data on Chandrayaan‑2 (Kumar et al., 21 Aug 2025)) utilize Python-based, openly available analysis stacks (NumPy, SciPy, Astropy, Pandas), enhancing reproducibility and providing community access to both methodology and derived products.
• Machine Learning and Unsupervised Clustering: Dimensionality reduction (using convolutional variational autoencoders) followed by unsupervised clustering (k-means), as applied to M³ data, enables unbiased global mineralogical mapping (Thoresen et al., 5 Nov 2024).

3. Global Patterns and Geological Correlations

Global lunar chemical maps reveal pronounced geochemical variability correlating with geological terranes and surface features:

• Terrane Structure: Mg/Al, Mg/Si, and Al/Si ratio maps consistently delineate the feldspathic highlands (low Mg/Al, high Al/Si) from the basaltic maria (high Mg/Al, low Al/Si), with the South Pole–Aitken (SPA) basin and KREEP-rich terranes (elevated Th, Fe, K, P) exhibiting distinctive signatures (Kumar et al., 21 Aug 2025, Dong et al., 2015, Gloudemans et al., 2021).
• High-Resolution Features: Advanced spectral deconvolution and spatial binning (e.g., 5.3 km grids (Kumar et al., 21 Aug 2025)) enable the identification of geochemically unique features, such as Eratosthenian basalts in Mare Imbrium, compositional heterogeneity in the Tycho crater ejecta (linked to oblique impact hypotheses (Lu et al., 2020)), and nearly pure anorthosite exposures in specific basins (e.g., Hertzsprung and Schrödinger (Wilson et al., 2018)).
• Surface vs. Subsurface: Discrepancies between measures of regolith maturity and hydrogen content across different depths, revealed by the contrast between optical albedo and neutron/epithermal counts, indicate a complex, layered regolith—most notably in basin structures like Orientale (Wilson et al., 2018).
• Correlation Patterns: Elemental flux ratio maps produced by independent XRF analyses show strong correspondence with established sample-based abundance maps, validating compositional interpretations and supporting the use of mixture modeling (e.g., Gaussian mixture models in Mg/Si vs. Al/Si space) to classify discrete lithological regions (Kumar et al., 21 Aug 2025).

4. Statistical Models and Machine Learning Applications

Statistical analysis and machine learning frameworks are increasingly central to the interpretation and utilization of global chemical maps:

• Gaussian Mixture Models (GMM): Applied to bivariate histograms of elemental flux ratios (e.g., Mg/Si vs. Al/Si), GMMs distinguish compositional clusters corresponding to major lunar terranes, supporting robust, data-driven geochemical unit classification (Kumar et al., 21 Aug 2025).
• Extreme Gradient Boosting (XGB): XGB regression models using elemental concentrations as features have been trained to predict surface albedo, offering predictive mapping of surface properties in areas where one type of data is missing or incomplete (Strukova et al., 8 Jul 2024).
• Interactive Analytical Tools: Real-time visualization of prediction errors and anomalies (e.g., albedo model errors mapped spatially) provide geologists with diagnostic perspectives on both measurement uncertainties and genuine compositional anomalies (Strukova et al., 8 Jul 2024).
• Dimensionality Reduction and Spectral Clustering: Convolutional variational autoencoders reduce hyperspectral data into latent variables that retain dominant mineralogical information, which are then clustered (optimally into five groups) to create global mineralogical maps that align with known concentrations of plagioclase, clinopyroxene, and olivine (Thoresen et al., 5 Nov 2024).

