1Hyunseob Baik, 2Eun-Jin Cho, 1Kwang-Eun Kim
Icarus (in Press)
Open Access Link to Article [DOI: 10.1016/j.icarus.2026.117341]
1Resources Exploration & Development Research Division, Korea Institute of Geosciences and Mineral Resources (KIGAM), Daejeon, Republic of Korea
2Division of Space Exploration, Korea Astronomy and Space Science Institute (KASI), Daejeon, Republic of Korea
Copyright Elsevier
The Moon Mineralogy Mapper (M3) acquired 85-band hyperspectral data (430–3000 nm) at 140 m/pixel resolution between 2008 and 2009 for spectral imaging of the lunar surface. Conventional per-pixel methods, such as the Modified Gaussian Model (MGM) and Integrated Band Depth (IBD), have been widely used for lunar mineral detection but address complementary problems and do not provide sub-pixel abundance estimates across entire scenes. In this study, linear spectral unmixing (LSU) with Iterative Error Analysis (IEA) endmember extraction was applied to M3 Level 2 reflectance data for sub-pixel mineral mapping of Theophilus crater. After noise reduction via Minimum Noise Fraction (MNF) and enhancement of absorption features through continuum removal, IEA extracted five image-derived endmembers without prior mineralogical information or reference libraries. The extracted endmembers were subsequently identified against 23 reference spectra from the RELAB and USGS spectral libraries using explicit criteria based on diagnostic absorption-band positions and Pearson correlation (r = 0.87–0.97 for the assigned minerals). Four endmembers were identified as spinel, plagioclase, pyroxene, and OH/H2O-bearing materials, while the fifth remains unassigned and likely reflects space-weathering effects or mosaicking artifacts. The five-endmember model reconstructs the observed continuum-removed spectra with a mean RMSE of 0.010, and the derived relative abundance maps show a reasonable spatial correspondence with independent Kaguya Multiband Imager mineral maps (Spearman ρ up to 0.63). The resulting maps show spinel concentrated at the central peak, plagioclase dominating the peak and crater walls, and broadly distributed pyroxene, consistent with previous studies. Notably, the unmixing results reveal spatially distinct abundance maxima for spinel and plagioclase within the central peak, suggesting sub-kilometer compositional gradients within the Pink Spinel Anorthosite (PSA) lithology. This sub-pixel approach yields fractional abundance maps that complement conventional per-pixel methods. The combination of library-independent endmember extraction and library-based identification demonstrates that LSU with IEA provides an effective sub-pixel mapping framework for lunar regions where prior mineralogical knowledge is limited.
