Nonlinear mapping technique for data visualization and clustering assessment of LIBS data: application to ChemCam...

Area of Science:

• Planetary Science
• Analytical Chemistry
• Data Science

Background:

• ChemCam, a laser-induced breakdown spectroscopy (LIBS) instrument, will analyze Martian geology.
• LIBS is vital for identifying samples and quantifying elemental abundances remotely.
• Effective data visualization and clustering are essential for interpreting LIBS data.

Purpose of the Study:

• To compare linear and nonlinear multivariate techniques for visualizing ChemCam LIBS data.
• To assess the effectiveness of Principal Component Analysis (PCA), Independent Component Analysis (ICA), and Sammon's map projection for 2D data representation.
• To determine the optimal technique for combining data visualization and clustering assessment.

Main Methods:

• Utilized Principal Component Analysis (PCA) and Independent Component Analysis (ICA) as linear methods.
• Applied Sammon's map projection as a nonlinear technique.
• Evaluated 2D data representation using optimization values and clustering purity via entropy.

Main Results:

• Sammon's map projection provided the best 2D representation (2.8%-4.3% optimization) and clustering purity (entropy 0.81).
• Linear methods (ICA, PCA) showed significantly higher 'stress' and lower clustering purity.
• Sammon's map was faster when initialized with ICA projections.

Conclusions:

• Nonlinear Sammon's map projection is the superior technique for 2D visualization and clustering of ChemCam LIBS data.
• This method enhances the interpretation of elemental abundance data from Mars.
• Higher dimensional projections (PCA, ICA) may improve results but reduce intuitive 2D interpretability.