5. Calibration, Validation, and Integration with Sample Data

The accuracy and interpretability of global chemical maps hinge on rigorous calibration and continuous cross-validation against direct measurements:

• Sample Calibration: High-resolution optical and infrared maps (e.g., Gaofen‑4, M³) are quantitatively anchored to Apollo and Luna sample site chemistry through multivariate regression, with typical R² values for major elements exceeding 0.87 and standard deviations typically below 1.33 wt% (Lu et al., 2020).
• Instrument Corrections: Updated transmission models (e.g., Apollo XRS, with revised filter matrix inversions and corrections for solar spectral variations) have resulted in a downward revision (~30%) of Al/Si and Mg/Al intensity ratios, realigning remote sensing maps with laboratory concentrations (Gloudemans et al., 2021).
• Consistency Across Missions: Comparisons between maps derived from Apollo, Lunar Prospector, Chandrayaan-2 CLASS, Chang’E-2, and Gaofen‑4 consistently show both absolute and spatial agreement in major elemental trends, confirming both methodology and geochemical interpretation (Kumar et al., 21 Aug 2025, Gloudemans et al., 2021, Dong et al., 2015, Lu et al., 2020).

6. Implications for Lunar Evolution, Resource Assessment, and Future Directions

The synthesis of global chemical maps underpins both fundamental science and applied lunar exploration:

• Crustal Differentiation and Impact History: Compositional maps reinforce the scenario of a plagioclase-rich highlands crust formed from a lunar magma ocean, with mare basalts sourced from differentiated mantle. Features such as the preserved anorthosite in Hertzsprung and Schrödinger basins provide constraints on crustal thickness and impact excavation (Wilson et al., 2018).
• Volatile Inventories: The discovery of hydrated minerals like ULM‑1 [(NH₄)MgCl₃·6H₂O] in the Chang’e‑5 samples and isotopic measurements by upcoming missions (Luna-27) indicate that water and other volatiles are not restricted to permanently shadowed regions but may persist as hydrated phases even in sunlit terrains, influencing global volatile models and ISRU potential (Jin et al., 2023, Meshcherinov et al., 2023).
• Resource Prospecting and Site Selection: High-resolution global maps of Fe, Ti, Mg, Al, and associated mineral groupings inform site selection for both scientific exploration and in situ resource extraction strategies. Targeting regions of high Mg/Al or unique spectral clusters enables efficient exploration and maximizes scientific return (Kumar et al., 21 Aug 2025, Thoresen et al., 5 Nov 2024).
• Methodological Advances: The integration of open-source data processing, advanced statistical modeling, and multi-instrument calibration sets a precedent for reproducibility and multi-mission synergy. Anticipated mission datasets (e.g., Luna-27 isotopic mapping) will further enhance compositional resolution and depth.
• Pan-planetary Applicability: Techniques such as adaptive Gaussian blurring for cross-resolution data fusion and unsupervised spectral clustering are extensible to other airless bodies (e.g., Mercury), enabling broader comparative studies in planetary geochemistry (Strukova et al., 8 Jul 2024, Thoresen et al., 5 Nov 2024).

7. Representative Formulas and Conceptual Frameworks

Central analytic expressions utilized in the field include:

• XRF Line Ratio (as defined for CLASS):

$$\mathcal{R}_M^L = \frac{A_M \sigma_M}{A_L \sigma_L}$$

where $A_M, A_L$ are Gaussian amplitudes for elements M and L, and $\sigma_M, \sigma_L$ the corresponding standard deviations (Kumar et al., 21 Aug 2025).

• Reflectance-to-Abundance Regression:

$$\mathrm{Ref} = \frac{\mathrm{RAD}}{\cos i \cdot \cos e \cdot d^2 \cdot E}$$

where RAD is radiance, $i$ and $e$ are solar and instrument zenith angles, $d$ is Sun–Moon distance, and $E$ is solar irradiance (Lu et al., 2020).

• VAE Evidence Lower Bound (ELBO):

$$\mathcal{L}(x) = \mathbb{E}_{q_\phi(z|x)}[\log p_\theta(x|z)] - \mathcal{D}_{KL}(q_\phi(z|x)||p(z))$$

combining reconstruction fidelity with latent-space regularization for clustering hyperspectral data (Thoresen et al., 5 Nov 2024).

Through the continual refinement of measurement techniques, data analysis, statistical modeling, and cross-validation with sample and in situ measurements, global lunar chemical maps are evolving into multidimensional, high-resolution frameworks that reveal the compositional complexity of the Moon and inform planetary science and exploration agendas for decades to come.