Quantitative Inversion of Martian Hydrous Minerals Based on LSTM-1DCNN Model

Hydrous minerals are significant indicators of the ancient aqueous environment on Mars, and orbital hyperspectral data are one of the most effective tools for obtaining information about the distribution of hydrous minerals on the Martian surface. However, prolonged weathering, erosion, and other external forces result in complex mixing effects, often weakening the spectral absorption features of individual minerals. This study proposes a quantitative inversion method for Martian hydrous minerals by integrating a radiative transfer model with a deep learning network. Based on the physics of the Hapke radiative transfer model, the single-scattering albedo spectra of mineral end members were obtained. Additionally, the Linear Spectral Mixture Model was employed to generate a large number of fully constrained mineral mixture samples, providing theoretical support for experimental data. An LSTM-1DCNN model was trained to establish a data-driven quantitative inversion framework. CRISM data were applied to the Eberswalde Crater region to retrieve the abundances of 21 hydrous minerals, including tremolite, opal, and serpentine. The average abundance of hydrous minerals was calculated to be 0.018, with a total area proportion of approximately 8%. Additionally, by analyzing the distribution areas of hydrous silicates, hydrous sulfates, and hydrous hydroxides, the water activity history of the region was inferred. The results align with findings from related studies and mineral spectral index results. By incorporating deep learning into traditional mixing models, this study identifies the distribution of various low-abundance hydrous minerals, enhancing the accuracy of Martian hydrous mineral inversion. It is expected to provide valuable references for the selection of landing sites for Tianwen-3 and support the smooth implementation of China’s Mars exploration mission.