Exploring Alternative Data-Analysis Strategies for Mineralogical Identification in Hyperspectral Images of Oxia Planum

Hyperspectral imaging has become one of the main tools for mineralogical mapping on Mars, particularly through CRISM observations acquired by the Mars Reconnaissance Orbiter (MRO). These datasets have enabled the identification of hydrated minerals, phyllosilicates, sulfates, and silica-bearing deposits associated with ancient aqueous environments, including candidate landing sites such as Oxia Planum [1,2]. Indeed, hydrated silica has previously been reported in this region within the Hydrated Silica-bearing Unit (HSU) [3], together with possible sulfates associated with an absorption feature around 1.75 µm [3].

In the literature, several approaches have been employed for mineralogical determination in satellite images, including DAFA/TT (Dynamic Aperture Factor Analysis/Target Transformation) [4], Spectral Angle Mapper (SAM) [5], abundance-based mapping methods using peak or maximum abundance criteria [6], and spectral unmixing models [7].

These approaches differ in stability, sensitivity to preprocessing, and susceptibility to false positives, especially under variable image quality conditions. SAM is simple and relatively robust to illumination effects but can become unstable when spectra are noisy or highly mixed, leading to ambiguous classifications in spectrally similar materials. Abundance-based mapping methods relying on peak abundance values are sensitive to endmember selection and may produce false positives when noise or residual atmospheric effects generate spurious maxima. DAFA/TT improves robustness through local subspace modelling, but results depend on window size, noise levels, and preprocessing choices, affecting detection consistency. Non-negative Matrix Factorization provides physically interpretable endmembers and abundance maps and can better represent mixed systems, but it is sensitive to initialization, constraints, and convergence issues, which may lead to different solutions under different preprocessing pipelines.

Consequently, definitive validation remains difficult due to the lack of Martian samples and in situ ground-truth analyses in Oxia Planum.

In this study, five CRISM images (frt0000810d_07_if166j_ter3, frt00009a16_07_if166s_trr3, frt0000810d_07_if166j_ter3, frt00009a16_07_if166j_mtr3, and frt00009a16_07_if166j_ter3) were analyzed after atmospheric and photometric correction. Different preprocessing strategies, including ratioing against the lowest-albedo spectrum within each region, Savitzky–Golay smoothing, first derivatives, and Standard Normal Variate (SNV) normalization, were applied to evaluate their effect on noise reduction and spectral feature enhancement under different data quality conditions.

To improve the robustness of mineralogical detections, supervised DAFA/TT and unsupervised MCR-ALS (Multivariate Curve Resolution–Alternating Least Squares) with non-negativity constraints were compared under different preprocessing strategies and image qualities, together with Spectral Angle Mapper (SAM) and abundance-based mapping approaches. Convergent detections across methods were used as an indicator of stability and reduced likelihood of false positives within the analyzed ROIs [5].

Data quality is the primary limiting factor in the interpretation of the results. DAFA/TT yielded positive detections compatible with serpentine and carbonate phases; however, MCR-ALS results showed significant ambiguity, preventing an unequivocal confirmation of these mineral phases. Pixel-by-pixel spectral analysis indicates that the investigated regions are characterized by complex mineralogical mixtures combined with residual spectral noise. These mixtures appear dominated by hydrated silica, as indicated by absorption features around 1.9 µm, consistent with previous studies [3], although contributions from Al–Fe phyllosilicates (~2.3 µm), carbonates (~2.3 and 2.5 µm), and sulfates (~1.7 µm) may also be present.

The results highlight several key limitations in mineralogical discrimination: (i) absence of pure mineral phases within ROIs; (ii) limited spatial resolution (~18 m/pixel), leading to subpixel mixing; (iii) residual spectral noise, which can exceed diagnostic absorption features; (iv) lack of realistic mineral mixtures in spectral libraries used for semi-supervised models; (v) intrinsic rotational ambiguity in MCR-ALS even under non-negativity constraints [7]; and (vi) strong dependence on image quality and dataset selection. Additionally, preprocessing choices significantly influence mineral identification and can affect the occurrence of false positives across methods.

Overall, the applied methods show complementary strengths but also distinct limitations. Rather than providing a single definitive solution, they offer complementary and sometimes inconsistent perspectives on the same data. In this context, cross-disciplinary chemometric strategies that combine and critically compare multiple approaches, instead of relying on a single method, improve interpretability and enhance the robustness of mineralogical conclusions derived from hyperspectral data.

Keywords: Remote sensing, Chemometrics, Spectral Unmixing Models, DAFA/TT, Mineralogical Identification

Acknowledgements: Work supported through the PAMMAT project (Grant No. PID2022-142750OB-I00), funded by the Spanish Agency for Research (through the Spanish Ministry of Science and Innovation, MCIN, and the European Regional Development Fund, FEDER).

References

[1] Murchie, S., Arvidson, R., Bedini, P., Beisser, K., Bibring, J. P., Bishop, J., ... & Wolff, M. (2007). Compact reconnaissance imaging spectrometer for Mars (CRISM) on Mars reconnaissance orbiter (MRO). Journal of Geophysical Research: Planets, 112(E5).

[2] Quantin-Nataf, C., Carter, J., Mandon, L., Thollot, P., Balme, M., Volat, M., ... & Broyer, J. (2021). Oxia Planum: The landing site for the ExoMars “Rosalind Franklin” rover mission: Geological context and prelanding interpretation. Astrobiology, 21(3), 345-366.

[3] McNeil, J. D., Grindrod, P., Tornabene, L. L., Fawdon, P., & Rangarajan, V. G. (2025). Hydrated Silica in Oxia Planum, Mars. Journal of Geophysical Research: Planets, 130(9), e2025JE008989.

[4] Lin, H., Tarnas, J. D., Mustard, J. F., Zhang, X., Wei, Y., Wan, W., ... & Kellner, J. R. (2021). Dynamic aperture factor analysis/target transformation (DAFA/TT) for Mg-serpentine and Mg-carbonate mapping on Mars with CRISM near-infrared data. Icarus, 355, 114168.

[5] Yuhas, R. H., Goetz, A. F., & Boardman, J. W. (1992, June). Discrimination among semi-arid landscape endmembers using the spectral angle mapper (SAM) algorithm. In JPL, summaries of the third annual JPL airborne geoscience workshop. volume 1: AVIRIS workshop.

[6] Clark, R. N., Swayze, G. A., Livo, K. E., Kokaly, R. F., Sutley, S. J., Dalton, J. B., ... & Gent, C. A. (2003). Imaging spectroscopy: Earth and planetary remote sensing with the USGS Tetracorder and expert systems. Journal of Geophysical Research: Planets, 108(E12).

[7] De Juan, A., & Tauler, R. (2006). Multivariate curve resolution (MCR) from 2000: progress in concepts and applications. Critical reviews in analytical chemistry, 36(3-4), 163-176.