Speaker
Description
Planetary missions targeted at rocky bodies generate large amounts of geological data. For example, geochemical information is obtained by spectrometers installed on Martian rovers. Also, microscopic images acquired by cameras on Martian rovers allow individual grain segmentation and subsequent calculation of useful sedimentological parameters (e.g., particle size distributions, circularity, roundness, etc.). Both geochemical composition and individual particle morphometry are multivariate data by nature, i.e., they consist of at least three variables (e.g., elemental concentrations or particle dimensions). Although these kinds of datasets are often analyzed with standard bivariate statistical methods (such as Pearson correlation), the use of dedicated multivariate statistical methods allows capturing more complex relations in data. Other applications of multivariate data analysis include dimensionality reduction and hypothesis generation.
To popularize multivariate statistical methods, we developed an R/shiny application called “pebbler”. It is graphical, cross-platform software which allows easy visualization, parameters tuning and export of results. Although initially designed as a tool to perform morphometric analysis of pebble grains, it can be successfully used for other instances of multivariate data. “Pebbler” supports Principal Component Analysis (PCA) and Linear Discriminant Analysis (LDA) as well as multiple data preprocessing options (including preprocessing of compositional data). We will demonstrate its capabilities using geochemical and granulometric data derived from a Martian exploration mission. Contrary to the most common approach in which statistical quantities of particle size distribution (e.g., mean, standard deviation, skewness, etc.) are analyzed, we will use morphometric information of individual particles and treat them as multivariate data points.