15:30 〜 17:00
[PPS06-P12] Classification Map of the lunar VNIR spectra obtained by the Kaguya/SP data.
キーワード：月、分類、近赤外可視光スペクトル、地質、鉱物マップ
A geological map gives an us the important information for understanding a terrestrial body. The VIS-NIR reflectance spectra of the Moon include key information of the distributed minerals. A global lunar classification map of absorption spectra by K-means was originally proposed by [Hareyama et al., 2019]. They used the spectral data set obtained by the Spectral Profiler (SP) onboard the lunar orbiter KAGUYA after averaging 0.5 deg by 0.5 deg.
We used the clustering tool RasterMiner [Rage, 2021] and applied the latest algorithms of classification to the same SP data set. A non-hierarchical clustering, Fuzzy K-means is applied after determining the optimal number of classifications (K) in advance by Elbow K-means. We applied Fuzzy K-means, a non-hierarchical clustering, and DBSCAN, a hierarchical and density-based spatial clustering both to the lunar reflectance spectral data [Shibakura et al., 2021, JpGU]. The results of applying both algorithms were compared with Hareyama et al. [2019] and existing geological maps for evaluation, respectively. We observed good agreement of our results with the plagioclase abundance (wt%) in FHT (Farside Highlands Terrane). However, maria region did not become clusters including a large number of elements with any parameter set, showing fewer classes than expected. Some over-classification was also observed.
We focus on the two areas: maria and the SPA (South Pole Aitken) regions expecting the possibility of more detailed classification. We also apply clustering algorithms to the other limited region of highland and compare the classifications with our previous results. The updated classification results are compared with existing geological map (or mineral distribution map such as FeO or TiO map) quantitatively by examining the corresponding spots, then calculating the area ratio to evaluate the consistency. We also plan to apply clustering algorithms to original SP data after screening the data under good lightning condition for a more detailed analysis.
This study aims to update the global classification map based on hyperspectral data and contribute to the generation of the new lunar geologic in future.
We used the clustering tool RasterMiner [Rage, 2021] and applied the latest algorithms of classification to the same SP data set. A non-hierarchical clustering, Fuzzy K-means is applied after determining the optimal number of classifications (K) in advance by Elbow K-means. We applied Fuzzy K-means, a non-hierarchical clustering, and DBSCAN, a hierarchical and density-based spatial clustering both to the lunar reflectance spectral data [Shibakura et al., 2021, JpGU]. The results of applying both algorithms were compared with Hareyama et al. [2019] and existing geological maps for evaluation, respectively. We observed good agreement of our results with the plagioclase abundance (wt%) in FHT (Farside Highlands Terrane). However, maria region did not become clusters including a large number of elements with any parameter set, showing fewer classes than expected. Some over-classification was also observed.
We focus on the two areas: maria and the SPA (South Pole Aitken) regions expecting the possibility of more detailed classification. We also apply clustering algorithms to the other limited region of highland and compare the classifications with our previous results. The updated classification results are compared with existing geological map (or mineral distribution map such as FeO or TiO map) quantitatively by examining the corresponding spots, then calculating the area ratio to evaluate the consistency. We also plan to apply clustering algorithms to original SP data after screening the data under good lightning condition for a more detailed analysis.
This study aims to update the global classification map based on hyperspectral data and contribute to the generation of the new lunar geologic in future.