Scholarlink

Understanding the distribution of titanium dioxide (TiO \(_2\) ) and iron oxide (FeO) in the lunar regolith is crucial for future lunar exploration and in-situ resource utilization (ISRU). This study integrates spectral reflectance data from the RELAB database with geochemical compositions from the Lunar Sample Compendium, which compiles analyses of Apollo mission samples. By training machine learning models, specifically Random Forest, on reflectance-concentration relationships derived from these datasets, we develop predictive models for TiO \(_2\) and FeO concentrations. These models are then applied to spectral data from the Clementine mission to estimate the distribution of both oxides across different lunar regions. Instead of validating the models against discrete sample concentrations alone, we assess their reliability by analyzing the correlation between TiO \(_2\) and FeO in well-characterized areas, leveraging established geochemical relationships. Additionally, we perform Principal Component Analysis (PCA) on the available datasets to verify that the most relevant wavelengths for model construction align with those identified in the scientific literature as key for TiO \(_2\) and FeO detection. This approach enhances the robustness of our methodology, ensuring its applicability to regions where compositional data remain limited.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Predicting TiO2 and FeO Concentrations in Lunar Regolith Using Machine Learning Models: A Spectral Reflectance Approach

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Julia Fernández Díaz,
  • Francisco Javier de Cos Juez,
  • Fernando Sánchez Lasheras,
  • Javier Gracia Rodriguez,
  • Santiago Iglesias,
  • Javier Rodriguez,
  • Saul Perez,
  • Alejandro Buendia

摘要

Understanding the distribution of titanium dioxide (TiO \(_2\) ) and iron oxide (FeO) in the lunar regolith is crucial for future lunar exploration and in-situ resource utilization (ISRU). This study integrates spectral reflectance data from the RELAB database with geochemical compositions from the Lunar Sample Compendium, which compiles analyses of Apollo mission samples. By training machine learning models, specifically Random Forest, on reflectance-concentration relationships derived from these datasets, we develop predictive models for TiO \(_2\) and FeO concentrations. These models are then applied to spectral data from the Clementine mission to estimate the distribution of both oxides across different lunar regions. Instead of validating the models against discrete sample concentrations alone, we assess their reliability by analyzing the correlation between TiO \(_2\) and FeO in well-characterized areas, leveraging established geochemical relationships. Additionally, we perform Principal Component Analysis (PCA) on the available datasets to verify that the most relevant wavelengths for model construction align with those identified in the scientific literature as key for TiO \(_2\) and FeO detection. This approach enhances the robustness of our methodology, ensuring its applicability to regions where compositional data remain limited.