Blind Source Separation for Compositional Time Series.

Area of Science:

• Geosciences
• Data Science
• Time Series Analysis

Background:

• Geological phenomena are often measured over time, yielding compositional time series data where relative values are key.
• Analyzing multivariate, high-dimensional time series requires specialized methods that account for both serial dependence and compositional geometry.

Purpose of the Study:

• To review existing blind source separation (BSS) techniques for time series analysis.
• To demonstrate the application of BSS methods to high-dimensional compositional time series.
• To introduce a novel, flexible BSS method for latent time series analysis.

Main Methods:

• Review of established blind source separation techniques.
• Application of BSS to high-dimensional compositional time series.
• Development and simulation of a new flexible BSS method.
• Analysis of light absorbance data from water samples.

Main Results:

• Existing BSS methods can be adapted for compositional time series.
• The proposed novel BSS method offers flexibility in latent time series assumptions.
• Methodology validated through simulations and real-world environmental data.

Conclusions:

• Blind source separation is a powerful tool for analyzing compositional time series in geosciences.
• The novel BSS method provides a versatile approach for uncovering latent structures in complex environmental data.
• This work enhances the analysis of geological changes and water quality monitoring.